refactor(agent): read WiFi info via native WLAN API instead of netsh parsing

netsh wlan show interfaces output is localized: "Radio type"/"Channel"
(and on some languages even "Signal") never matched on non-English
images, silently dropping fields from wifi_signal_info and network
change snapshots. It also spawned a child process per collection.

Replace both netsh call sites with wlanapi.dll P/Invoke
(WlanQueryInterface: current_connection + channel_number):

- New WlanNativeMethods (raw interop) + WifiInfoReader (fail-soft,
  prefers the active NIC's interface GUID, falls back to the first
  connected WLAN interface; returns null on machines without WLAN).
- DeviceInfoCollector.CollectWiFiSignalInfo and
  NetworkChangeDetector.CollectWiFiInfo now use the reader; RadioType
  strings stay netsh-compatible ("802.11ax" ...), event fields
  (wifiSsid/wifiSignalPercent/wifiRadioType/wifiChannel) unchanged.
- SystemPaths.Netsh removed (no remaining references).

V2.Core suite 2085/2085 green incl. new WifiInfoReader tests; struct
marshaling verified against native sizeof (604/68/36/532).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -63,7 +63,7 @@ jobs:
 
       - uses: actions/setup-node@v7
         with:
-          node-version: '20'
+          node-version: '22'
           cache: npm
           cache-dependency-path: src/Web/autopilot-monitor-web/package-lock.json
 
@@ -86,7 +86,7 @@ jobs:
 
       - uses: actions/setup-node@v7
         with:
-          node-version: '20'
+          node-version: '22'
           cache: npm
           cache-dependency-path: src/McpServer/autopilot-monitor-mcp/package-lock.json
 
@@ -106,7 +106,7 @@ jobs:
 
       - uses: actions/setup-node@v7
         with:
-          node-version: '20'
+          node-version: '22'
 
       - name: Regenerate
         run: node rules/scripts/combine.js

.github/workflows/combine-rules.yml

@@ -26,7 +26,7 @@ jobs:
 
       - uses: actions/setup-node@v7
         with:
-          node-version: '20'
+          node-version: '22'
 
       - name: Install ajv-cli
         run: npm install -g ajv-cli ajv-formats

.github/workflows/deploy-web.yml

@@ -21,6 +21,11 @@ jobs:
     if: github.event_name == 'push' || github.event_name == 'workflow_dispatch' || (github.event_name == 'pull_request' && github.event.action != 'closed')
     runs-on: ubuntu-latest
     name: Build and Deploy Job
+    # Dependabot PRs run without repo secrets, so the deploy token is absent.
+    # Surface it as an env var so the deploy step can gate on it — secrets
+    # cannot be referenced directly in an `if:` expression.
+    env:
+      DEPLOY_TOKEN: ${{ secrets.AZURE_SWA_DEPLOY_TOKEN }}
     steps:
       - uses: actions/checkout@v7
         with:
@@ -28,6 +33,9 @@ jobs:
           fetch-depth: 0  # Full history needed for git-based sitemap lastmod dates
       - name: Build And Deploy
         id: builddeploy
+        # Skip (green) instead of failing when the token is missing (Dependabot
+        # PRs); real deploys on main and internal PRs still have it.
+        if: env.DEPLOY_TOKEN != ''
         uses: Azure/static-web-apps-deploy@v1
         env:
           NEXT_PUBLIC_ENTRA_CLIENT_ID: ${{ vars.NEXT_PUBLIC_ENTRA_CLIENT_ID }}
@@ -39,9 +47,6 @@ jobs:
         with:
           azure_static_web_apps_api_token: ${{ secrets.AZURE_SWA_DEPLOY_TOKEN }}
           repo_token: ${{ secrets.GITHUB_TOKEN }} # Used for Github integrations (i.e. PR comments)
-          # Dependabot PRs run without repo secrets, so the deploy token is absent.
-          # Skip (green) instead of failing the check; real deploys on main still have the token.
-          skip_deploy_on_missing_secrets: true
           action: "upload"
           ###### Repository/Build Configurations - These values can be configured to match your app requirements. ######
           # For more information regarding Static Web App workflow configurations, please visit: https://aka.ms/swaworkflowconfig
@@ -54,9 +59,12 @@ jobs:
     if: github.event_name == 'pull_request' && github.event.action == 'closed'
     runs-on: ubuntu-latest
     name: Close Pull Request Job
+    env:
+      DEPLOY_TOKEN: ${{ secrets.AZURE_SWA_DEPLOY_TOKEN }}
     steps:
       - name: Close Pull Request
         id: closepullrequest
+        if: env.DEPLOY_TOKEN != ''
         uses: Azure/static-web-apps-deploy@v1
         with:
           azure_static_web_apps_api_token: ${{ secrets.AZURE_SWA_DEPLOY_TOKEN }}

docs/agent/esp-policy-provider-stall.md

@@ -1,7 +1,7 @@
 ---
 type: Concept
 title: ESP Policy-Provider Stall Detection — the no-timeout Setup/Apps wait and the co-management Sidecar gap
-description: Why a registered EnrollmentStatusTracking CSP policy provider that never sets TrackingPoliciesCreated parks the user ESP at "Apps (Identifying)" indefinitely, the pre-provisioning + ConfigMgr field case, and how the always-on esp_policy_provider_stalled tripwire (15-min wall-clock dwell, StartupEventGate cross-restart dedup) surfaces it.
+description: Why the ESP's Apps wait is keyed to the Intune-registered "Sidecar" provider by name — a foreign provider (ConfigMgr) with TrackingPoliciesCreated=1 does not satisfy it — plus the classic never-completes stall, and how the always-on two-arm esp_policy_provider_stalled tripwire (reason=provider_incomplete | sidecar_provider_missing, 15-min wall-clock dwell, StartupEventGate cross-restart dedup) surfaces both.
 resource: /src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/SystemSignals/EspPolicyProviderProbe.cs
 tags:
   - agent
@@ -9,7 +9,7 @@ tags:
   - enrollment-status-tracking
   - co-management
   - stall-detection
-timestamp: 2026-07-23T00:00:00+02:00
+timestamp: 2026-07-24T00:00:00+02:00
 ---
 
 # ESP Policy-Provider Stall Detection
@@ -42,37 +42,59 @@ exactly these registry paths. The same document states that ESP policy providers
 aware of each other** and that **pre-provisioning is unsupported** with Autopilot into
 co-management.
 
-## The field case
+## The field case (issue #106)
 
 A tenant pre-provisions devices and installs the ConfigMgr client as a blocking app.
 Occasionally the user-phase ESP reaches "Apps (Identifying)" and never leaves it — for
 hours to days. On those devices `Device\Setup\Apps\PolicyProviders` contains **only a
-`ConfigMgr` key (no `Sidecar`)** and `TrackingPoliciesCreated` is never set. Manually
-renaming the key to `Sidecar` unblocks the ESP immediately (the IME then finds its
-provider node and reports tracking-policies-created). The mechanism is exactly the
-documented no-timeout wait; the unsupported pre-provisioning + co-management combination
-explains why only some devices are affected. Why the Sidecar registration is missing or
-displaced on those devices is not publicly documented — irrelevant for detection, since
-the signature is deterministic.
+`ConfigMgr` key (no `Sidecar`)** — and, per the customer's follow-up, **`TrackingPoliciesCreated=1`
+IS set under it**. The provider is "complete" by the CSP value contract, yet the ESP
+still waits. Manually renaming the key to `Sidecar` — which creates no tracking entries,
+only satisfies the name — unblocks the ESP instantly.
+
+The naive CSP-doc reading ("any provider with TrackingPoliciesCreated=1 satisfies the
+wait") is therefore wrong for this case. What the evidence supports instead: provider
+nodes are registered **by the Intune service at enrollment**, and its ESP-policy WAP
+payload always registers `Sidecar` (ConfigMgr joins only via a co-management settings
+policy — see the payload documented by Niehaus). The ESP waits on the provider it
+expects **by name**; a foreign provider occupying the slot does not satisfy it,
+regardless of its completion values. On the affected devices the CM client (installed
+as a plain Win32 app, no co-management settings policy) evidently displaced the
+Sidecar node — consistent with Microsoft's warning that providers "aren't currently
+aware of others" and that pre-provisioning is unsupported with co-management. The exact
+CloudExperienceHost mechanism is not publicly documented; the registry signature —
+setupApps providers present, `Sidecar` absent — is deterministic either way.
 
 ## Detection
 
 `EspPolicyProviderProbe` (fail-soft, static, `ScopedOverride` test seam — structural clone
 of `EspTrackingInfoProbe`) enumerates all providers in both kinds and scopes.
 `EspPolicyProviderStallDetector` evaluates a pure transition function on a 60-second tick
-from the always-on `EspPolicyProviderStallHost`:
+from the always-on `EspPolicyProviderStallHost` — **two arms**, same 15-minute wall-clock
+dwell, one shared event type discriminated by `reason`:
 
-* A provider **continuously incomplete for ≥ 15 minutes** (wall-clock) is stalled. The
-  dwell clock starts at first incomplete observation and resets on completion or
-  disappearance. Missing values parse as incomplete (fail-toward-observation).
+* **Arm 1, `reason=provider_incomplete`** — a provider **continuously incomplete for
+  ≥ 15 minutes**. The dwell clock starts at first incomplete observation and resets on
+  completion or disappearance. Missing values parse as incomplete
+  (fail-toward-observation). Catches e.g. Sidecar itself never reporting.
+* **Arm 2, `reason=sidecar_provider_missing`** — **≥ 1 `Setup\Apps` provider registered
+  but none named `Sidecar`**, continuously for ≥ 15 minutes. Deliberately ignores
+  provider completeness (the field case had `TrackingPoliciesCreated=1`). Sidecar
+  registering later — the legitimate startup ordering — clears the condition and resets
+  the clock. DevicePreparation-only snapshots don't activate it.
+* When both arms cross in the same round, ONE event is emitted and
+  `sidecar_provider_missing` wins as `reason` (it explains the hang regardless of the
+  foreign provider's state).
 * Missing root key / empty provider list = normal early enrollment: clears all dwell
-  state, never fires.
-* One `esp_policy_provider_stalled` **Warning** per newly stalled provider set
-  (`ImmediateUpload`), one-shot per provider in-process, cross-restart deduped via a
-  `StartupEventGate` fingerprint over the sorted currently-stalled key set — a restart
-  into the same stall stays silent, a new provider joining the stall re-reports.
-* Payload: `providers[]` (full table: name, scope, kind, raw values, complete flag),
-  `stalledProviders[]` with `stalledForMinutes`, `sidecarRegistered`, `dwellMinutes`.
+  state (both arms), never fires.
+* One `esp_policy_provider_stalled` **Warning** per newly stalled condition set
+  (`ImmediateUpload`), one-shot per condition in-process, cross-restart deduped via a
+  `StartupEventGate` fingerprint over the sorted currently-active fired conditions
+  (provider keys + a `sidecarMissing` sentinel) — a restart into the same stall stays
+  silent, a new condition joining re-reports.
+* Payload: `reason`, `providers[]` (full table: name, scope, kind, raw values, complete
+  flag), `stalledProviders[]` with `stalledForMinutes`, `sidecarMissingForMinutes`
+  (arm 2 only), `sidecarRegistered`, `dwellMinutes`.
 * **Purely observational**: emitted via `InformationalEventPost` (dispatch-guard exempt
   pass-through) — never touches decision state. No config toggle, no ConfigVersion bump
   (precedent: `disk_space_low`).
@@ -93,33 +115,37 @@ resource), so both the detector and the guardrail change ship with the agent rel
 
 ## Examples
 
-Customer-case event (abbreviated):
+Customer-case event (abbreviated) — note the provider is "complete" by CSP values, the
+hang is carried entirely by `reason=sidecar_provider_missing`:
 
 ```json
 {
   "eventType": "esp_policy_provider_stalled",
   "severity": "Warning",
   "data": {
+    "reason": "sidecar_provider_missing",
     "dwellMinutes": 15,
     "sidecarRegistered": false,
+    "sidecarMissingForMinutes": 15.0,
     "providers": [
       { "name": "ConfigMgr", "scope": "device", "kind": "setupApps",
-        "trackingPoliciesCreated": null, "installationState": null, "complete": false }
+        "trackingPoliciesCreated": 1, "installationState": null, "complete": true }
     ],
-    "stalledProviders": [
-      { "name": "ConfigMgr", "scope": "device", "kind": "setupApps", "stalledForMinutes": 15.0 }
-    ]
+    "stalledProviders": []
   }
 }
 ```
 
-`sidecarRegistered=false` while `setupApps` providers exist is the co-management
-signature; `sidecarRegistered=true` with a stalled third-party provider generalizes the
-same contract violation.
+`reason=sidecar_provider_missing` is the co-management signature;
+`reason=provider_incomplete` with `sidecarRegistered=true` covers a stalled Sidecar or
+third-party provider — the generalized contract violation.
 
 # Citations
 
 * EnrollmentStatusTracking CSP — <https://learn.microsoft.com/en-us/windows/client-management/mdm/enrollmentstatustracking-csp>
 * Windows Autopilot into co-management (requirements, provider-conflict warning, troubleshooting registry paths) — <https://learn.microsoft.com/en-us/intune/configmgr/comanage/autopilot-enrollment>
+* Troubleshoot the ESP (models app tracking exclusively via the `Sidecar` subkey) — <https://learn.microsoft.com/en-us/troubleshoot/mem/intune/device-enrollment/understand-troubleshoot-esp>
+* Niehaus: the Intune ESP-policy WAP payload registering the provider nodes (`Sidecar` always, `ConfigMgr` only with a co-management settings policy) — <https://oofhours.com/2023/08/25/run-an-sccm-task-sequence-during-autopilot/>
+* Niehaus: ESP tracks arbitrarily named providers when they are properly registered — <https://oofhours.com/2023/09/18/track-anything-using-esp/>
 * Agent code: `EspPolicyProviderProbe.cs`, `EspPolicyProviderStallDetector.cs`, `EspPolicyProviderStallHost.cs` (SystemSignals / Orchestration.Hosts)
 * Related: [MDM Reboot Coalescing](mdm-reboot-coalescing.md), [Gather-Rule Guardrails](../rules/gather-rule-guardrails.md)

docs/agent/secureboot-ca2023-detection.md

@@ -37,6 +37,21 @@ Fail-soft contract like `OobeStateReader`: never throws; failures map to status
 (`not_uefi` = legacy BIOS via `ERROR_INVALID_FUNCTION`, `variable_not_found`,
 `privilege_denied`, `error_<code>`).
 
+This is the repo's only privilege-adjustment code, so it also carries the ownership rules
+any future one should copy:
+
+* The process token lives in a `SafeHandle` assigned by the P/Invoke marshaler, so the CLR
+  owns it from the moment `OpenProcessToken` returns — the critical finalizer releases it
+  even if `Dispose` never runs (async exception before the instance reaches its `using`).
+* The restore obligation is recorded immediately after the successful `AdjustTokenPrivileges`,
+  and only when the privilege was not *already* enabled; nothing that can throw sits between
+  the adjustment and that assignment. `Dispose` is idempotent, swallows a failed restore, and
+  releases the handle in a `finally`.
+* The privilege is adjusted on the **process** token, so enable → read → restore is a
+  process-wide side effect and runs under a static gate. Without it a concurrent reader would
+  see "already enabled", correctly skip its own restore, and then lose the privilege mid-read
+  when the first reader disposes — a spurious `privilege_denied` on a machine that holds it.
+
 Fields land on the existing `secureboot_status` event (once per agent start,
 restart-deduped by the StartupEventGate): `uefiFirmwareReadStatus`, per-cert booleans
 (`uefiDbHasWindowsUefiCa2023`, `uefiDbHasWindowsProductionPca2011`,

docs/backend/device-journeys.md

@@ -0,0 +1,122 @@
+---
+type: Concept
+title: Device Journeys & First-Time-Right (F2)
+description: How a device's terminal enrollment attempts are grouped into journeys (device key, junk-serial exclusion, first-success end, 30-day gap), where the DeviceHistories chain and the daily FTR aggregates are computed, stored and pruned, and which surfaces serve them.
+resource: src/Backend/AutopilotMonitor.Functions/Helpers/DeviceJourneyCalculator.cs
+tags:
+  - backend
+  - metrics
+  - device-history
+  - first-time-right
+  - insights
+timestamp: 2026-07-27T12:00:00+02:00
+---
+
+# Device Journeys & First-Time-Right (F2)
+
+Answers "how many devices enrolled right on the first try?" and "show me every attempt
+this device made". Session success rate hides wipe-and-retry loops — every retry that
+eventually succeeds counts as a success; First-Time-Right (FTR) exposes the real pain.
+All grouping semantics live server-side in `DeviceJourneyCalculator` (JourneyVersion 1);
+no consumer re-derives them.
+
+# Schema
+
+## Device key
+
+`(TenantId, normalized serial)` — normalization is trim + lower-case. Junk/placeholder
+serials are excluded entirely (never a chain, never in FTR, disclosed as a per-day
+exclusion count): `System Serial Number`, `To Be Filled By O.E.M.`, `Default string`,
+`0`, `None`, `INVALID`, `Unknown` (the agent's WMI-failure sentinel), empty, or shorter
+than 4 characters. A motherboard swap changes the serial and therefore the device key;
+re-enrollment under a different tenant is a separate key by design.
+
+## Journey
+
+Terminal sessions of a device key ordered by StartedAt. The terminal set is
+**Succeeded / Failed / Incomplete** — `Pending` (a sealed WhiteGlove part 1),
+`AwaitingUser`, `Stalled` and `InProgress` are OPEN sessions and never attempts.
+
+* A journey **ends** with the first terminal success; the next session starts a new
+  journey (redeployment).
+* A gap of **more than 30 days** since the previous attempt's terminal moment
+  (`CompletedAt`, fallback StartedAt) starts a new journey even without a success
+  (device shelved/repurposed). Constant, not a setting — production gaps: median 4.9 h,
+  92 % < 7 d.
+* WhiteGlove part 1 + part 2 share ONE session row, so a completed WG enrollment is ONE
+  attempt; a WG device still waiting for its user session leaves the journey open —
+  never a failed attempt.
+* **Attempt count** = terminal sessions in the journey (`Failed` and `Incomplete` are
+  non-successful attempts).
+* **FTR rate** = completed journeys with attempt count 1 ÷ all completed journeys.
+  Open and gap-abandoned journeys never enter numerator or denominator.
+
+Durations shown for attempts are the sessions' authoritative `DurationSeconds` verbatim
+(WhiteGlove pause excluded; `Incomplete` deliberately has none) — never
+`CompletedAt − StartedAt`, which is later in ~25 % of terminal sessions.
+
+## Persistence & compute
+
+* `DeviceHistories` (PK = TenantId, RK = percent-encoded normalized serial): the chain of
+  the device's terminal session refs `{sessionId, startedAt, completedAt, status,
+  enrollmentType, isPreProvisioned, durationSeconds, adminMarked}` capped at the 20 most
+  recent, plus derived `CurrentJourneyAttempts` (the LAST journey's attempts — the
+  banner's "Attempt N"), `JourneyCount` (chain-scoped) and `JourneyVersion`. Written
+  inline at the session-terminal single-writer seam
+  (`TableSessionRepository.UpdateSessionStatusAsync`, all three terminal statuses,
+  DeletionState-guarded) so the session banner is fresh; healed by the maintenance sweep.
+  Deliberately NOT part of the per-session deletion manifest — the row aggregates many
+  sessions. Wiped on tenant offboarding.
+* Sweep (`SweepDeviceJourneysAsync`, 30-day rolling window, 2h tick): (1) drops chain
+  refs of deleted sessions **tombstone-driven** — every session deletion (cascade and
+  retention both run through the cascade worker) leaves a `SessionTombstones` marker for
+  ~7 days, far longer than the sweep cadence, and even an expired-but-unpruned marker is
+  valid deletion evidence; a chain left empty deletes its row. (2) Merges the window's
+  terminal sessions into their chains (backfill, inline-miss healing, re-terminal
+  reclassifications). (3) Recomputes the daily FTR aggregates idempotently.
+* `DeviceJourneyAggregates` (PK = TenantId or `global`, RK = `{yyyy-MM-dd}`): per day —
+  completed journeys, first-time-right count, attempt histogram, junk-serial exclusion
+  count, `JourneyVersion`. A journey buckets on the StartedAt date of its completing
+  success session (the platform's StartedAt-date convention; late terminals converge via
+  the sweep). All counts are **additive**, so a window rate is the ratio of summed daily
+  rows — there is no rolling-window row (unlike the median-based time-attribution
+  aggregates, which cannot merge). 180-day retention. Rows exist below the ≥20 UI gate
+  on purpose (the UI needs the n for "insufficient data (n=…)").
+
+# Surfaces
+
+* `GET api/metrics/device-history?serialNumber=&sessionId=` (MemberRead + tenant query
+  param — the same route serves the web banner and MCP; a Global Admin passes `tenantId`)
+  — the chain plus, with `sessionId`, that session's server-computed attempt number
+  (live sessions get their would-be position via a virtual non-successful attempt, so the
+  redeploy and gap rules apply). `history: null` = no recorded history.
+* `GET api/metrics/device-journeys?days=` (MemberRead) /
+  `GET api/global/metrics/device-journeys?days=&tenantId=` (GlobalReadOrAdmin) —
+  `{ windowDays, totals, daily, repeatDevices }`. The days selector is honored because
+  counts are additive. Without `tenantId` the global variant serves the cross-tenant
+  aggregate rows and `repeatDevices: null` — a per-device list would require scanning
+  every tenant's partition.
+* Repeat devices (violator list, capped 10): devices whose LAST journey took ≥2 attempts
+  and whose newest terminal session falls inside the window, ordered by attempts then
+  recency; the failure reason comes from the newest failed attempt (bounded point-reads).
+* Web: "Attempt N for this device · View history" banner on session detail (expandable
+  chain); "First-time-right" section on Fleet Health (rate + weekly trend + attempt
+  histogram + repeat devices). MCP: `get_device_history` (serial mode / fleet FTR mode).
+
+# Examples
+
+A device fails ESP twice and succeeds on the third try within one week: one journey,
+three attempts, completed — counts once in the FTR denominator, not in the numerator,
+and the histogram's bucket `3` gains one. The same device redeployed a month later and
+succeeding immediately starts journey 2 with attempt count 1 — first-time-right. A
+device with `serialNumber = "Unknown"` (WMI failure) never gets a chain; its terminal
+sessions surface only as that day's disclosed exclusion count.
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Helpers/DeviceJourneyCalculator.cs` — device key, junk list, grouping, attempt numbers (JourneyVersion 1).
+* `src/Backend/AutopilotMonitor.Functions/Services/TableStorageService.DeviceHistory.cs` — persistence + inline seam entry.
+* `src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.DeviceJourneys.cs` — sweep, tombstone pruning, daily aggregation.
+* `src/Backend/AutopilotMonitor.Functions/Functions/Metrics/GetDeviceJourneyFunctions.cs` — read surfaces.
+* `src/Backend/AutopilotMonitor.Functions.Tests/DeviceJourneyAndFtrTests.cs` — table-driven grouping/aggregation/attempt-number tests.
+* `tasks/insights-expansion-spec.md` — F2 specification and source-data audit (§0.5).

docs/backend/rule-regression-radar.md

@@ -0,0 +1,112 @@
+---
+type: Concept
+title: Rule Regression Radar (F3)
+description: How the radar detects analyze rules whose hit rate suddenly rose (7-day window vs 28-day baseline, Wilson-separated), which suppressions keep it quiet, how alert episodes live in the notification tracker, and which surfaces show them.
+resource: src/Backend/AutopilotMonitor.Functions/Helpers/RuleRegressionRadar.cs
+tags:
+  - backend
+  - rules
+  - regression-radar
+  - insights
+timestamp: 2026-07-27T16:00:00+02:00
+---
+
+# Rule Regression Radar (F3)
+
+Detects "this analyze rule suddenly fires much more often" before helpdesk volume does, and
+says what the spike correlates with — with honest statistics and honest wording. Analyze
+rules only in v1: their daily stats are audit-verified session-deduplicated
+(`FireCount` ≈ distinct sessions with a hit, `SessionsEvaluated` the same-source
+denominator). Gather rules are excluded until their per-batch stats dedup is fixed —
+never silently included with bad math.
+
+# Schema
+
+## Signal & detection
+
+Per (tenant, ruleId), daily hit rate = `FireCount ÷ SessionsEvaluated` from the
+`RuleStats` rows — RATES, not counts (absolute counts spike with every rollout wave).
+A regression fires only when ALL hold for the trailing 7-day window vs. the prior
+28-day baseline:
+
+1. window has ≥ 5 hit sessions AND ≥ 20 evaluated sessions;
+2. window rate ≥ 2× baseline rate;
+3. the Wilson 95 % intervals are disjoint in the increase direction
+   (`MetricsMath.WilsonInterval` / `RateIncreaseSeparated` — deterministic, unit-pinned),
+   so small-n noise never alerts.
+
+## Suppression (false positives are trust damage)
+
+* Rule entity missing (deleted) — never alerts.
+* Rule younger than baseline + window (`CreatedAt` within 35 days — grace period for new
+  rules; the entity timestamp is the honest gate, a stats-row count would fail on
+  low-volume tenants whose quiet days simply have no rows).
+* Rule edited inside the window (`UpdatedAt` ≥ window start) — edits legitimately change
+  hit rates.
+* Empty baseline denominator (rule disabled throughout the baseline).
+* Platform kill switch: app setting `RuleRegressionRadarDisabled=true` skips the radar
+  (fail-open — it only notifies, never mutates data).
+
+## Alert episodes (tracker keyspace)
+
+An episode lives as `ruleregression|{ruleId}` in the notification-tracker table
+(the `tpmpss|` pattern): the row IS the dedup (bell + ops event fire exactly once per
+episode), the badge state and the `regressions[]` payload — window/baseline counts,
+rates, lift (null for a zero-baseline "new signal" — no finite lift is invented), window
+dates, and the dimension concentration captured at first fire. Numbers refresh on every
+radar pass while the episode stays active; `FirstNotifiedAt` never moves. The episode
+closes (badge clears, dedup re-arms) when the window rate falls back under 1.5× baseline
+or fires stop entirely; the tracker's 30-day retention sweep also re-arms, so a
+month-old still-burning regression rings once more by design.
+
+## Dimension correlation (on fire only)
+
+Hit sessions (RuleResults scan: tenant partition range + RowKey = ruleId + DetectedAt in
+window, intersected with the window's sessions) are compared against all window sessions
+across `osBuild`, `model` (manufacturer + model), `agentVersion` and `imeVersion`. The
+top value by lift is reported with BOTH shares — wording contract: "X % of affected
+sessions are on … vs Y % of all sessions — correlated, not necessarily causal". No value
+with ≥ 5 hit sessions at ≥ 2× lift → "no clear dimension concentration"; a correlation
+failure also yields no claim. App-version correlation for `ANALYZE-APP-*` rules is a
+follow-up (needs rule→app matching semantics), tracked with the gather-rule fix.
+
+## Compute & alert path
+
+Runs in the daily maintenance pass right after the rule-stats aggregation
+(`RunRuleRegressionRadarAsync`), anchored on YESTERDAY — whole days only, a partial
+today would understate the window rate. One 35-day stats read covers all tenants;
+evaluation is the pure `RuleRegressionRadar` core. On a NEW episode:
+
+1. `RuleFrequencyRegression` ops event (category Tenant, Warning) with the full numbers
+   — dual-registered in the web `OPS_EVENT_TYPES` catalog for alert-rule routing.
+2. Tenant bell notification (type `rule_frequency_regression`, Admin audience) with the
+   same numbers and a deep link to the rule's card (`/analyze-rules#rule-card-{ruleId}`).
+   A sessions-list pre-filtered to rule hits needs a rule filter the session search does
+   not have yet — follow-up alongside the gather-rule work.
+3. v1 channel: bell + ops event only. Provider fan-out (webhook/Telegram) follows once
+   alert quality is proven — a noisy first version burns trust permanently.
+
+# Surfaces
+
+* `GET api/metrics/rule-stats` (and the global variant with `tenantId`) carries
+  `regressions[]` — the active episodes. MCP `get_rule_stats` inherits the block.
+* Analyze-rules page: per-rule 30-day fire-count sparkline (densified from the `trend`
+  rows the endpoint already served) and a red "↑ Regression" badge while an episode is
+  active, with the full numbers in the tooltip.
+
+# Examples
+
+A rule firing on 2 % of sessions for a month jumps to 15 % over a week (15/100 vs
+10/500): lift 7.5×, Wilson-separated → one bell, one ops event, badge until the rate
+falls back under 3 %. The same jump on 5/20 sessions with a 12 % baseline stays silent —
+the intervals overlap, so the spike is not statistically real yet. A rule that never
+fired in the baseline and now hits 15 % alerts as a "new signal" with no lift figure.
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Helpers/RuleRegressionRadar.cs` — gates, suppression, re-arm, dimension concentration.
+* `src/Backend/AutopilotMonitor.Functions/Helpers/MetricsMath.cs` — Wilson interval + separation primitive.
+* `src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.RuleRegression.cs` — sweep, episode reconcile, bell/ops wiring.
+* `src/Backend/AutopilotMonitor.Functions/DataAccess/TableStorage/TableHardwareRejectionNotificationTracker.cs` — episode keyspace.
+* `src/Backend/AutopilotMonitor.Functions.Tests/RuleRegressionRadarTests.cs` — hand-computed Wilson vectors + every suppression branch.
+* `tasks/insights-expansion-spec.md` — F3 specification and source-data audit (§0.5).

docs/backend/server-actions-on-demand-diagnostics.md

@@ -0,0 +1,102 @@
+---
+type: Concept
+title: ServerActions & On-Demand Diagnostics Collection
+description: The queue-on-session, deliver-on-ingest ServerAction channel (terminate_session / rotate_config / request_diagnostics), the operator-tier authorization split, the portal's Collect Logs flow including the admin quick-config path, and why the offboarding cascade sweeps the hosted diagnostics prefix.
+resource: /src/Backend/AutopilotMonitor.Functions/Functions/Sessions/QueueSessionActionFunction.cs
+tags:
+  - backend
+  - agent
+  - web
+  - server-actions
+  - diagnostics
+  - authorization
+timestamp: 2026-07-24T00:00:00+02:00
+---
+
+# Schema
+
+**Channel.** A `ServerAction` (`Type`, `Reason`, `RuleId`, `Params`, `QueuedAt`) is queued onto
+`Sessions.PendingActionsJson` by server-side writers — the admin endpoint
+`POST /api/sessions/{sessionId}/actions`, the RuleEngine, maintenance — and delivered
+piggy-backed on the agent's next ingest response. No polling channel exists; delivery latency
+equals the agent's telemetry cadence (seconds to ~1 minute during enrollment). Delivery is
+at-least-once: the agent's `ServerActionDispatcher` dedups on `Type|RuleId|QueuedAt` and
+processes a batch sequentially in queue order. Unknown types are logged and skipped, so new
+types roll out server-first. Every action emits `server_action_received` and then
+`server_action_executed` or `server_action_failed` into the session timeline.
+
+**Authorization (two layers).** The route sits on `EndpointPolicy.TenantAdminOrOperator`
+(tenant Admin, tenant Operator, or Global Admin — no Viewer, no Global Reader: it is a write
+tier). Because the route serves multiple action kinds, `QueueSessionActionFunction`
+re-gates per type via `IsTypeAllowedForCaller`: Operators may queue only
+`request_diagnostics`; `terminate_session` and `rotate_config` require
+`RequestContext.IsTenantAdmin || IsGlobalAdmin` (403 otherwise). Any route on this tier with
+a multi-kind body MUST follow this same in-function re-gating pattern.
+
+**On-demand collection (`request_diagnostics`).** The dispatcher calls
+`DiagnosticsPackageService.CreateAndUploadAsync(enrollmentSucceeded: false, suffix:
+"server-requested")` — producing `AgentDiagnostics-{sessionId}-{timestamp}-server-requested.zip`.
+Passing `enrollmentSucceeded: false` is what makes on-demand collection work under
+`DiagnosticsUploadMode=OnFailure`; only `Off` (or an unconfigured destination) blocks it,
+surfacing as `server_action_failed`. The `diagnostics_uploaded` event stamps
+`DiagnosticsBlobName` + `Destination` onto the Sessions row (latest wins — an on-demand name
+is later replaced by the terminal package), and the SignalR `sessionUpdate` delta carries the
+field, so the portal's Download button flips live.
+
+**Portal flow (Collect Logs, session detail header).** The button is always rendered; the
+disabled states carry tooltips (inactive session, cross-tenant view, missing role, diagnostics
+not configured) as a deliberate nudge. Members below Admin cannot read the tenant config, so
+the member-readable feature-flags endpoint exposes `diagnosticsUploadConfigured`
+(`mode != Off && (customer SAS present || destination Hosted)`). Progress tracking is
+sequence-based (backend-assigned, monotonic), never timestamp-based — agent clock skew and
+at-least-once redelivery of older actions cannot produce false matches
+(`collectLogsLogic.ts`). Cross-tenant views are gated off because `TenantHelper.GetTenantId`
+resolves the tenant strictly from the caller's JWT.
+
+**Admin quick-config.** For an unconfigured tenant, a Tenant Admin's click opens a dialog
+that switches the tenant to `Destination=Hosted`, `Mode=OnFailure` (GET full config → patch
+exactly those two fields → PUT verbatim), then queues `rotate_config` **before**
+`request_diagnostics` — the dispatcher's sequential ordering guarantees the agent refetches
+the now-enabled config (the `RemoteConfigMerger` applies diagnostics fields live, remote wins)
+before building the package. This works against the already-deployed agent fleet; a
+"one-time override" variant would not, because the `mode=Off` check is client-side in
+`CreateAndUploadAsync` and the deployed handler reads no params.
+
+**Offboarding cascade.** Latest-wins on `DiagnosticsBlobName` means a session can leave more
+hosted blobs behind than the deletion manifest references. The cascade's Hosted branch
+therefore sweeps the canonical prefix `{tenantId}/AgentDiagnostics-{sessionId}-`
+(`HostedDiagnosticsBlobService.DeleteBySessionPrefixAsync`) plus the manifest-named blob as
+belt-and-braces; a non-GUID legacy `SessionId` degrades to the single-name delete instead of
+poisoning the cascade. The CustomerSas branch is unchanged (delete only with the `d`
+permission; the customer's lifecycle rules stay authoritative).
+
+# Examples
+
+Queue an on-demand collection (Admin or Operator, own tenant):
+
+```
+POST /api/sessions/{sessionId}/actions
+{ "type": "request_diagnostics", "reason": "On-demand log collection from portal" }
+→ 202 { "success": true, "queuedAt": "..." }
+```
+
+Operator queueing an admin-only type:
+
+```
+POST /api/sessions/{sessionId}/actions
+{ "type": "terminate_session" }
+→ 403 { "message": "Action type 'terminate_session' requires the Tenant Admin role" }
+```
+
+Timeline for a successful round trip: `server_action_received` → `server_action_executed`
+(with `blobName`) → `diagnostics_uploaded` → Sessions row + SignalR delta carry the new
+`DiagnosticsBlobName`.
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Functions/Sessions/QueueSessionActionFunction.cs` — allowed types, per-type caller gate.
+* `src/Backend/AutopilotMonitor.Functions/Security/EndpointAccessPolicyCatalog.cs` — `TenantAdminOrOperator` tier definition.
+* `src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Runtime/ServerActionDispatcher.cs` — dedup, sequential dispatch, telemetry contract.
+* `src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Runtime/DiagnosticsPackageService.cs` — mode gates, archive build, suffix.
+* `src/Backend/AutopilotMonitor.Functions/Services/Diagnostics/DiagnosticsBlobCascadeDeleter.cs` — hosted prefix sweep.
+* `src/Web/autopilot-monitor-web/app/sessions/[sessionId]/components/collectLogsLogic.ts` — button gating matrix, sequence-based progress evaluation.

docs/backend/time-attribution.md

@@ -0,0 +1,124 @@
+---
+type: Concept
+title: Enrollment Time Attribution (F1)
+description: How a terminal session's wall clock is partitioned into named segments with an explicit unattributed remainder, how ESP-blocking apps and reboots are attributed, and where breakdowns and fleet aggregates are computed, stored and served.
+resource: src/Backend/AutopilotMonitor.Functions/Helpers/TimeAttributionCalculator.cs
+tags:
+  - backend
+  - metrics
+  - time-attribution
+  - insights
+timestamp: 2026-07-26T20:00:00+02:00
+---
+
+# Enrollment Time Attribution (F1)
+
+Answers "where did the 47 minutes go?" per session and per fleet. Every number is
+traceable to observed events; a missing signal produces "unknown" or an explicit
+unattributed remainder — never a guess, and the partition is never normalized to 100 %.
+
+# Schema
+
+## Invariant
+
+For every computed breakdown:
+
+```
+sum(span seconds) + UnattributedSeconds == WallClockSeconds == session DurationSeconds
+```
+
+`DurationSeconds` is the authoritative wall clock — for WhiteGlove it is part 1 + part 2
+WITHOUT the pause, and re-terminal stamps move `CompletedAt`, so `CompletedAt − StartedAt`
+is never used (it diverges in ~25 % of terminal sessions). Span seconds are floored;
+sub-second dust lands in the unattributed remainder.
+
+## Observation windows
+
+Windows are end-anchored on authoritative timestamps and constructed so their lengths sum
+exactly to `DurationSeconds`:
+
+* Non-WhiteGlove: one window `[CompletedAt − DurationSeconds, CompletedAt]`.
+* WhiteGlove (ONE session row, two windows): part 2 = `[ResumedAt, CompletedAt]`; part 1 =
+  `[part1End − (DurationSeconds − part2), part1End]` where `part1End` is anchored on the
+  `whiteglove_part1_complete` event. The pause between the windows is never attributed.
+  Fallback anchors set the `WhiteGloveAnchorsIncomplete` quality flag.
+
+## Segments
+
+Spans are sliced at phase-declaration events (only declaration events carry
+`Phase != Unknown`) plus `desktop_arrived`, in canonical Sequence order; anchors whose
+timestamp runs backward are dropped (`ClockSkewDropped`).
+
+| EnrollmentPhase | Segment key |
+|---|---|
+| Start / DevicePreparation / DeviceSetup | `device_prep` |
+| AppsDevice | `esp_apps` |
+| AccountSetup / FinalizingSetup | `identity_hello` |
+| AppsUser | `user_esp` |
+| Complete, `desktop_arrived` event | `desktop_handoff` |
+| Failed | attribution ends — tail stays unattributed |
+
+## ESP-blocking apps (positive evidence only)
+
+Per-app intervals use EVENT timestamps (first started/download event → LAST terminal
+event, covering IME retries) — never the agent payload timing, which freezes at the first
+terminal transition. Blocking membership joins the app id against the session's latest
+list-carrying `esp_config_detected` emission: listed ⇒ blocking; absent ⇒ unknown, never
+false (`BlockingSetUnknown` / `BlockingSetTruncated` flags). Critical-path occupancy =
+overlap-merged union of blocking intervals clipped to the `esp_apps` spans;
+`EspAppsOccupancySeconds` is null (unknown) when no lists were observed.
+
+The fleet what-if bound per app X is `max(0, cpEnd − cpEndWithoutX)` (latest interval end
+with vs. without X; the idle gap before X counts — the ESP also waits for X to start).
+It is an upper bound by construction; every consumer says "up to", never "you will save".
+
+## Reboots
+
+`system_reboot_detected` is detection-time-stamped by the NEXT agent run; the real boot
+moment is its `lastBootUtc` payload. Reboot spans are therefore the event-stream gap
+bracketing `lastBootUtc`, deduped per boot, clipped to the observation windows, and carried
+as a cross-cutting annotation (they overlap the segment they started in — never a slice of
+the partition).
+
+## Persistence & compute
+
+* `SessionTimeBreakdowns` (PK = TenantId, RK = SessionId): spans, reboot spans, blocking
+  intervals (top 20 + uncapped count), quality flags, `AttributionVersion`. Written once at
+  the session-terminal single-writer seam (`TableSessionRepository.UpdateSessionStatusAsync`,
+  Succeeded/Failed only — Incomplete deliberately has no duration). Deleted with its session
+  (deletion-manifest step) and on tenant offboarding.
+* Self-healing 30-day maintenance sweep (2h tick): backfills missing breakdowns, folds
+  late-terminating sessions into their StartedAt-date aggregates, and enforces version
+  purity — a stale-version row whose events aged out counts as missing, never mixed.
+* `TimeAttributionAggregates` (PK = TenantId or `global`, RK = `{yyyy-MM-dd}|{class}` and
+  `rolling30|{class}`): per enrollment class (`user_driven`, `whiteglove`, `self_deploying`,
+  `device_preparation` — never mixed) median/p75/p90 per segment over the fixed six-segment
+  stack (incl. `unattributed`; a session without a segment contributes 0), plus top blocking
+  apps (≥5 sessions per app row). Clean/flagged/missing counts are part of every row; rows
+  exist below the ≥20 UI gate on purpose (the UI needs the n for "insufficient data (n=…)").
+  Daily rows retain 180 days; `rolling30` rows are refreshed whole (a median of per-day
+  medians is not the range median — the rolling rows are the honest range statistics).
+
+## Surfaces
+
+* `GET api/sessions/{sessionId}/time-attribution` (MemberRead + tenant query param) —
+  breakdown row or `breakdown: null` (pre-feature / non-terminal / Incomplete = no lane).
+* `GET api/metrics/time-attribution` (MemberRead) / `GET api/global/metrics/time-attribution`
+  (GlobalReadOrAdmin, optional `tenantId`) — `{ windowDays, classes (rolling), daily }`.
+* Web: attribution lane under the session-detail phase timeline; "Time attribution" section
+  on Fleet Health. MCP: `get_time_attribution` (sessionId mode / fleet mode).
+
+# Examples
+
+A 30-minute user-driven session: `device_prep` 5m → `esp_apps` 10m (occupancy 4m by one
+blocking app) → `identity_hello` 8m → `user_esp` 5m → `desktop_handoff` 2m, unattributed 0.
+A WhiteGlove session with a 2-day pause reports wall clock = part1 + part2 only; the resume
+gap before the first part-2 phase declaration is unattributed, not guessed.
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Helpers/TimeAttributionCalculator.cs` — calculator + semantics (AttributionVersion 1).
+* `src/Backend/AutopilotMonitor.Functions/Services/TableStorageService.TimeAttribution.cs` — persistence.
+* `src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.Aggregation.cs` — sweep + aggregation.
+* `src/Backend/AutopilotMonitor.Functions.Tests/TimeAttributionCalculatorTests.cs` — golden fixtures pinning the invariant.
+* `tasks/insights-expansion-spec.md` — F1 specification and source-data audit (§0.5).

docs/index.md

@@ -22,7 +22,7 @@ https://docs.autopilotmonitor.com (separate repository).
 * [Agent Endpoint Migration](agent/endpoint-migration.md) - How the backend re-homes agents to a new API base URL via the config channel (backend move or per-tenant region move): live-fetch-only, allowlist-validated on both sides, one hop, kill wins.
 * [MDM Reboot Coalescing](agent/mdm-reboot-coalescing.md) - Attributing the mid-ESP coalesced reboot ("second sign-in") to the device-assigned policy URIs that forced it: aggregated DM-Enterprise 2800 observations (neutral, flush-time watermark) + the ANALYZE-ESP-005 advisory gated on an actually-observed reboot.
 * [Secure Boot CA-2023 Detection](agent/secureboot-ca2023-detection.md) - Why ANALYZE-SEC-001 v3 reads the UEFI db/KEK variables directly instead of trusting the SecureBoot\Servicing registry, and how the one-sided uefiCa2023FirmwareConfirmed marker plus a not_exists precondition give absence-tolerant suppression across old and new agents.
-* [ESP Policy-Provider Stall Detection](agent/esp-policy-provider-stall.md) - The EnrollmentStatusTracking CSP's no-timeout Setup/Apps wait: a registered provider that never sets TrackingPoliciesCreated (co-management "ConfigMgr" without the IME's "Sidecar") parks the user ESP at "Apps (Identifying)" for days; the always-on esp_policy_provider_stalled tripwire and the SOFTWARE\Microsoft\Windows\Autopilot gather-rule allowlist entry that surface it.
+* [ESP Policy-Provider Stall Detection](agent/esp-policy-provider-stall.md) - The EnrollmentStatusTracking CSP's no-timeout Setup/Apps wait is keyed to the Intune-registered "Sidecar" provider by NAME: co-management's "ConfigMgr" parks the user ESP at "Apps (Identifying)" for days even with TrackingPoliciesCreated=1 (issue #106); the always-on two-arm esp_policy_provider_stalled tripwire (provider_incomplete | sidecar_provider_missing) and the SOFTWARE\Microsoft\Windows\Autopilot gather-rule allowlist entry that surface it.
 
 # Rules
 
@@ -32,11 +32,19 @@ https://docs.autopilotmonitor.com (separate repository).
 # Backend
 
 * [Business Timestamps](backend/business-timestamps.md) - Why the Azure Tables system Timestamp is never authoritative (migrations reset it), and the OccurredUtc + RowKey-decode compensation for AuditLogs/OpsEvents/Events after the 2026-07-18 migration.
+* [Device Journeys & First-Time-Right](backend/device-journeys.md) - How a device's terminal enrollment attempts are grouped into journeys (trim+casefold device key with junk-serial exclusion, first-success end, 30-day gap, WhiteGlove = one attempt), the DeviceHistories chain with its tombstone-driven deleted-session pruning, the additive daily FTR aggregates, and the banner/fleet/MCP surfaces.
+* [Enrollment Time Attribution](backend/time-attribution.md) - How a terminal session's authoritative wall clock (DurationSeconds, WhiteGlove pause excluded) is partitioned into named segments with an explicit unattributed remainder, ESP-blocking apps as positive evidence with "up to" what-if bounds, reboot spans from event gaps, and the breakdown/aggregate tables the sweep maintains.
+* [Rule Regression Radar](backend/rule-regression-radar.md) - How analyze-rule hit-rate regressions are detected (7-day window ≥2× the 28-day baseline with disjoint Wilson intervals), which suppressions keep the radar quiet (grace period, edit-in-window, kill switch), how alert episodes live in the ruleregression tracker keyspace with re-arm semantics, and the on-fire dimension correlation with its "correlated, not necessarily causal" wording contract.
+* [ServerActions & On-Demand Diagnostics](backend/server-actions-on-demand-diagnostics.md) - The queue-on-session, deliver-on-ingest action channel, the TenantAdminOrOperator tier with in-function per-type re-gating, the portal's Collect Logs flow (quick-config orders rotate_config before request_diagnostics), and the hosted-prefix sweep in the offboarding cascade.
 
 # MCP Server
 
 * [MCP Docs Corpus](mcp/docs-corpus.md) - How the published customer documentation becomes the `search_docs` corpus: CRLF-safe heading chunking, build-time embedding as a third precomputed section, and why a docs edit needs an MCP redeploy.
 
+# Web Portal
+
+* [Portal Navigation Prefetch & the SWA Runtime Bottleneck](web/portal-navigation-prefetch.md) - Why every navigation Link sets `prefetch={false}`: the App Router's viewport prefetch fans out one RSC request per visible link against an HTTP/1.1 Static Web Apps runtime, and its latency tail starves the six connections that also serve the document and the route chunks. Includes where browser telemetry actually lives.
+
 # Security & Identity
 
 * [MCP OAuth Flow](mcp-oauth-flow.md) - Who authenticates where when connecting an AI client to the MCP server; two identities, three parties plus Entra ID.

docs/log.md

@@ -1,6 +1,22 @@
 # Log
 
-## 2026-07-23
+## 2026-07-27
+
+* **Creation**: Added `web/portal-navigation-prefetch.md` (and a `Web Portal` section to the index) — root cause of the minute-long portal freezes. `GlobalSidebar` and `Navbar` render the whole navigation on every page, so the App Router's default viewport prefetch issued one `?_rsc=` request per visible link against `portal.autopilotmonitor.com` (~10,400 in 14 days). That origin serves the document, the route chunks and the RSC payloads over **HTTP/1.1**, so the six-connection-per-origin cap applies; when the Static Web Apps managed runtime queued the prefetches (p50 91 ms, p95 896 ms, but max 234 s / 21 samples over 30 s), the in-flight prefetches held every socket and the navigation the user wanted could not get one. Measured on the 2026-07-27 17:25 CEST incident: `/admin/presence?_rsc=` 151 s, three siblings finishing within 340 ms of each other at ~37.5 s, all `200`, while the backend API answered in 1.6–2.2 s and `browserTimings` showed `networkDuration` 16 ms against `processingDuration` 225 s. Every navigation Link now sets `prefetch={false}` — these routes are client components behind `ProtectedRoute` that fetch their own data, so the prefetched RSC payload carried no benefit. The doc also records that browser telemetry lives in a **separate** App Insights resource (`AUTOPILOT_MONITOR_WEB_APPINSIGHTS_ID`) that the MCP `query_backend_logs` tool cannot reach, which is why the freezes left no trace in the backend logs.
+
+* **Creation**: Added `backend/rule-regression-radar.md` — F3 regression radar (analyze rules only; gather rules wait for their per-batch stats dedup fix): per (tenant, rule) daily hit rate from `RuleStats`, firing only when the trailing 7-day window has ≥5 hit sessions and ≥20 evaluated sessions, the rate is ≥2× the prior 28-day baseline AND the Wilson 95 % intervals are disjoint (deterministic `MetricsMath` primitive with hand-computed test vectors). Suppressed for deleted rules, rules younger than baseline+window (`CreatedAt` grace), rules edited inside the window (`UpdatedAt`), empty baselines, and the `RuleRegressionRadarDisabled` app-setting kill switch. Alert episodes live as `ruleregression|{ruleId}` tracker rows (one bell + one `RuleFrequencyRegression` ops event per episode; numbers refreshed while active; re-armed under 1.5× baseline, on fire-stop, or by the 30-day tracker retention). On-fire dimension correlation over osBuild/model/agentVersion/imeVersion with the "correlated, not necessarily causal" wording contract. Surfaces: `regressions[]` in the rule-stats response (MCP inherits), the analyze-rules page's 30-day sparkline and regression badge.
+
+* **Creation**: Added `backend/device-journeys.md` — F2 device history / First-Time-Right: the device key (trim + case-fold serial, junk/placeholder exclusion incl. the agent's `Unknown` WMI sentinel), journey grouping over the terminal set Succeeded/Failed/Incomplete (first success ends a journey, a >30-day gap from the previous attempt's CompletedAt starts one, WhiteGlove's single session row is ONE attempt, open sessions are never attempts), the `DeviceHistories` chain (cap 20, written inline at the terminal seam, healed by the sweep, deleted-session refs pruned tombstone-driven, deliberately not in the per-session deletion manifest) and the additive `DeviceJourneyAggregates` daily rows (window rate = ratio of sums — no rolling row needed, 180-day retention). Surfaces: the session-detail "Attempt N" banner with server-computed attempt numbers (virtual-attempt rule for live sessions), the Fleet Health First-time-right section (≥20-journey gate, attempt histogram, repeat-devices violator list), and the MCP `get_device_history` tool (serial mode / fleet FTR mode).
+
+## 2026-07-25
+
+* **Update**: `agent/secureboot-ca2023-detection.md` — review follow-up on 90012d9 hardened the privilege scope, the repo's only privilege-adjustment code and therefore the precedent for any future one. The process token moved from a raw `IntPtr` to a `SafeHandle` assigned by the P/Invoke marshaler, so the CLR owns it from the moment `OpenProcessToken` returns and the critical finalizer releases it even when `Dispose` never runs — previously an async exception between the successful open and the caller's `using` leaked the handle *with the privilege enabled and no owner left to restore it*. `Dispose` became idempotent and now releases the handle in a `finally`, so a throwing restore cannot strand it; the restore obligation is recorded immediately after the successful `AdjustTokenPrivileges` with nothing throwable in between. The whole enable → read → restore sequence runs under a static gate because the privilege is adjusted on the *process* token: a concurrent reader would see "already enabled", correctly skip its own restore, and then lose the privilege mid-read when the first reader disposed, reporting `privilege_denied` on a machine that holds it.
+
+## 2026-07-24
+
+* **Creation**: Added `backend/server-actions-on-demand-diagnostics.md` — the ServerAction channel (queue on `Sessions.PendingActionsJson`, deliver piggy-backed on the next ingest response, at-least-once with `Type|RuleId|QueuedAt` dedup and sequential dispatch) and the on-demand log collection built on it: the portal's always-visible **Collect Logs** button, the new `EndpointPolicy.TenantAdminOrOperator` tier with the mandatory in-function per-type re-gate (Operators queue only `request_diagnostics`), the member-readable `diagnosticsUploadConfigured` feature flag, the admin quick-config path that must queue `rotate_config` before `request_diagnostics` (works against the deployed fleet — a one-time override would not, the mode gate is agent-side), sequence-based progress matching immune to agent clock skew, and the offboarding cascade's hosted-prefix sweep closing the latest-wins orphan-blob gap.
+
+* **Update**: `agent/esp-policy-provider-stall.md` — the customer's follow-up on issue #106 corrected the field-case mechanics: `TrackingPoliciesCreated=1` WAS set under the foreign `ConfigMgr` provider, so the ESP ignores a foreign provider's completion values and waits on the Intune-registered `Sidecar` provider by name (the rename to `Sidecar` creates no tracking entries yet unblocks instantly). The completeness-keyed detector would have missed exactly this — the incident it was built for. The tripwire is now two-armed under one event type discriminated by `reason`: `provider_incomplete` (unchanged dwell on continuously-incomplete providers) and `sidecar_provider_missing` (≥ 1 Setup/Apps provider registered without `Sidecar`, completeness deliberately ignored, clock reset when Sidecar registers later). Sidecar-missing wins the `reason` when both arms cross in one round; the cross-restart fingerprint gains a `sidecarMissing` sentinel; payload adds `reason` and `sidecarMissingForMinutes`. Evidence for the name-keyed wait: Niehaus' documented Intune WAP payload always registers `Sidecar` (ConfigMgr only via a co-management settings policy) and Microsoft's ESP troubleshooting models app tracking exclusively through the Sidecar subkey.
 
 * **Creation**: Added `agent/esp-policy-provider-stall.md` — a customer diagnosed that pre-provisioned devices with the ConfigMgr client as a blocking app occasionally park the user ESP at "Apps (Identifying)" for days: only a `ConfigMgr` provider is registered under `EnrollmentStatusTracking\Device\Setup\Apps\PolicyProviders` (no `Sidecar` = IME) and `TrackingPoliciesCreated` never becomes 1 — a wait the CSP defines without any timeout; renaming the key to `Sidecar` unblocks immediately. The agent now ships the always-on `esp_policy_provider_stalled` tripwire (60-s tick, 15-min continuous-incompleteness dwell, one-shot per provider, `StartupEventGate` fingerprint across restarts, purely observational via the InformationalEvent rail — no toggle, no ConfigVersion bump), generalized to any provider in both `Setup/Apps` and `DevicePreparation` kinds rather than hardcoding the ConfigMgr case. Companion change: `SOFTWARE\Microsoft\Windows\Autopilot` joins the gather-rule registry allowlist so tenants can target the subtree with custom rules (e.g. `not_exists` on the Sidecar key).
 
@@ -44,6 +60,10 @@
 
 * **Creation**: Added `agent/endpoint-migration.md` — config-channel agent re-home (`MigrateToApiBaseUrl` on `GET agent/config`, ConfigVersion 33): backend or per-tenant API-URL moves without DNS indirection (impossible due to mTLS/Flex-TLS). Live-fetch-only with cache strip, `AgentEndpointMigrationRules` allowlist validated on both sides, one hop, kill-switch wins. Built as the lesson from the 2026-07-18 WEU→GWC cutover, which hard-cut all in-flight agents.
 
+## 2026-07-26
+
+* **Creation**: Added `backend/time-attribution.md` — the F1 enrollment time attribution: exact-partition invariant against the session's authoritative DurationSeconds, two-window WhiteGlove model, segment mapping from phase-declaration events, positive-evidence ESP-blocking join with "up to" what-if bounds, reboot spans from event gaps around `lastBootUtc`, and the SessionTimeBreakdowns / TimeAttributionAggregates tables with their terminal-seam + maintenance-sweep compute paths and read surfaces (web lane, fleet panel, MCP `get_time_attribution`).
+
 ## 2026-07-17
 
 * **Creation**: Added `agent/build-counter-blob.md` — agent build numbers now come from a shared public counter blob (`buildcounter-v{N}.txt`, reserve-before-build with ETag-CAS) so local `build.ps1` and CI `build-agent.yml` builds can never mint the same number. Documents the CI parity fixes (SummaryDialog.exe.config in the ZIP, EXE-vs-manifest version check, hash-oracle update gated on stable cutover) and the Sigstore provenance attestation on CI-built ZIPs.

docs/web/portal-navigation-prefetch.md

@@ -0,0 +1,88 @@
+---
+type: Concept
+title: Portal Navigation Prefetch & the SWA Runtime Bottleneck
+description: Why every navigation Link in the portal sets prefetch={false} — the App Router's viewport prefetch fires one RSC request per visible link against an HTTP/1.1 Static Web Apps runtime, and its latency tail starves the connection pool that serves the document and the route chunks.
+resource: src/Web/autopilot-monitor-web/components/GlobalSidebar.tsx
+tags:
+  - web
+  - performance
+  - nextjs
+  - static-web-apps
+  - observability
+timestamp: 2026-07-27T18:00:00+02:00
+---
+
+# Context
+
+The portal is a Next.js 15 App Router application hosted on Azure Static Web Apps. Every page under
+the authenticated shell is a client component behind `ProtectedRoute` that fetches its own data from
+the Function App (`autopilotmonitor-api-eu`) after mount. The RSC payload a route returns on
+navigation therefore carries no page data — only the shell.
+
+`GlobalSidebar` and `Navbar` render the full navigation on every page. With the App Router's default
+prefetch behaviour, each visible `<Link>` triggers a background `GET /<route>?_rsc=<hash>` as soon as
+it enters the viewport.
+
+# Mechanism
+
+Three properties combine into a portal-wide freeze:
+
+1. **Fan-out.** The navigation is fully rendered on each page load, so a single navigation issues one
+   RSC request per visible link — measured at ~10,400 requests over 14 days.
+2. **Shared origin over HTTP/1.1.** `portal.autopilotmonitor.com` serves the document, the JS route
+   chunks *and* the RSC payloads, and it negotiates HTTP/1.1 — not HTTP/2. Browsers cap concurrent
+   connections per origin at six.
+3. **A heavy latency tail on the SSR runtime.** RSC responses are prerendered and edge-cached
+   (`x-nextjs-cache: HIT`, `s-maxage=31536000`), so the median is fast. On a cache miss or cold start
+   the Static Web Apps managed runtime queues them.
+
+When the tail hits, the in-flight prefetches hold all six connections. The document request and the
+route chunks of the page the user actually wants then have no socket available, and the tab shows a
+loading spinner or a blank page until the queue drains. Nothing fails — the requests eventually
+return `200`, which is why the failure leaves no error trace anywhere.
+
+# Evidence
+
+Incident of 2026-07-27, 17:25–17:29 CEST. Browser-side dependencies during the freeze:
+
+| Request | Duration | Status |
+| --- | --- | --- |
+| `/admin/presence?_rsc=…` | 151,461 ms | 200 |
+| `/gather-rules?_rsc=…` | 52,283 ms | 200 |
+| `/progress?_rsc=…` | 37,878 ms | 200 |
+| `/analyze-rules?_rsc=…` | 37,559 ms | 200 |
+| `/audit?_rsc=…` | 37,539 ms | 200 |
+| Backend API, same window | 1,600–2,200 ms | 200 |
+
+The three ~37.5 s responses complete within 340 ms of each other — a queue draining, not independent
+slow requests. The corresponding `browserTimings` row for `/geographic-performance` reports
+`networkDuration` = 16 ms against `processingDuration` = 225,095 ms: the network delivered the
+document immediately and the browser then waited nearly four minutes for the `load` event.
+
+Fourteen-day distribution of RSC prefetches: n = 10,428, p50 = 91 ms, p95 = 896 ms, max = 234,230 ms,
+38 over 10 s, 21 over 30 s.
+
+# Diagnosing this class of problem
+
+The backend Application Insights resource is the wrong place to look — a freeze appears there as an
+*absence* of requests, never as an error. Browser telemetry lives in a **separate** Application
+Insights resource addressed by `AUTOPILOT_MONITOR_WEB_APPINSIGHTS_ID`, which the MCP
+`query_backend_logs` tool does not reach. Query it through
+`.claude/commands/scripts/query-appinsights.sh` with that ID exported, and read `dependencies`
+(client-side fetches, including RSC prefetches) and `browserTimings` (the network/processing split of
+a document load).
+
+# Contract
+
+Navigation `<Link>` elements in `GlobalSidebar` and `Navbar` set `prefetch={false}`. The prefetched
+RSC payload buys nothing for these routes, so the only remaining cost of a click is the route's JS
+chunk, which the browser caches after first use.
+
+Do not remove the prop to "speed up navigation" without first re-measuring the tail in the web
+Application Insights resource — the prefetch is cheap only while the runtime is warm.
+
+# Citations
+
+* `src/Web/autopilot-monitor-web/components/GlobalSidebar.tsx` — nav link rendering, rationale comment
+* `src/Web/autopilot-monitor-web/components/Navbar.tsx` — top-bar and notification links
+* [SWA and Oryx node pinning](../architecture.md) — hosting model for the web app

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/DeviceInfo/DeviceInfoCollectorEspRefreshTests.cs

@@ -0,0 +1,157 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Threading;
+using System.Threading.Tasks;
+using AutopilotMonitor.Agent.V2.Core.Logging;
+using AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals;
+using AutopilotMonitor.Agent.V2.Core.Orchestration;
+using AutopilotMonitor.Agent.V2.Core.Persistence;
+using AutopilotMonitor.Agent.V2.Core.Tests.Harness;
+using AutopilotMonitor.Agent.V2.Core.Tests.Orchestration;
+using AutopilotMonitor.DecisionCore.Engine;
+using AutopilotMonitor.DecisionCore.Signals;
+using Xunit;
+using SharedEventTypes = AutopilotMonitor.Shared.Constants.EventTypes;
+using DeviceInfoCollector = AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.DeviceInfo.DeviceInfoCollector;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.DeviceInfo
+{
+    /// <summary>
+    /// Audit Q2 — <see cref="DeviceInfoCollector.RefreshEspConfiguration"/>
+    /// re-reads the ESP blocking lists and re-emits <c>esp_config_detected</c> exactly when the
+    /// payload changed: the registry lists grow progressively (one timestamped subkey per CSP
+    /// status write) and the user-scope lists appear only after sign-in, so the early emissions
+    /// are structurally partial. The StartupEventGate keeps an unchanged re-read silent.
+    /// Probe overrides are static slots → serialized with the other probe-driven suites.
+    /// </summary>
+    [Collection("SerialThreading")]
+    public sealed class DeviceInfoCollectorEspRefreshTests
+    {
+        private static readonly DateTime T0 = new DateTime(2026, 7, 27, 10, 0, 0, DateTimeKind.Utc);
+
+        private const string DeviceApp1 = "11111111-1111-1111-1111-111111111111";
+        private const string DeviceApp2 = "22222222-2222-2222-2222-222222222222";
+        private const string UserApp1 = "33333333-3333-3333-3333-333333333333";
+
+        private static EspTrackingInfoSnapshot Snapshot(string[] deviceApps, string[] userApps)
+        {
+            var all = deviceApps.Concat(userApps).Distinct().ToArray();
+            return new EspTrackingInfoSnapshot(
+                msiProductCodes: Array.Empty<string>(),
+                modernAppPfns: Array.Empty<string>(),
+                win32AppIds: all,
+                userWin32AppIds: userApps,
+                msiCount: 0,
+                modernCount: 0,
+                win32Count: all.Length);
+        }
+
+        private static IReadOnlyList<FakeSignalIngressSink.PostedSignal> EspConfigEvents(FakeSignalIngressSink ingress) =>
+            ingress.Posted
+                .Where(p => p.Kind == DecisionSignalKind.InformationalEvent
+                            && p.Payload != null
+                            && p.Payload.TryGetValue(SignalPayloadKeys.EventType, out var et)
+                            && et == SharedEventTypes.EspConfigDetected)
+                .ToList();
+
+        private static IReadOnlyList<string> TypedList(FakeSignalIngressSink.PostedSignal signal, string key)
+        {
+            var data = Assert.IsAssignableFrom<IReadOnlyDictionary<string, object>>(signal.TypedPayload);
+            return Assert.IsAssignableFrom<IReadOnlyList<string>>(data[key]);
+        }
+
+        [Fact]
+        public void Refresh_reemits_when_lists_grew_and_stays_silent_when_unchanged()
+        {
+            using var tmp = new TempDirectory();
+            var logger = new AgentLogger(tmp.Path, AgentLogLevel.Info);
+            var ingress = new FakeSignalIngressSink();
+            var clock = new VirtualClock(T0);
+            var collector = new DeviceInfoCollector(
+                "session-1", "tenant-1",
+                new InformationalEventPost(ingress, clock), logger,
+                signalIngress: null, clock: clock,
+                startupGate: new StartupEventGate(tmp.Path, logger));
+
+            using var skip = new EspSkipConfigurationProbe.ScopedFullOverride(
+                _ => new EspFirstSyncSnapshot(
+                    skipUser: false, skipDevice: false,
+                    blockInStatusPage: null, syncFailureTimeoutMinutes: null));
+
+            // Early read (agent start / DeviceSetup detection): partial device list, empty user list.
+            using (new EspTrackingInfoProbe.ScopedOverride(_ => Snapshot(new[] { DeviceApp1 }, Array.Empty<string>())))
+            {
+                collector.CollectEspConfiguration();
+            }
+            var afterEarly = EspConfigEvents(ingress);
+            Assert.Single(afterEarly);
+            Assert.Empty(TypedList(afterEarly[0], "espTrackedUserWin32AppIds"));
+
+            // Apps sub-phase opened: the device list grew and sign-in populated the user list.
+            using (new EspTrackingInfoProbe.ScopedOverride(_ => Snapshot(new[] { DeviceApp1, DeviceApp2 }, new[] { UserApp1 })))
+            {
+                collector.RefreshEspConfiguration(DeviceInfoHost.EspConfigTriggerAppsUser);
+            }
+            var afterGrowth = EspConfigEvents(ingress);
+            Assert.Equal(2, afterGrowth.Count);
+            Assert.Contains(DeviceApp2, TypedList(afterGrowth[1], "espTrackedWin32AppIds"));
+            Assert.Equal(new[] { UserApp1 }, TypedList(afterGrowth[1], "espTrackedUserWin32AppIds"));
+
+            // A later trigger with an UNCHANGED registry state re-reads but emits nothing —
+            // the StartupEventGate suppresses the identical payload.
+            using (new EspTrackingInfoProbe.ScopedOverride(_ => Snapshot(new[] { DeviceApp1, DeviceApp2 }, new[] { UserApp1 })))
+            {
+                collector.RefreshEspConfiguration(DeviceInfoHost.EspConfigTriggerAccountSetup);
+            }
+            Assert.Equal(2, EspConfigEvents(ingress).Count);
+        }
+
+        [Fact]
+        public async Task Concurrent_refreshes_never_emit_an_older_snapshot_after_a_newer_one()
+        {
+            // Codex review: the refresh triggers each run on their own Task.Run — without the
+            // collector-side lock, two collects can interleave read-before-emit and the LATER
+            // emission carries the OLDER (smaller) list. The backend keeps the latest
+            // list-bearing emission as positive evidence, so that would silently shrink it.
+            // The probe returns a strictly growing list per read; with the read→emit critical
+            // section, emission order must equal read order — sizes strictly ascending.
+            using var tmp = new TempDirectory();
+            var logger = new AgentLogger(tmp.Path, AgentLogLevel.Info);
+            var ingress = new FakeSignalIngressSink();
+            var clock = new VirtualClock(T0);
+            var collector = new DeviceInfoCollector(
+                "session-1", "tenant-1",
+                new InformationalEventPost(ingress, clock), logger,
+                signalIngress: null, clock: clock,
+                startupGate: new StartupEventGate(tmp.Path, logger));
+
+            using var skip = new EspSkipConfigurationProbe.ScopedFullOverride(
+                _ => new EspFirstSyncSnapshot(
+                    skipUser: false, skipDevice: false,
+                    blockInStatusPage: null, syncFailureTimeoutMinutes: null));
+
+            var readCounter = 0;
+            using var tracking = new EspTrackingInfoProbe.ScopedOverride(_ =>
+            {
+                var n = Interlocked.Increment(ref readCounter);
+                var ids = Enumerable.Range(1, n)
+                    .Select(i => $"{i:d8}-1111-2222-3333-444444444444")
+                    .ToArray();
+                return Snapshot(ids, Array.Empty<string>());
+            });
+
+            var tasks = Enumerable.Range(0, 12)
+                .Select(i => Task.Run(() => collector.RefreshEspConfiguration($"stress-{i}")))
+                .ToArray();
+            await Task.WhenAll(tasks);
+
+            var sizes = EspConfigEvents(ingress)
+                .Select(e => TypedList(e, "espTrackedWin32AppIds").Count)
+                .ToList();
+            // Every refresh saw a strictly larger list (gate never suppresses) and each
+            // read→emit is atomic, so the emission sequence is exactly 1..12.
+            Assert.Equal(Enumerable.Range(1, 12), sizes);
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Monitoring/Interop/WifiInfoReaderTests.cs

@@ -0,0 +1,97 @@
+#nullable enable
+using System.Text;
+using AutopilotMonitor.Agent.V2.Core.Monitoring.Interop;
+using Xunit;
+using static AutopilotMonitor.Agent.V2.Core.Monitoring.Interop.WlanNativeMethods;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Interop
+{
+    /// <summary>
+    /// Tests for the native-WLAN-API WiFi reader. The WLAN state of the test machine is
+    /// unknown (CI runners/VMs have no WLAN service), so the live path only pins the
+    /// fail-soft contract; the pure helpers (SSID decode, PHY-type mapping) are pinned
+    /// exactly — their output feeds wifi_signal_info event payloads and must stay
+    /// consistent with the strings netsh historically reported.
+    /// </summary>
+    public sealed class WifiInfoReaderTests
+    {
+        [Fact]
+        public void TryGetCurrentConnection_never_throws_and_returns_sane_values()
+        {
+            var result = WifiInfoReader.TryGetCurrentConnection();
+
+            if (result != null)
+            {
+                Assert.InRange(result.SignalPercent, 0, 100);
+                if (result.Channel.HasValue)
+                    Assert.True(result.Channel.Value > 0, "channel must be positive when present");
+            }
+        }
+
+        // ----- DecodeSsid ---------------------------------------------------
+
+        [Fact]
+        public void DecodeSsid_decodes_utf8_bytes()
+        {
+            var raw = new byte[32];
+            var ssid = Encoding.UTF8.GetBytes("Contoso-Büro");
+            ssid.CopyTo(raw, 0);
+
+            Assert.Equal("Contoso-Büro", WifiInfoReader.DecodeSsid(raw, (uint)ssid.Length));
+        }
+
+        [Fact]
+        public void DecodeSsid_returns_null_for_empty_or_missing()
+        {
+            Assert.Null(WifiInfoReader.DecodeSsid(null, 5));
+            Assert.Null(WifiInfoReader.DecodeSsid(new byte[32], 0));
+        }
+
+        [Fact]
+        public void DecodeSsid_clamps_length_to_buffer_size()
+        {
+            var raw = Encoding.UTF8.GetBytes("abc");
+
+            // A corrupt native length beyond the buffer must not throw.
+            Assert.Equal("abc", WifiInfoReader.DecodeSsid(raw, 200));
+        }
+
+        [Fact]
+        public void DecodeSsid_uses_only_declared_length()
+        {
+            var raw = new byte[32];
+            Encoding.UTF8.GetBytes("HomeNet-Rest-Ignored").CopyTo(raw, 0);
+
+            Assert.Equal("HomeNet", WifiInfoReader.DecodeSsid(raw, 7));
+        }
+
+        // ----- PhyTypeToRadioType -------------------------------------------
+
+        // int parameters (cast inside): DOT11_PHY_TYPE is internal and may not appear in a
+        // public test method signature (CS0051).
+        [Theory]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_ofdm, "802.11a")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_hrdsss, "802.11b")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_erp, "802.11g")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_ht, "802.11n")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_vht, "802.11ac")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_dmg, "802.11ad")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_he, "802.11ax")]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_eht, "802.11be")]
+        public void PhyTypeToRadioType_maps_known_generations(int phyType, string expected)
+        {
+            Assert.Equal(expected, WifiInfoReader.PhyTypeToRadioType((DOT11_PHY_TYPE)phyType));
+        }
+
+        [Theory]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_unknown)]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_fhss)]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_dsss)]
+        [InlineData((int)DOT11_PHY_TYPE.dot11_phy_type_irbaseband)]
+        [InlineData(9999)]
+        public void PhyTypeToRadioType_returns_null_for_legacy_or_unknown(int phyType)
+        {
+            Assert.Null(WifiInfoReader.PhyTypeToRadioType((DOT11_PHY_TYPE)phyType));
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Monitoring/SystemSignals/EspPolicyProviderStallDetectorTests.cs

@@ -74,14 +74,17 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             var previous = new Dictionary<string, DateTime> { ["setupApps|device|ConfigMgr"] = T0 };
 
             var missing = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                EspPolicyProviderSnapshot.Empty, previous, new HashSet<string>(), T0.AddHours(2), Dwell);
+                EspPolicyProviderSnapshot.Empty, previous, new HashSet<string>(), T0, false, T0.AddHours(2), Dwell);
             Assert.Empty(missing.UpdatedFirstSeenUtc);
             Assert.Empty(missing.NewlyStalled);
+            Assert.Null(missing.UpdatedSidecarMissingSinceUtc); // arm 2 resets too
+            Assert.Null(missing.SidecarMissingStalledForMinutes);
 
             var registeredButEmpty = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                Snapshot(), previous, new HashSet<string>(), T0.AddHours(2), Dwell);
+                Snapshot(), previous, new HashSet<string>(), T0, false, T0.AddHours(2), Dwell);
             Assert.Empty(registeredButEmpty.UpdatedFirstSeenUtc);
             Assert.Empty(registeredButEmpty.NewlyStalled);
+            Assert.Null(registeredButEmpty.UpdatedSidecarMissingSinceUtc);
         }
 
         [Fact]
@@ -90,14 +93,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             var snapshot = Snapshot(SetupApps("ConfigMgr", tracking: null));
 
             var round1 = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                snapshot, new Dictionary<string, DateTime>(), new HashSet<string>(), T0, Dwell);
+                snapshot, new Dictionary<string, DateTime>(), new HashSet<string>(), null, false, T0, Dwell);
             Assert.Equal(T0, round1.UpdatedFirstSeenUtc["setupApps|device|ConfigMgr"]);
             Assert.Empty(round1.NewlyStalled);
 
             var round2 = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                snapshot, round1.UpdatedFirstSeenUtc, new HashSet<string>(), T0.AddMinutes(14), Dwell);
+                snapshot, round1.UpdatedFirstSeenUtc, new HashSet<string>(),
+                round1.UpdatedSidecarMissingSinceUtc, false, T0.AddMinutes(14), Dwell);
             Assert.Equal(T0, round2.UpdatedFirstSeenUtc["setupApps|device|ConfigMgr"]); // first-seen preserved
             Assert.Empty(round2.NewlyStalled);
+            Assert.Null(round2.SidecarMissingStalledForMinutes); // 14 min < dwell on arm 2 as well
         }
 
         [Fact]
@@ -107,7 +112,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             var seen = new Dictionary<string, DateTime> { ["setupApps|device|ConfigMgr"] = T0 };
 
             var eval = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                snapshot, seen, new HashSet<string>(), T0.AddMinutes(15), Dwell);
+                snapshot, seen, new HashSet<string>(), null, false, T0.AddMinutes(15), Dwell);
 
             var stalled = Assert.Single(eval.NewlyStalled);
             Assert.Equal("ConfigMgr", stalled.Provider.Name);
@@ -122,20 +127,20 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
 
             // Completes mid-dwell → dropped from the map.
             var completed = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                Snapshot(SetupApps("ConfigMgr", tracking: 1)), seen, new HashSet<string>(), T0.AddMinutes(10), Dwell);
+                Snapshot(SetupApps("ConfigMgr", tracking: 1)), seen, new HashSet<string>(), null, false, T0.AddMinutes(10), Dwell);
             Assert.Empty(completed.UpdatedFirstSeenUtc);
 
             // Re-appears incomplete much later → dwell restarts from the new now, no stall yet.
             var reappeared = EspPolicyProviderStallDetector.EvaluateStallTransition(
                 Snapshot(SetupApps("ConfigMgr", tracking: 0)), completed.UpdatedFirstSeenUtc,
-                new HashSet<string>(), T0.AddHours(3), Dwell);
+                new HashSet<string>(), null, false, T0.AddHours(3), Dwell);
             Assert.Equal(T0.AddHours(3), reappeared.UpdatedFirstSeenUtc[key]);
             Assert.Empty(reappeared.NewlyStalled);
 
             // Disappearance behaves the same as completion.
             var disappeared = EspPolicyProviderStallDetector.EvaluateStallTransition(
                 Snapshot(SetupApps("Sidecar", tracking: 1)), new Dictionary<string, DateTime> { [key] = T0 },
-                new HashSet<string>(), T0.AddMinutes(20), Dwell);
+                new HashSet<string>(), null, false, T0.AddMinutes(20), Dwell);
             Assert.False(disappeared.UpdatedFirstSeenUtc.ContainsKey(key));
         }
 
@@ -147,7 +152,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             var fired = new HashSet<string>(StringComparer.OrdinalIgnoreCase) { "setupApps|device|ConfigMgr" };
 
             var eval = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                snapshot, seen, fired, T0.AddHours(5), Dwell);
+                snapshot, seen, fired, null, false, T0.AddHours(5), Dwell);
 
             Assert.Empty(eval.NewlyStalled);
             Assert.True(eval.UpdatedFirstSeenUtc.ContainsKey("setupApps|device|ConfigMgr")); // still tracked
@@ -162,7 +167,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
                 DevicePrep("ConfigMgr", installState: 1));
 
             var eval = EspPolicyProviderStallDetector.EvaluateStallTransition(
-                snapshot, new Dictionary<string, DateTime>(), new HashSet<string>(), T0, Dwell);
+                snapshot, new Dictionary<string, DateTime>(), new HashSet<string>(), null, false, T0, Dwell);
 
             Assert.Equal(3, eval.UpdatedFirstSeenUtc.Count);
             Assert.Contains("setupApps|device|ConfigMgr", eval.UpdatedFirstSeenUtc.Keys);
@@ -179,6 +184,71 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             Assert.False(Snapshot(DevicePrep("Sidecar", installState: 3)).SidecarRegistered); // wrong kind
             Assert.False(EspPolicyProviderSnapshot.Empty.SidecarRegistered);
         }
+
+        // ------------------------------------------------- arm 2: sidecar-missing transition
+
+        [Fact]
+        public void Issue106_foreign_provider_complete_but_sidecar_missing_stalls_after_dwell()
+        {
+            // The actual field case: ConfigMgr with TrackingPoliciesCreated=1 — "complete" by the
+            // CSP value contract, arm 1 never fires — yet the ESP waits on the absent Sidecar.
+            var snapshot = Snapshot(SetupApps("ConfigMgr", tracking: 1));
+
+            var round1 = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                snapshot, new Dictionary<string, DateTime>(), new HashSet<string>(), null, false, T0, Dwell);
+            Assert.Empty(round1.NewlyStalled);                       // arm 1 stays silent...
+            Assert.Equal(T0, round1.UpdatedSidecarMissingSinceUtc);  // ...but arm 2 starts its clock
+            Assert.Null(round1.SidecarMissingStalledForMinutes);
+
+            var round2 = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                snapshot, round1.UpdatedFirstSeenUtc, new HashSet<string>(),
+                round1.UpdatedSidecarMissingSinceUtc, false, T0.AddMinutes(15), Dwell);
+            Assert.Empty(round2.NewlyStalled);
+            Assert.Equal(15.0, round2.SidecarMissingStalledForMinutes!.Value, precision: 3);
+        }
+
+        [Fact]
+        public void Sidecar_registering_resets_the_sidecar_missing_clock()
+        {
+            // ConfigMgr first (legitimate startup ordering) → clock runs...
+            var round1 = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                Snapshot(SetupApps("ConfigMgr", tracking: 1)),
+                new Dictionary<string, DateTime>(), new HashSet<string>(), null, false, T0, Dwell);
+            Assert.Equal(T0, round1.UpdatedSidecarMissingSinceUtc);
+
+            // ...Sidecar joins → condition clears, state resets.
+            var round2 = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                Snapshot(SetupApps("ConfigMgr", tracking: 1), SetupApps("Sidecar", tracking: 0)),
+                round1.UpdatedFirstSeenUtc, new HashSet<string>(),
+                round1.UpdatedSidecarMissingSinceUtc, false, T0.AddMinutes(10), Dwell);
+            Assert.Null(round2.UpdatedSidecarMissingSinceUtc);
+            Assert.Null(round2.SidecarMissingStalledForMinutes);
+
+            // Sidecar gone again much later → clock restarts from the new now, no instant stall.
+            var round3 = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                Snapshot(SetupApps("ConfigMgr", tracking: 1)),
+                round2.UpdatedFirstSeenUtc, new HashSet<string>(),
+                round2.UpdatedSidecarMissingSinceUtc, false, T0.AddHours(3), Dwell);
+            Assert.Equal(T0.AddHours(3), round3.UpdatedSidecarMissingSinceUtc);
+            Assert.Null(round3.SidecarMissingStalledForMinutes);
+        }
+
+        [Fact]
+        public void Sidecar_missing_one_shot_and_deviceprep_only_scoping()
+        {
+            // Already fired → never re-reported, but the clock keeps running.
+            var latched = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                Snapshot(SetupApps("ConfigMgr", tracking: 1)),
+                new Dictionary<string, DateTime>(), new HashSet<string>(), T0, true, T0.AddHours(5), Dwell);
+            Assert.Null(latched.SidecarMissingStalledForMinutes);
+            Assert.Equal(T0, latched.UpdatedSidecarMissingSinceUtc);
+
+            // DevicePreparation providers alone don't constitute the Setup/Apps wait — arm 2 off.
+            var devicePrepOnly = EspPolicyProviderStallDetector.EvaluateStallTransition(
+                Snapshot(DevicePrep("ConfigMgr", installState: 1)),
+                new Dictionary<string, DateTime>(), new HashSet<string>(), null, false, T0, Dwell);
+            Assert.Null(devicePrepOnly.UpdatedSidecarMissingSinceUtc);
+        }
     }
 
     /// <summary>
@@ -243,7 +313,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             Assert.Equal(EspPolicyProviderStallDetector.SourceLabel, evt.Payload[SignalPayloadKeys.Source]);
 
             var data = Assert.IsAssignableFrom<IReadOnlyDictionary<string, object>>(evt.TypedPayload);
+            // Both arms cross in the same round (ConfigMgr incomplete AND Sidecar absent) →
+            // ONE event; sidecar-missing is the dominant reason.
+            Assert.Equal(EspPolicyProviderStallDetector.ReasonSidecarMissing, data["reason"]);
             Assert.Equal(false, data["sidecarRegistered"]);
+            Assert.Equal(15.0, data["sidecarMissingForMinutes"]);
             Assert.Equal(EspPolicyProviderStallDetector.DwellMinutes, data["dwellMinutes"]);
             var stalled = Assert.IsAssignableFrom<IReadOnlyList<Dictionary<string, object>>>(data["stalledProviders"]);
             var entry = Assert.Single(stalled);
@@ -252,6 +326,36 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
             Assert.Equal(15.0, entry["stalledForMinutes"]);
         }
 
+        [Fact]
+        public void Issue106_configmgr_complete_without_sidecar_still_emits_sidecar_missing()
+        {
+            // Regression test for the customer's follow-up on issue #106: TrackingPoliciesCreated=1
+            // WAS present under ConfigMgr — the provider looks "complete", arm 1 never fires, yet
+            // the ESP hangs because Sidecar is absent. Arm 2 must catch exactly this.
+            var configMgrComplete = new EspPolicyProviderSnapshot(new[]
+            {
+                new PolicyProviderState("ConfigMgr", "device", EspPolicyProviderProbe.KindSetupApps,
+                    trackingPoliciesCreated: 1, installationState: null),
+            });
+            using var _probe = new EspPolicyProviderProbe.ScopedOverride(_ => configMgrComplete);
+            var (sut, sink, clock) = BuildDetector();
+
+            sut.Tick();
+            Assert.Empty(StalledEvents(sink));
+
+            clock.Advance(TimeSpan.FromMinutes(EspPolicyProviderStallDetector.DwellMinutes));
+            sut.Tick();
+
+            var evt = Assert.Single(StalledEvents(sink));
+            var data = Assert.IsAssignableFrom<IReadOnlyDictionary<string, object>>(evt.TypedPayload);
+            Assert.Equal(EspPolicyProviderStallDetector.ReasonSidecarMissing, data["reason"]);
+            Assert.Equal(false, data["sidecarRegistered"]);
+            Assert.Equal(15.0, data["sidecarMissingForMinutes"]);
+            Assert.Empty(Assert.IsAssignableFrom<IReadOnlyList<Dictionary<string, object>>>(data["stalledProviders"]));
+            var providers = Assert.IsAssignableFrom<IReadOnlyList<Dictionary<string, object?>>>(data["providers"]);
+            Assert.Equal(true, Assert.Single(providers)["complete"]); // the misleading "complete" is visible
+        }
+
         [Fact]
         public void Sidecar_complete_configmgr_incomplete_reports_only_configmgr_with_sidecar_flag()
         {
@@ -264,7 +368,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.SystemSignals
 
             var evt = Assert.Single(StalledEvents(sink));
             var data = Assert.IsAssignableFrom<IReadOnlyDictionary<string, object>>(evt.TypedPayload);
+            Assert.Equal(EspPolicyProviderStallDetector.ReasonProviderIncomplete, data["reason"]); // arm 2 off — Sidecar is there
             Assert.Equal(true, data["sidecarRegistered"]);
+            Assert.False(data.ContainsKey("sidecarMissingForMinutes"));
             var stalled = Assert.IsAssignableFrom<IReadOnlyList<Dictionary<string, object>>>(data["stalledProviders"]);
             Assert.Equal("ConfigMgr", Assert.Single(stalled)["name"]);
             var providers = Assert.IsAssignableFrom<IReadOnlyList<Dictionary<string, object?>>>(data["providers"]);

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Monitoring/UefiSecureBootCertReaderTests.cs

@@ -161,6 +161,40 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring
             Assert.False(string.IsNullOrEmpty(snapshot.KekStatus));
         }
 
+        /// <summary>
+        /// SeSystemEnvironmentPrivilege is adjusted on the PROCESS token, so the enable →
+        /// read → restore sequence is a process-wide side effect. Unserialized, a second
+        /// reader observes "already enabled", skips its own restore, and then loses the
+        /// privilege mid-read when the first reader disposes — surfacing as a privilege_denied
+        /// that disagrees with the other readers. The machine's actual privilege state does
+        /// not matter here: whatever it is, every reader must report the SAME status.
+        /// </summary>
+        [Fact]
+        public void Read_is_serialized_across_concurrent_callers()
+        {
+            const int readers = 8;
+            var snapshots = new UefiSecureBootCertSnapshot[readers];
+            var failures = new System.Exception[readers];
+
+            var threads = new System.Threading.Thread[readers];
+            for (var i = 0; i < readers; i++)
+            {
+                var index = i;
+                threads[i] = new System.Threading.Thread(() =>
+                {
+                    try { snapshots[index] = UefiSecureBootCertReader.Read(); }
+                    catch (System.Exception ex) { failures[index] = ex; }
+                });
+                threads[i].Start();
+            }
+            foreach (var thread in threads) thread.Join();
+
+            Assert.All(failures, Assert.Null);
+            Assert.All(snapshots, s => Assert.False(string.IsNullOrEmpty(s.DbStatus)));
+            Assert.All(snapshots, s => Assert.Equal(snapshots[0].DbStatus, s.DbStatus));
+            Assert.All(snapshots, s => Assert.Equal(snapshots[0].KekStatus, s.KekStatus));
+        }
+
         private static byte[] Concat(params byte[][] parts)
         {
             var total = 0;

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Orchestration/DeviceInfoHostTriggerTests.cs

@@ -37,5 +37,41 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Orchestration
             Assert.False(DeviceInfoHost.IsEndTrigger(DecisionSignalKind.EspPhaseChanged, null));
             Assert.False(DeviceInfoHost.IsEndTrigger(DecisionSignalKind.InformationalEvent, Phase("FinalizingSetup")));
         }
+
+        // Audit Q2 — the payload shape the adapter's apps sub-phase declaration produces on the
+        // ingress rail (InformationalEventPost: eventType + declared phase enum name).
+        private static IReadOnlyDictionary<string, string> SubPhaseDeclaration(string declaredPhase) =>
+            new Dictionary<string, string>
+            {
+                { SignalPayloadKeys.EventType, AutopilotMonitor.Shared.Constants.EventTypes.PhaseTransition },
+                { AutopilotMonitor.Agent.V2.Core.Telemetry.Events.EventTimelineEmitter.PhaseParamKey, declaredPhase },
+            };
+
+        [Fact]
+        public void Apps_subphase_declarations_classify_as_esp_config_refresh_triggers()
+        {
+            Assert.Equal(DeviceInfoHost.EspConfigTriggerAppsDevice,
+                DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.InformationalEvent, SubPhaseDeclaration("AppsDevice")));
+            Assert.Equal(DeviceInfoHost.EspConfigTriggerAppsUser,
+                DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.InformationalEvent, SubPhaseDeclaration("AppsUser")));
+            // Non-apps phase declarations (e.g. the reducer's Finalizing entry) never classify.
+            Assert.Null(DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.InformationalEvent, SubPhaseDeclaration("FinalizingSetup")));
+            // A phase_transition without a declared phase never classifies.
+            Assert.Null(DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.InformationalEvent,
+                new Dictionary<string, string> { { SignalPayloadKeys.EventType, AutopilotMonitor.Shared.Constants.EventTypes.PhaseTransition } }));
+        }
+
+        [Fact]
+        public void AccountSetup_phase_classifies_as_the_user_setup_refresh_trigger()
+        {
+            Assert.Equal(DeviceInfoHost.EspConfigTriggerAccountSetup,
+                DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.EspPhaseChanged, Phase("AccountSetup")));
+            // DeviceSetup is the enrollment-start trigger, not an ESP-config refresh.
+            Assert.Null(DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.EspPhaseChanged, Phase("DeviceSetup")));
+            Assert.Null(DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.EspPhaseChanged, null));
+            // Unrelated informational events never classify.
+            Assert.Null(DeviceInfoHost.ClassifyEspConfigRefreshTrigger(DecisionSignalKind.InformationalEvent,
+                new Dictionary<string, string> { { SignalPayloadKeys.EventType, "esp_config_detected" } }));
+        }
     }
 }

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Security/HardwareInfoNormalizeTests.cs

@@ -0,0 +1,33 @@
+using AutopilotMonitor.Agent.V2.Core.Security;
+using Xunit;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Security
+{
+    /// <summary>
+    /// Audit Q5 — WMI hardware values are trimmed at the source. BIOS serials become device
+    /// identity keys downstream (F2 device history joins on TenantId + serial), so padding
+    /// must never leave the agent; whitespace-only values collapse to the "Unknown" sentinel
+    /// the retry logic already treats as unresolved.
+    /// </summary>
+    public sealed class HardwareInfoNormalizeTests
+    {
+        [Theory]
+        [InlineData(" PF60H941 ", "PF60H941")]
+        [InlineData("PF60H941", "PF60H941")]
+        [InlineData("\tLENOVO \r\n", "LENOVO")]
+        [InlineData("Unknown", "Unknown")]
+        public void Trims_whitespace_at_the_source(string raw, string expected)
+            => Assert.Equal(expected, HardwareInfo.NormalizeWmiValue(raw));
+
+        [Theory]
+        [InlineData("")]
+        [InlineData("   ")]
+        [InlineData("\t\r\n")]
+        public void Empty_and_whitespace_only_collapse_to_the_Unknown_sentinel(string raw)
+            => Assert.Equal("Unknown", HardwareInfo.NormalizeWmiValue(raw));
+
+        [Fact]
+        public void Null_collapses_to_the_Unknown_sentinel()
+            => Assert.Equal("Unknown", HardwareInfo.NormalizeWmiValue(null));
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/SignalAdapters/ImeLogTrackerAdapterTimingTests.cs

@@ -243,5 +243,87 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.SignalAdapters
 
             Assert.Equal(firstStamp, adapter.AppTimings["app-P"].StartedAtUtc);
         }
+
+        [Fact]
+        public void Terminal_to_active_retry_rearms_CompletedAt_and_final_terminal_covers_total_occupancy()
+        {
+            // Audit Q1: the IME retry path (Error -> Downloading -> Installed is permitted by
+            // AppPackageState.UpdateState). Pre-fix, CompletedAt froze at the first Error, so
+            // the final Installed payload carried the first failure's completedAt/duration.
+            using var f = new ImeLogTrackerAdapterFixture(T0);
+            using var adapter = new ImeLogTrackerAdapter(f.Tracker, f.Ingress, f.Clock);
+            var app = new AppPackageState("app-R", 0);
+
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Unknown, AppInstallationState.Installing);
+            f.Clock.Advance(TimeSpan.FromSeconds(5));
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Installing, AppInstallationState.Error);
+            f.Clock.Advance(TimeSpan.FromSeconds(10));
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Error, AppInstallationState.Downloading);
+
+            // Re-armed while active again: the retry's download event carries no completedAt,
+            // and the FIRST start stamp is kept.
+            Assert.Null(adapter.AppTimings["app-R"].CompletedAtUtc);
+            var retryDownload = f.InfoEvent(SharedEventTypes.AppDownloadStarted);
+            Assert.False(retryDownload.Payload!.ContainsKey("completedAt"));
+            Assert.False(retryDownload.Payload.ContainsKey("durationSeconds"));
+            Assert.Equal(T0.ToString("o"), retryDownload.Payload["startedAt"]);
+
+            f.Clock.Advance(TimeSpan.FromSeconds(20));
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Downloading, AppInstallationState.Installed);
+
+            var timing = adapter.AppTimings["app-R"];
+            Assert.Equal(T0, timing.StartedAtUtc);
+            Assert.Equal(T0.AddSeconds(35), timing.CompletedAtUtc);
+            Assert.Equal(35.0, timing.DurationSeconds);
+
+            var completed = Assert.Single(f.InfoEvents(SharedEventTypes.AppInstallComplete));
+            Assert.Equal("35.00", completed.Payload!["durationSeconds"]);
+            Assert.Equal(T0.AddSeconds(35).ToString("o"), completed.Payload["completedAt"]);
+        }
+
+        [Fact]
+        public void Skipped_then_retried_app_emits_positive_duration_from_retry_start()
+        {
+            // Production regression (session 66f1c5e1, audit Q1): a user-scope app evaluated
+            // Skipped during DeviceSetup (terminal, no start observed), retried after sign-in
+            // (Skipped -> Downloading is permitted), Installed later. Pre-fix payload:
+            // completedAt frozen at the early Skip + startedAt from the retry ->
+            // durationSeconds=-2342. Post-fix: retry re-arms, duration covers retry-start ->
+            // final terminal.
+            using var f = new ImeLogTrackerAdapterFixture(T0);
+            using var adapter = new ImeLogTrackerAdapter(f.Tracker, f.Ingress, f.Clock);
+            var app = new AppPackageState("app-S", 0);
+
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Unknown, AppInstallationState.Skipped);
+            Assert.Null(adapter.AppTimings["app-S"].StartedAtUtc);
+            Assert.Equal(T0, adapter.AppTimings["app-S"].CompletedAtUtc);
+
+            f.Clock.Advance(TimeSpan.FromSeconds(60));
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Skipped, AppInstallationState.Downloading);
+            f.Clock.Advance(TimeSpan.FromSeconds(30));
+            adapter.TriggerAppStateFromTest(app, AppInstallationState.Downloading, AppInstallationState.Installed);
+
+            var timing = adapter.AppTimings["app-S"];
+            Assert.Equal(T0.AddSeconds(60), timing.StartedAtUtc);
+            Assert.Equal(T0.AddSeconds(90), timing.CompletedAtUtc);
+            Assert.Equal(30.0, timing.DurationSeconds);
+
+            var completedEvents = f.InfoEvents(SharedEventTypes.AppInstallComplete);
+            Assert.Equal(2, completedEvents.Count);
+            Assert.Equal("30.00", completedEvents[1].Payload!["durationSeconds"]);
+        }
+
+        [Fact]
+        public void Inverted_endpoints_yield_null_DurationSeconds()
+        {
+            // Negative guard (audit Q1): endpoints stamped from out-of-order source lines must
+            // never surface as a negative durationSeconds (observed -2342s in production). The
+            // payload emitter omits the field when DurationSeconds is null.
+            var inverted = new AppInstallTiming(startedAtUtc: T0.AddSeconds(100), completedAtUtc: T0);
+            Assert.Null(inverted.DurationSeconds);
+
+            var wellOrdered = new AppInstallTiming(startedAtUtc: T0, completedAtUtc: T0.AddSeconds(100));
+            Assert.Equal(100.0, wellOrdered.DurationSeconds);
+        }
     }
 }

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Termination/ObservationCoverageTests.cs

@@ -1,5 +1,7 @@
 using System;
+using AutopilotMonitor.Agent.V2.Core.Orchestration;
 using AutopilotMonitor.Agent.V2.Core.Termination;
+using AutopilotMonitor.Shared.Models;
 using Xunit;
 
 namespace AutopilotMonitor.Agent.V2.Core.Tests.Termination
@@ -70,5 +72,82 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Termination
             Assert.Equal(0, bootToStart, 0);
             Assert.Equal(10, uptime, 0);
         }
+
+        // --- DescribeLateStart: outcome-calibrated severity + phrasing (fleet 2026-07-28) ---
+
+        private static void Describe(
+            EnrollmentTerminationOutcome outcome, string oobeState,
+            out EventSeverity severity, out string message, out string note)
+            => ObservationCoverage.DescribeLateStart(
+                outcome, oobeState, bootToStartSeconds: 34 * 60, uptimeSeconds: 30,
+                out severity, out message, out note);
+
+        [Fact]
+        public void Failed_outcome_is_a_warning_with_the_hung_script_post_mortem_note()
+        {
+            Describe(EnrollmentTerminationOutcome.Failed, "in_progress",
+                out var severity, out var message, out var note);
+
+            Assert.Equal(EventSeverity.Warning, severity);
+            Assert.Contains("low coverage of the enrollment window", message);
+            Assert.Contains("post-mortem", note);
+            Assert.Contains("hung ahead of the bootstrap", note);
+        }
+
+        [Fact]
+        public void TimedOut_outcome_is_a_warning_like_a_failure()
+        {
+            Describe(EnrollmentTerminationOutcome.TimedOut, "in_progress",
+                out var severity, out _, out var note);
+
+            Assert.Equal(EventSeverity.Warning, severity);
+            Assert.Contains("post-mortem", note);
+        }
+
+        [Fact]
+        public void Succeeded_outcome_is_an_info_about_tail_only_coverage_without_alarm_framing()
+        {
+            // The 659c3a90 shape: healthy enrollment, bootstrap merely ran late in the ESP queue.
+            Describe(EnrollmentTerminationOutcome.Succeeded, "in_progress",
+                out var severity, out _, out var note);
+
+            Assert.Equal(EventSeverity.Info, severity);
+            Assert.Contains("succeeded", note);
+            Assert.Contains("IME log replay", note);
+            Assert.DoesNotContain("post-mortem", note);
+            Assert.DoesNotContain("hung", note);
+        }
+
+        [Fact]
+        public void Oobe_completed_at_start_phrases_a_pure_post_mortem()
+        {
+            Describe(EnrollmentTerminationOutcome.Failed, "completed",
+                out _, out var message, out _);
+
+            Assert.Contains("OOBE was already completed when the agent started", message);
+        }
+
+        [Fact]
+        public void Oobe_in_progress_at_start_phrases_the_live_tail()
+        {
+            Describe(EnrollmentTerminationOutcome.Failed, "in_progress",
+                out _, out var message, out _);
+
+            Assert.Contains("OOBE was still in progress at agent start", message);
+            Assert.Contains("final 30s live", message);
+        }
+
+        [Theory]
+        [InlineData("not_started")]  // flips in after a mid-OOBE reboot — not interpretable
+        [InlineData("unavailable")]  // WinRT contract absent / read failed
+        [InlineData("unknown_3")]    // unexpected enum value
+        public void Uninterpretable_oobe_states_add_no_claim_to_the_message(string oobeState)
+        {
+            Describe(EnrollmentTerminationOutcome.Failed, oobeState,
+                out _, out var message, out _);
+
+            Assert.DoesNotContain("OOBE", message);
+            Assert.EndsWith("low coverage of the enrollment window.", message);
+        }
     }
 }

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/Ime/AppInstallTiming.cs

@@ -10,9 +10,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
     /// the <c>app_tracking_summary</c> terminal event.
     /// <para>
     /// <b>StartedAtUtc</b> is recorded on the first lifecycle transition into Downloading /
-    /// Installing / InProgress. <b>CompletedAtUtc</b> is recorded on the first transition to a
-    /// terminal state (Installed / Skipped / Postponed / Error). Both are "set-once" — subsequent
-    /// identical-state events do not overwrite an earlier stamp.
+    /// Installing / InProgress and never overwritten. <b>CompletedAtUtc</b> is recorded on a
+    /// transition to a terminal state (Installed / Skipped / Postponed / Error) and re-armed
+    /// to <c>null</c> when the app goes terminal → active again (IME retry / re-evaluation,
+    /// audit Q1) — so the final terminal event carries first-start → last-terminal.
     /// </para>
     /// </summary>
     public sealed class AppInstallTiming
@@ -30,10 +31,14 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
         /// <summary>
         /// Elapsed time from first Downloading/Installing transition to terminal state, in
         /// seconds. <c>null</c> when either endpoint is missing (e.g. app still installing,
-        /// or the adapter missed the start event because it came online mid-install).
+        /// or the adapter missed the start event because it came online mid-install) — and
+        /// <c>null</c> when the endpoints are inverted (CompletedAt before StartedAt: log lines
+        /// from different source files can be processed out of chronological order; observed in
+        /// production as durationSeconds=-2342 on a Skipped-then-retried user-scope app,
+        /// audit Q1). A negative duration is never a real measurement, so it is not emitted.
         /// </summary>
         public double? DurationSeconds =>
-            StartedAtUtc.HasValue && CompletedAtUtc.HasValue
+            StartedAtUtc.HasValue && CompletedAtUtc.HasValue && CompletedAtUtc.Value >= StartedAtUtc.Value
                 ? (CompletedAtUtc.Value - StartedAtUtc.Value).TotalSeconds
                 : (double?)null;
     }

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/SystemSignals/EspPolicyProviderProbe.cs

@@ -35,13 +35,19 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
     /// (registered during Autopilot-into-co-management, see the troubleshooting registry paths in
     /// https://learn.microsoft.com/en-us/intune/configmgr/comanage/autopilot-enrollment). Microsoft
     /// warns that ESP policy providers are not aware of each other and that pre-provisioning is
-    /// unsupported with co-management. Field case that motivated this probe: a pre-provisioned
-    /// device with the ConfigMgr client installed as a blocking app ended up with ONLY
-    /// <c>ConfigMgr</c> registered under <c>Device\Setup\Apps\PolicyProviders</c> (no
-    /// <c>Sidecar</c>) and <c>TrackingPoliciesCreated</c> never set — the user ESP hung at
-    /// "Apps (Identifying)" for hours to days; manually renaming the key to <c>Sidecar</c>
-    /// unblocked it immediately. <see cref="EspPolicyProviderStallDetector"/> turns this contract
-    /// into the <c>esp_policy_provider_stalled</c> tripwire.
+    /// unsupported with co-management. Field case that motivated this probe (issue #106): a
+    /// pre-provisioned device with the ConfigMgr client installed as a blocking app ended up with
+    /// ONLY <c>ConfigMgr</c> registered under <c>Device\Setup\Apps\PolicyProviders</c> (no
+    /// <c>Sidecar</c>) — and although <c>TrackingPoliciesCreated=1</c> WAS set under it, the user
+    /// ESP hung at "Apps (Identifying)" for hours to days; manually renaming the key to
+    /// <c>Sidecar</c> unblocked it immediately. So the ESP's Apps wait is keyed to the provider
+    /// the Intune service registered at enrollment (always <c>Sidecar</c>; ConfigMgr only joins
+    /// via a co-management settings policy — see the WAP payload documented in
+    /// https://oofhours.com/2023/08/25/run-an-sccm-task-sequence-during-autopilot/), and a
+    /// foreign provider's completion values do not satisfy it. Detection therefore keys on the
+    /// provider NAME topology, not only on completeness:
+    /// <see cref="EspPolicyProviderStallDetector"/> turns this contract into the two-arm
+    /// <c>esp_policy_provider_stalled</c> tripwire.
     /// </para>
     /// <para>
     /// Fail-soft like <see cref="EspTrackingInfoProbe"/>: any failure is Debug-logged and yields

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/SystemSignals/EspPolicyProviderStallDetector.cs

@@ -13,18 +13,30 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
 {
     /// <summary>
     /// Turns the ESP policy-provider contract read by <see cref="EspPolicyProviderProbe"/> into
-    /// the one-shot <c>esp_policy_provider_stalled</c> Warning: a registered provider that stays
-    /// continuously incomplete for <see cref="DwellMinutes"/> minutes is reported once, with the
-    /// full provider table in the payload. See the probe's header for the CSP contract, the
-    /// official sources, and the co-management field case (ConfigMgr registered without Sidecar
-    /// → user ESP parked at "Apps (Identifying)" with no OS timeout) this tripwire exists for.
+    /// the one-shot <c>esp_policy_provider_stalled</c> Warning via two independent arms, each
+    /// with the same <see cref="DwellMinutes"/>-minute continuous-dwell rule and the full
+    /// provider table in the payload:
+    /// <list type="bullet">
+    ///   <item><b>Arm 1 — <c>reason=provider_incomplete</c></b>: a registered provider stays
+    ///     continuously incomplete (e.g. Sidecar itself never sets TrackingPoliciesCreated).</item>
+    ///   <item><b>Arm 2 — <c>reason=sidecar_provider_missing</c></b>: at least one
+    ///     <c>Setup\Apps</c> provider is registered but none of them is <c>Sidecar</c> (the IME).
+    ///     This is the actual co-management field-case signature (issue #106): the foreign
+    ///     <c>ConfigMgr</c> provider had <c>TrackingPoliciesCreated=1</c> — i.e. was "complete"
+    ///     by the CSP value contract — yet the user ESP stayed parked at "Apps (Identifying)"
+    ///     until the key was renamed to <c>Sidecar</c>. The ESP's Apps wait is keyed to the
+    ///     Intune-registered provider by NAME, so provider completeness alone must not gate this
+    ///     detection. See the probe's header for sources.</item>
+    /// </list>
     /// <para>
-    /// Dwell semantics: the dwell clock per provider starts when it is first observed incomplete
-    /// and RESETS when the provider completes or disappears — only <i>continuously</i> incomplete
-    /// providers fire. A missing root key / empty provider list is normal early enrollment and
-    /// clears all dwell state (probe D4). Purely observational: emission goes through
-    /// <see cref="InformationalEventPost"/> (InformationalEvent pass-through, dispatch-guard
-    /// exempt), never into decision state.
+    /// Dwell semantics: arm 1's clock per provider starts when it is first observed incomplete
+    /// and RESETS when the provider completes or disappears; arm 2's clock starts when the
+    /// sidecar-missing condition is first observed and RESETS when Sidecar registers (covers the
+    /// legitimate "ConfigMgr registered before Sidecar" startup ordering) — only
+    /// <i>continuously</i> bad states fire. A missing root key / empty provider list is normal
+    /// early enrollment and clears all dwell state (probe D4). Purely observational: emission
+    /// goes through <see cref="InformationalEventPost"/> (InformationalEvent pass-through,
+    /// dispatch-guard exempt), never into decision state.
     /// </para>
     /// <para>
     /// Duplicate suppression: in-process one-shot per provider key, plus a
@@ -44,6 +56,18 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
 
         internal const string SourceLabel = "EspPolicyProviderStallDetector";
 
+        /// <summary>Payload <c>reason</c>: a registered provider stayed continuously incomplete (arm 1).</summary>
+        internal const string ReasonProviderIncomplete = "provider_incomplete";
+
+        /// <summary>Payload <c>reason</c>: Setup\Apps providers registered but Sidecar absent (arm 2 — dominant when both arms fire in one round).</summary>
+        internal const string ReasonSidecarMissing = "sidecar_provider_missing";
+
+        /// <summary>
+        /// Sentinel token representing the arm-2 condition in the cross-restart fingerprint.
+        /// Cannot collide with provider keys — those always contain <c>|</c> separators.
+        /// </summary>
+        private const string SidecarMissingFingerprintToken = "sidecarMissing";
+
         private readonly string _sessionId;
         private readonly string _tenantId;
         private readonly InformationalEventPost _post;
@@ -57,6 +81,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
         private readonly HashSet<string> _firedKeys =
             new HashSet<string>(StringComparer.OrdinalIgnoreCase);
 
+        /// <summary>Arm-2 dwell state: first observation of "setupApps providers without Sidecar"; null = condition not active.</summary>
+        private DateTime? _sidecarMissingSinceUtc;
+
+        /// <summary>Arm-2 in-process one-shot latch (mirror of <see cref="_firedKeys"/>).</summary>
+        private bool _sidecarMissingFired;
+
         public EspPolicyProviderStallDetector(
             string sessionId,
             string tenantId,
@@ -79,13 +109,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
         internal void Tick()
         {
             var snapshot = EspPolicyProviderProbe.Read(_logger);
-            var eval = EvaluateStallTransition(snapshot, _firstSeenIncompleteUtc, _firedKeys, _clock.UtcNow, _dwell);
+            var eval = EvaluateStallTransition(
+                snapshot, _firstSeenIncompleteUtc, _firedKeys,
+                _sidecarMissingSinceUtc, _sidecarMissingFired, _clock.UtcNow, _dwell);
 
             _firstSeenIncompleteUtc.Clear();
             foreach (var kv in eval.UpdatedFirstSeenUtc)
                 _firstSeenIncompleteUtc[kv.Key] = kv.Value;
+            _sidecarMissingSinceUtc = eval.UpdatedSidecarMissingSinceUtc;
 
-            if (eval.NewlyStalled.Count > 0)
+            if (eval.NewlyStalled.Count > 0 || eval.SidecarMissingStalledForMinutes != null)
                 EmitStalled(snapshot, eval);
         }
 
@@ -94,24 +127,29 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
         /// registry, timer, or ingress (pattern: <c>PerformanceCollector.EvaluateDiskLowTransition</c>).
         /// </summary>
         /// <param name="snapshot">Current provider snapshot.</param>
-        /// <param name="firstSeenIncompleteUtc">Dwell state from the previous round (provider key → first-seen-incomplete).</param>
+        /// <param name="firstSeenIncompleteUtc">Arm-1 dwell state from the previous round (provider key → first-seen-incomplete).</param>
         /// <param name="alreadyFired">Provider keys already reported this run (never re-report).</param>
+        /// <param name="sidecarMissingSinceUtc">Arm-2 dwell state from the previous round; null = condition was not active.</param>
+        /// <param name="sidecarMissingAlreadyFired">Arm-2 one-shot latch (never re-report).</param>
         /// <param name="nowUtc">Evaluation timestamp.</param>
-        /// <param name="dwell">Continuous-incompleteness threshold.</param>
+        /// <param name="dwell">Continuous-condition threshold (shared by both arms).</param>
         internal static StallEvaluation EvaluateStallTransition(
             EspPolicyProviderSnapshot snapshot,
             IReadOnlyDictionary<string, DateTime> firstSeenIncompleteUtc,
             ISet<string> alreadyFired,
+            DateTime? sidecarMissingSinceUtc,
+            bool sidecarMissingAlreadyFired,
             DateTime nowUtc,
             TimeSpan dwell)
         {
             var result = new StallEvaluation();
 
             // Missing root key or no registered providers = normal early enrollment: clear all
-            // dwell state (a provider must be continuously OBSERVED incomplete to accrue dwell).
+            // dwell state (a condition must be continuously OBSERVED to accrue dwell).
             if (!snapshot.HasData || snapshot.Providers.Count == 0)
                 return result;
 
+            // Arm 1 — a registered provider continuously incomplete.
             foreach (var provider in snapshot.Providers)
             {
                 if (provider.IsComplete)
@@ -127,22 +165,44 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
                     result.NewlyStalled.Add(new StalledProvider(provider, stalledFor.TotalMinutes));
             }
 
+            // Arm 2 — Setup\Apps providers registered, but none of them is Sidecar. Deliberately
+            // ignores provider completeness: the field case (issue #106) had ConfigMgr with
+            // TrackingPoliciesCreated=1 and the ESP still waited. Sidecar registering later
+            // (legitimate startup ordering) clears the condition and resets the clock.
+            if (HasSetupAppsProvider(snapshot) && !snapshot.SidecarRegistered)
+            {
+                var since = sidecarMissingSinceUtc ?? nowUtc;
+                result.UpdatedSidecarMissingSinceUtc = since;
+
+                var missingFor = nowUtc - since;
+                if (missingFor >= dwell && !sidecarMissingAlreadyFired)
+                    result.SidecarMissingStalledForMinutes = missingFor.TotalMinutes;
+            }
+
             return result;
         }
 
+        private static bool HasSetupAppsProvider(EspPolicyProviderSnapshot snapshot)
+            => snapshot.Providers.Any(p => p.Kind == EspPolicyProviderProbe.KindSetupApps);
+
         private void EmitStalled(EspPolicyProviderSnapshot snapshot, StallEvaluation eval)
         {
             foreach (var stalled in eval.NewlyStalled)
                 _firedKeys.Add(stalled.Provider.Key);
+            if (eval.SidecarMissingStalledForMinutes != null)
+                _sidecarMissingFired = true;
 
-            // Cross-restart dedup: fingerprint = sorted set of all fired keys that are still
-            // incomplete. A restart into the same stall re-accrues dwell and lands on the same
-            // fingerprint → suppressed; a new provider joining the stall changes it → re-emitted.
+            // Cross-restart dedup: fingerprint = sorted set of all fired conditions that are
+            // still active (incomplete providers + the sidecar-missing sentinel). A restart into
+            // the same stall re-accrues dwell and lands on the same fingerprint → suppressed; a
+            // new condition joining the stall changes it → re-emitted.
             var currentlyStalledKeys = snapshot.Providers
                 .Where(p => !p.IsComplete && _firedKeys.Contains(p.Key))
                 .Select(p => p.Key)
-                .OrderBy(k => k, StringComparer.OrdinalIgnoreCase)
                 .ToList();
+            if (_sidecarMissingFired && HasSetupAppsProvider(snapshot) && !snapshot.SidecarRegistered)
+                currentlyStalledKeys.Add(SidecarMissingFingerprintToken);
+            currentlyStalledKeys.Sort(StringComparer.OrdinalIgnoreCase);
             var fingerprint = StartupEventGate.HashString(string.Join(";", currentlyStalledKeys));
             if (_startupGate != null &&
                 !_startupGate.ShouldEmit(Constants.EventTypes.EspPolicyProviderStalled, fingerprint))
@@ -153,11 +213,30 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
                 return;
             }
 
+            // Sidecar-missing is the dominant reason when both arms fire in one round: it
+            // explains the ESP hang regardless of the foreign providers' completeness.
+            var reason = eval.SidecarMissingStalledForMinutes != null
+                ? ReasonSidecarMissing
+                : ReasonProviderIncomplete;
+
+            var messageParts = new List<string>();
+            if (eval.SidecarMissingStalledForMinutes != null)
+            {
+                var registeredNames = string.Join(", ", snapshot.Providers
+                    .Where(p => p.Kind == EspPolicyProviderProbe.KindSetupApps)
+                    .Select(p => $"{p.Name} ({p.Scope})"));
+                messageParts.Add(
+                    $"expected IME provider '{EspPolicyProviderProbe.SidecarProviderName}' not registered for " +
+                    $">= {DwellMinutes} min while Setup/Apps providers exist: {registeredNames}");
+            }
             var stalledNames = string.Join(", ",
                 eval.NewlyStalled.Select(s => $"{s.Provider.Name} ({s.Provider.Scope}/{s.Provider.Kind})"));
+            if (eval.NewlyStalled.Count > 0)
+                messageParts.Add($"provider(s) incomplete for >= {DwellMinutes} min: {stalledNames}");
 
             var data = new Dictionary<string, object>(StringComparer.Ordinal)
             {
+                ["reason"] = reason,
                 ["dwellMinutes"] = DwellMinutes,
                 ["sidecarRegistered"] = snapshot.SidecarRegistered,
                 ["providers"] = snapshot.Providers.Select(p => new Dictionary<string, object?>(StringComparer.Ordinal)
@@ -177,6 +256,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
                     ["stalledForMinutes"] = Math.Round(s.StalledForMinutes, 1),
                 }).ToList(),
             };
+            if (eval.SidecarMissingStalledForMinutes != null)
+                data["sidecarMissingForMinutes"] = Math.Round(eval.SidecarMissingStalledForMinutes.Value, 1);
+
+            var message =
+                "ESP Setup/Apps tracking stalled: " + string.Join("; ", messageParts) +
+                " — the ESP waits on its app-tracking providers without any timeout";
 
             _post.Emit(new EnrollmentEvent
             {
@@ -188,27 +273,31 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.SystemSignals
                 Source = SourceLabel,
                 Phase = EnrollmentPhase.Unknown,
                 ImmediateUpload = true,
-                Message = $"ESP policy provider(s) incomplete for >= {DwellMinutes} min: {stalledNames} " +
-                          $"(sidecarRegistered={snapshot.SidecarRegistered}) — the ESP waits on " +
-                          "Setup/Apps providers without any timeout",
+                Message = message,
                 Data = data
             });
             _startupGate?.MarkEmitted(Constants.EventTypes.EspPolicyProviderStalled);
 
             _logger?.Warning(
-                $"EspPolicyProviderStallDetector: {stalledNames} incomplete for >= {DwellMinutes} min " +
-                $"(sidecarRegistered={snapshot.SidecarRegistered}).");
+                $"EspPolicyProviderStallDetector: {string.Join("; ", messageParts)} " +
+                $"(reason={reason}, sidecarRegistered={snapshot.SidecarRegistered}).");
         }
 
         /// <summary>Result of one <see cref="EvaluateStallTransition"/> round.</summary>
         internal sealed class StallEvaluation
         {
-            /// <summary>Next round's dwell state — only currently-incomplete providers survive.</summary>
+            /// <summary>Next round's arm-1 dwell state — only currently-incomplete providers survive.</summary>
             public Dictionary<string, DateTime> UpdatedFirstSeenUtc { get; } =
                 new Dictionary<string, DateTime>(StringComparer.OrdinalIgnoreCase);
 
-            /// <summary>Providers that crossed the dwell threshold this round and were not yet reported.</summary>
+            /// <summary>Providers that crossed the dwell threshold this round and were not yet reported (arm 1).</summary>
             public List<StalledProvider> NewlyStalled { get; } = new List<StalledProvider>();
+
+            /// <summary>Next round's arm-2 dwell state; null = sidecar-missing condition not active (reset).</summary>
+            public DateTime? UpdatedSidecarMissingSinceUtc { get; set; }
+
+            /// <summary>Non-null when the sidecar-missing condition crossed the dwell threshold this round and was not yet reported (arm 2).</summary>
+            public double? SidecarMissingStalledForMinutes { get; set; }
         }
 
         internal readonly struct StalledProvider

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Interop/UefiSecureBootCertReader.cs

@@ -1,5 +1,6 @@
 using System;
 using System.Runtime.InteropServices;
+using Microsoft.Win32.SafeHandles;
 
 namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
 {
@@ -56,6 +57,13 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
         private const int InitialBufferSize = 64 * 1024;
         private const int RetryBufferSize = 256 * 1024;
 
+        // SeSystemEnvironmentPrivilege is adjusted on the PROCESS token, so enable → read →
+        // restore is a process-wide side effect and must run as one atomic sequence. Without
+        // this gate a second reader would observe "already enabled", correctly skip its own
+        // restore, and then have the privilege disabled underneath it when the first reader
+        // finishes — a spurious privilege_denied on a machine that actually holds it.
+        private static readonly object FirmwareReadGate = new object();
+
         /// <summary>
         /// Reads db and KEK and evaluates the certificate presence flags. Never throws;
         /// per-variable failures surface as <see cref="UefiSecureBootCertSnapshot.DbStatus"/> /
@@ -67,6 +75,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
             var snapshot = new UefiSecureBootCertSnapshot();
             try
             {
+                // The critical section covers the whole privileged window, not just the
+                // adjustment — see FirmwareReadGate. Only synchronous kernel calls and byte
+                // scans run inside it, so it cannot block on anything else the agent does.
+                lock (FirmwareReadGate)
                 using (var privilege = SystemEnvironmentPrivilege.TryEnable())
                 {
                     if (!privilege.Enabled)
@@ -168,23 +180,42 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
         /// of the instance and restores the previous privilege state on dispose. SYSTEM holds
         /// the privilege (default-disabled); without elevation <see cref="Enabled"/> stays
         /// false and the caller reports <see cref="StatusPrivilegeDenied"/>.
+        /// <para>
+        /// <b>Crash-safety.</b> This is the repo's only privilege-adjustment code, so the
+        /// ownership rules are spelled out: the token is held in a <see cref="SafeHandle"/>
+        /// assigned by the P/Invoke marshaler, so the CLR owns it from the moment
+        /// <c>OpenProcessToken</c> returns — it is released by the critical finalizer even if
+        /// this instance never reaches the caller's <c>using</c> (async exception between the
+        /// call and the assignment). <see cref="Dispose"/> is idempotent, never throws, and
+        /// releases the handle in a <c>finally</c> so a failed restore cannot leak it. The
+        /// restore obligation is recorded immediately after the successful adjustment, so no
+        /// throw in between can strand the process with the privilege left enabled.
+        /// </para>
         /// </summary>
         private sealed class SystemEnvironmentPrivilege : IDisposable
         {
             public bool Enabled { get; private set; }
 
-            private IntPtr _token = IntPtr.Zero;
+            private SafeTokenHandle _token;
             private NativeMethods.TOKEN_PRIVILEGES _previousState;
             private bool _restoreNeeded;
+            private bool _disposed;
 
             public static SystemEnvironmentPrivilege TryEnable()
             {
                 var instance = new SystemEnvironmentPrivilege();
                 try
                 {
+                    SafeTokenHandle token;
                     if (!NativeMethods.OpenProcessToken(NativeMethods.GetCurrentProcess(),
-                            NativeMethods.TOKEN_ADJUST_PRIVILEGES | NativeMethods.TOKEN_QUERY, out instance._token))
+                            NativeMethods.TOKEN_ADJUST_PRIVILEGES | NativeMethods.TOKEN_QUERY, out token))
+                    {
+                        // Failed open still yields a (closed/invalid) SafeHandle instance.
+                        if (token != null) token.Dispose();
                         return instance;
+                    }
+
+                    instance._token = token;
 
                     NativeMethods.LUID luid;
                     if (!NativeMethods.LookupPrivilegeValueW(null, "SeSystemEnvironmentPrivilege", out luid))
@@ -197,52 +228,80 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
                         Attributes = NativeMethods.SE_PRIVILEGE_ENABLED
                     };
 
+                    NativeMethods.TOKEN_PRIVILEGES previousState;
                     uint returnLength;
-                    if (!NativeMethods.AdjustTokenPrivileges(instance._token, false, ref newState,
+                    if (!NativeMethods.AdjustTokenPrivileges(token, false, ref newState,
                             (uint)Marshal.SizeOf(typeof(NativeMethods.TOKEN_PRIVILEGES)),
-                            out instance._previousState, out returnLength))
+                            out previousState, out returnLength))
                         return instance;
 
                     // AdjustTokenPrivileges returns TRUE even when the privilege is not held —
-                    // the real verdict is ERROR_NOT_ALL_ASSIGNED in the last error.
+                    // the real verdict is ERROR_NOT_ALL_ASSIGNED in the last error. Nothing was
+                    // adjusted in that case, so there is nothing to restore.
                     if (Marshal.GetLastWin32Error() == ErrorNotAllAssigned)
                         return instance;
 
-                    // Restore only when the privilege was not already enabled before.
-                    instance._restoreNeeded = instance._previousState.PrivilegeCount > 0
-                        && (instance._previousState.Attributes & NativeMethods.SE_PRIVILEGE_ENABLED) == 0;
+                    // The token is now modified: record the restore obligation before anything
+                    // else can throw. Restore only when the privilege was not already enabled.
+                    instance._previousState = previousState;
+                    instance._restoreNeeded = previousState.PrivilegeCount > 0
+                        && (previousState.Attributes & NativeMethods.SE_PRIVILEGE_ENABLED) == 0;
                     instance.Enabled = true;
                 }
                 catch
                 {
-                    // Fail-soft: Enabled stays false.
+                    // Fail-soft: Enabled stays false. Whatever was acquired before the failure
+                    // is already tracked on the instance and released by Dispose.
                 }
                 return instance;
             }
 
             public void Dispose()
             {
-                if (_token != IntPtr.Zero)
+                if (_disposed) return;
+                _disposed = true;
+                Enabled = false;
+
+                var token = _token;
+                _token = null;
+                if (token == null) return;
+
+                try
                 {
-                    try
+                    if (_restoreNeeded)
                     {
-                        if (_restoreNeeded)
-                        {
-                            NativeMethods.TOKEN_PRIVILEGES ignored;
-                            uint returnLength;
-                            NativeMethods.AdjustTokenPrivileges(_token, false, ref _previousState,
-                                (uint)Marshal.SizeOf(typeof(NativeMethods.TOKEN_PRIVILEGES)),
-                                out ignored, out returnLength);
-                        }
+                        _restoreNeeded = false;
+                        NativeMethods.TOKEN_PRIVILEGES ignored;
+                        uint returnLength;
+                        NativeMethods.AdjustTokenPrivileges(token, false, ref _previousState,
+                            (uint)Marshal.SizeOf(typeof(NativeMethods.TOKEN_PRIVILEGES)),
+                            out ignored, out returnLength);
                     }
-                    catch
-                    {
-                        // Best effort — the process token privilege state is not load-bearing
-                        // for anything else the agent does.
-                    }
-                    NativeMethods.CloseHandle(_token);
-                    _token = IntPtr.Zero;
                 }
+                catch
+                {
+                    // Best effort — the process token privilege state is not load-bearing for
+                    // anything else the agent does, and Dispose must never throw.
+                }
+                finally
+                {
+                    token.Dispose();
+                }
+            }
+        }
+
+        /// <summary>
+        /// Owns a process-token handle. <c>ReleaseHandle</c> runs on the finalizer thread when
+        /// <see cref="SystemEnvironmentPrivilege.Dispose"/> is skipped, so it must not throw —
+        /// a single <c>CloseHandle</c> P/Invoke satisfies that.
+        /// </summary>
+        private sealed class SafeTokenHandle : SafeHandleZeroOrMinusOneIsInvalid
+        {
+            public SafeTokenHandle() : base(true) { }
+
+            protected override bool ReleaseHandle()
+            {
+                return NativeMethods.CloseHandle(handle);
             }
         }
 
@@ -278,7 +337,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
 
             [DllImport("advapi32.dll", SetLastError = true)]
             [return: MarshalAs(UnmanagedType.Bool)]
-            public static extern bool OpenProcessToken(IntPtr processHandle, uint desiredAccess, out IntPtr tokenHandle);
+            public static extern bool OpenProcessToken(IntPtr processHandle, uint desiredAccess, out SafeTokenHandle tokenHandle);
 
             [DllImport("advapi32.dll", SetLastError = true, CharSet = CharSet.Unicode)]
             [return: MarshalAs(UnmanagedType.Bool)]
@@ -286,7 +345,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
 
             [DllImport("advapi32.dll", SetLastError = true)]
             [return: MarshalAs(UnmanagedType.Bool)]
-            public static extern bool AdjustTokenPrivileges(IntPtr tokenHandle, bool disableAllPrivileges,
+            public static extern bool AdjustTokenPrivileges(SafeTokenHandle tokenHandle, bool disableAllPrivileges,
                 ref TOKEN_PRIVILEGES newState, uint bufferLength, out TOKEN_PRIVILEGES previousState, out uint returnLength);
 
             [DllImport("kernel32.dll", SetLastError = true)]

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Interop/WifiInfoReader.cs

@@ -0,0 +1,194 @@
+using System;
+using System.Runtime.InteropServices;
+using System.Text;
+using static AutopilotMonitor.Agent.V2.Core.Monitoring.Interop.WlanNativeMethods;
+
+namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
+{
+    /// <summary>
+    /// Current WiFi connection details as read from the native WLAN API.
+    /// </summary>
+    public sealed class WifiConnectionInfo
+    {
+        /// <summary>SSID of the connected network; null when the AP broadcasts an empty SSID.</summary>
+        public string Ssid { get; internal set; }
+
+        /// <summary>Signal quality 0–100 (same semantics as the netsh "Signal" percentage).</summary>
+        public int SignalPercent { get; internal set; }
+
+        /// <summary>Human-readable PHY generation ("802.11ax" …); null for exotic/unknown PHYs.</summary>
+        public string RadioType { get; internal set; }
+
+        /// <summary>Current 802.11 channel number; null when the channel query fails.</summary>
+        public int? Channel { get; internal set; }
+    }
+
+    /// <summary>
+    /// Fail-soft reader over <see cref="WlanNativeMethods"/>: returns the current WiFi
+    /// connection or null — never throws, never blocks on a child process. Machines without
+    /// the WLAN AutoConfig service (VMs, wired-only devices) simply return null.
+    /// </summary>
+    public static class WifiInfoReader
+    {
+        /// <summary>
+        /// Reads the current WiFi connection. When <paramref name="preferredInterfaceId"/> is
+        /// given (the adapter GUID from <c>NetworkInterface.Id</c>), that interface is queried
+        /// first; otherwise — or when it is not connected — the first connected WLAN interface
+        /// wins. Returns null when no interface has an active connection.
+        /// </summary>
+        public static WifiConnectionInfo TryGetCurrentConnection(Guid? preferredInterfaceId = null)
+        {
+            var clientHandle = IntPtr.Zero;
+            try
+            {
+                if (WlanOpenHandle(WLAN_CLIENT_VERSION, IntPtr.Zero, out _, out clientHandle) != ERROR_SUCCESS)
+                    return null;
+
+                var interfaceList = IntPtr.Zero;
+                try
+                {
+                    if (WlanEnumInterfaces(clientHandle, IntPtr.Zero, out interfaceList) != ERROR_SUCCESS)
+                        return null;
+
+                    var count = (uint)Marshal.ReadInt32(interfaceList);
+                    var infoSize = Marshal.SizeOf(typeof(WLAN_INTERFACE_INFO));
+                    // List header: DWORD dwNumberOfItems + DWORD dwIndex, then the info array.
+                    var firstEntry = new IntPtr(interfaceList.ToInt64() + 8);
+
+                    WifiConnectionInfo fallback = null;
+                    for (var i = 0; i < count; i++)
+                    {
+                        var entryPtr = new IntPtr(firstEntry.ToInt64() + i * infoSize);
+                        var info = (WLAN_INTERFACE_INFO)Marshal.PtrToStructure(entryPtr, typeof(WLAN_INTERFACE_INFO));
+
+                        var connection = QueryCurrentConnection(clientHandle, info.InterfaceGuid);
+                        if (connection == null)
+                            continue;
+
+                        if (preferredInterfaceId.HasValue && info.InterfaceGuid == preferredInterfaceId.Value)
+                            return connection;
+
+                        if (fallback == null)
+                            fallback = connection;
+                    }
+
+                    return fallback;
+                }
+                finally
+                {
+                    if (interfaceList != IntPtr.Zero)
+                        WlanFreeMemory(interfaceList);
+                }
+            }
+            catch
+            {
+                return null;
+            }
+            finally
+            {
+                if (clientHandle != IntPtr.Zero)
+                    WlanCloseHandle(clientHandle, IntPtr.Zero);
+            }
+        }
+
+        private static WifiConnectionInfo QueryCurrentConnection(IntPtr clientHandle, Guid interfaceGuid)
+        {
+            var dataPtr = IntPtr.Zero;
+            try
+            {
+                if (WlanQueryInterface(clientHandle, ref interfaceGuid,
+                        WLAN_INTF_OPCODE.wlan_intf_opcode_current_connection,
+                        IntPtr.Zero, out _, out dataPtr, IntPtr.Zero) != ERROR_SUCCESS)
+                    return null;
+
+                var attributes = (WLAN_CONNECTION_ATTRIBUTES)Marshal.PtrToStructure(
+                    dataPtr, typeof(WLAN_CONNECTION_ATTRIBUTES));
+
+                if (attributes.isState != WLAN_INTERFACE_STATE.wlan_interface_state_connected)
+                    return null;
+
+                var assoc = attributes.wlanAssociationAttributes;
+                return new WifiConnectionInfo
+                {
+                    Ssid = DecodeSsid(assoc.dot11Ssid.ucSSID, assoc.dot11Ssid.uSSIDLength),
+                    SignalPercent = (int)Math.Min(assoc.wlanSignalQuality, 100),
+                    RadioType = PhyTypeToRadioType(assoc.dot11PhyType),
+                    Channel = QueryChannel(clientHandle, interfaceGuid),
+                };
+            }
+            finally
+            {
+                if (dataPtr != IntPtr.Zero)
+                    WlanFreeMemory(dataPtr);
+            }
+        }
+
+        private static int? QueryChannel(IntPtr clientHandle, Guid interfaceGuid)
+        {
+            var dataPtr = IntPtr.Zero;
+            try
+            {
+                if (WlanQueryInterface(clientHandle, ref interfaceGuid,
+                        WLAN_INTF_OPCODE.wlan_intf_opcode_channel_number,
+                        IntPtr.Zero, out var dataSize, out dataPtr, IntPtr.Zero) != ERROR_SUCCESS)
+                    return null;
+
+                if (dataSize < sizeof(uint) || dataPtr == IntPtr.Zero)
+                    return null;
+
+                return Marshal.ReadInt32(dataPtr);
+            }
+            catch
+            {
+                return null;
+            }
+            finally
+            {
+                if (dataPtr != IntPtr.Zero)
+                    WlanFreeMemory(dataPtr);
+            }
+        }
+
+        /// <summary>
+        /// Decodes a raw DOT11_SSID (UTF-8 by convention) into a string; null for empty SSIDs.
+        /// Internal for direct unit-testing via InternalsVisibleTo.
+        /// </summary>
+        internal static string DecodeSsid(byte[] raw, uint length)
+        {
+            if (raw == null || length == 0)
+                return null;
+
+            var byteCount = (int)Math.Min(length, (uint)raw.Length);
+            try
+            {
+                var ssid = Encoding.UTF8.GetString(raw, 0, byteCount);
+                return string.IsNullOrEmpty(ssid) ? null : ssid;
+            }
+            catch
+            {
+                return null;
+            }
+        }
+
+        /// <summary>
+        /// Maps a DOT11 PHY type to the same strings netsh historically reported as
+        /// "Radio type" — keeps event payloads consistent across the netsh→WLAN-API switch.
+        /// Internal for direct unit-testing via InternalsVisibleTo.
+        /// </summary>
+        internal static string PhyTypeToRadioType(DOT11_PHY_TYPE phyType)
+        {
+            switch (phyType)
+            {
+                case DOT11_PHY_TYPE.dot11_phy_type_ofdm: return "802.11a";
+                case DOT11_PHY_TYPE.dot11_phy_type_hrdsss: return "802.11b";
+                case DOT11_PHY_TYPE.dot11_phy_type_erp: return "802.11g";
+                case DOT11_PHY_TYPE.dot11_phy_type_ht: return "802.11n";
+                case DOT11_PHY_TYPE.dot11_phy_type_vht: return "802.11ac";
+                case DOT11_PHY_TYPE.dot11_phy_type_dmg: return "802.11ad";
+                case DOT11_PHY_TYPE.dot11_phy_type_he: return "802.11ax";
+                case DOT11_PHY_TYPE.dot11_phy_type_eht: return "802.11be";
+                default: return null;
+            }
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Interop/WlanNativeMethods.cs

@@ -0,0 +1,142 @@
+using System;
+using System.Runtime.InteropServices;
+
+namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Interop
+{
+    /// <summary>
+    /// Native WLAN API interop (<c>wlanapi.dll</c>) used to read the current WiFi connection
+    /// (SSID, signal quality, PHY type, channel) without spawning <c>netsh</c>.
+    /// <para>
+    /// The WLAN API is language-neutral — unlike parsing <c>netsh wlan show interfaces</c>
+    /// output, whose labels follow the OS UI language and silently drop fields on
+    /// non-English images. On machines without the WLAN AutoConfig service (VMs, servers,
+    /// wired-only devices) <c>WlanOpenHandle</c>/<c>WlanEnumInterfaces</c> fail with a
+    /// non-zero error code; callers treat that as "no WiFi" (see <see cref="WifiInfoReader"/>).
+    /// </para>
+    /// </summary>
+    internal static class WlanNativeMethods
+    {
+        internal const uint ERROR_SUCCESS = 0;
+
+        /// <summary>Client version 2 = Vista and later (the only relevant range).</summary>
+        internal const uint WLAN_CLIENT_VERSION = 2;
+
+        internal enum WLAN_INTF_OPCODE : uint
+        {
+            wlan_intf_opcode_current_connection = 7,
+            wlan_intf_opcode_channel_number = 8,
+        }
+
+        internal enum WLAN_INTERFACE_STATE
+        {
+            wlan_interface_state_not_ready = 0,
+            wlan_interface_state_connected = 1,
+            wlan_interface_state_ad_hoc_network_formed = 2,
+            wlan_interface_state_disconnecting = 3,
+            wlan_interface_state_disconnected = 4,
+            wlan_interface_state_associating = 5,
+            wlan_interface_state_discovering = 6,
+            wlan_interface_state_authenticating = 7,
+        }
+
+        internal enum DOT11_PHY_TYPE : uint
+        {
+            dot11_phy_type_unknown = 0,
+            dot11_phy_type_fhss = 1,
+            dot11_phy_type_dsss = 2,
+            dot11_phy_type_irbaseband = 3,
+            dot11_phy_type_ofdm = 4,    // 802.11a
+            dot11_phy_type_hrdsss = 5,  // 802.11b
+            dot11_phy_type_erp = 6,     // 802.11g
+            dot11_phy_type_ht = 7,      // 802.11n
+            dot11_phy_type_vht = 8,     // 802.11ac
+            dot11_phy_type_dmg = 9,     // 802.11ad
+            dot11_phy_type_he = 10,     // 802.11ax
+            dot11_phy_type_eht = 11,    // 802.11be
+        }
+
+        [StructLayout(LayoutKind.Sequential, CharSet = CharSet.Unicode)]
+        internal struct WLAN_INTERFACE_INFO
+        {
+            public Guid InterfaceGuid;
+            [MarshalAs(UnmanagedType.ByValTStr, SizeConst = 256)]
+            public string strInterfaceDescription;
+            public WLAN_INTERFACE_STATE isState;
+        }
+
+        [StructLayout(LayoutKind.Sequential)]
+        internal struct DOT11_SSID
+        {
+            public uint uSSIDLength;
+            [MarshalAs(UnmanagedType.ByValArray, SizeConst = 32)]
+            public byte[] ucSSID;
+        }
+
+        [StructLayout(LayoutKind.Sequential)]
+        internal struct WLAN_ASSOCIATION_ATTRIBUTES
+        {
+            public DOT11_SSID dot11Ssid;
+            public int dot11BssType;
+            [MarshalAs(UnmanagedType.ByValArray, SizeConst = 6)]
+            public byte[] dot11Bssid;
+            public DOT11_PHY_TYPE dot11PhyType;
+            public uint uDot11PhyIndex;
+            public uint wlanSignalQuality; // 0..100, same semantics as netsh "Signal"
+            public uint ulRxRate;
+            public uint ulTxRate;
+        }
+
+        [StructLayout(LayoutKind.Sequential)]
+        internal struct WLAN_SECURITY_ATTRIBUTES
+        {
+            [MarshalAs(UnmanagedType.Bool)]
+            public bool bSecurityEnabled;
+            [MarshalAs(UnmanagedType.Bool)]
+            public bool bOneXEnabled;
+            public int dot11AuthAlgorithm;
+            public int dot11CipherAlgorithm;
+        }
+
+        [StructLayout(LayoutKind.Sequential, CharSet = CharSet.Unicode)]
+        internal struct WLAN_CONNECTION_ATTRIBUTES
+        {
+            public WLAN_INTERFACE_STATE isState;
+            public int wlanConnectionMode;
+            [MarshalAs(UnmanagedType.ByValTStr, SizeConst = 256)]
+            public string strProfileName;
+            public WLAN_ASSOCIATION_ATTRIBUTES wlanAssociationAttributes;
+            public WLAN_SECURITY_ATTRIBUTES wlanSecurityAttributes;
+        }
+
+        [DllImport("wlanapi.dll")]
+        internal static extern uint WlanOpenHandle(
+            uint dwClientVersion,
+            IntPtr pReserved,
+            out uint pdwNegotiatedVersion,
+            out IntPtr phClientHandle);
+
+        [DllImport("wlanapi.dll")]
+        internal static extern uint WlanCloseHandle(
+            IntPtr hClientHandle,
+            IntPtr pReserved);
+
+        [DllImport("wlanapi.dll")]
+        internal static extern uint WlanEnumInterfaces(
+            IntPtr hClientHandle,
+            IntPtr pReserved,
+            out IntPtr ppInterfaceList);
+
+        [DllImport("wlanapi.dll")]
+        internal static extern uint WlanQueryInterface(
+            IntPtr hClientHandle,
+            [In] ref Guid pInterfaceGuid,
+            WLAN_INTF_OPCODE OpCode,
+            IntPtr pReserved,
+            out uint pdwDataSize,
+            out IntPtr ppData,
+            IntPtr pWlanOpcodeValueType);
+
+        [DllImport("wlanapi.dll")]
+        internal static extern void WlanFreeMemory(IntPtr pMemory);
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Telemetry/DeviceInfo/DeviceInfoCollector.NetworkAndSecurity.cs

@@ -706,10 +706,27 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.DeviceInfo
         /// the desktop via "Continue anyway".
         /// </para>
         /// </summary>
+        // Serializes the ESP-config read+emit across ALL callers (CollectAll on the startup
+        // ThreadPool task, CollectAtEnrollmentStart and RefreshEspConfiguration on their own
+        // Task.Run) — without it, two concurrent collects can interleave read-before-emit and
+        // the LATER emission can carry the OLDER (smaller) registry snapshot. The backend keeps
+        // the latest list-bearing emission as positive blocking evidence, so an out-of-order
+        // emission would silently shrink it (Codex review). Inside the lock each collect is
+        // read→emit atomic: a later emission always carries an equal-or-newer snapshot.
+        private readonly object _espConfigCollectLock = new object();
+
         // Internal (not private) as a test seam: unlike CollectAll/CollectAtEnrollmentStart this
         // method touches no live system state once both probes are overridden, so tests can
         // drive the esp_config_detected event surface in isolation (InternalsVisibleTo).
         internal (bool? skipUserStatusPage, bool? skipDeviceStatusPage) CollectEspConfiguration()
+        {
+            lock (_espConfigCollectLock)
+            {
+                return CollectEspConfigurationLocked();
+            }
+        }
+
+        private (bool? skipUserStatusPage, bool? skipDeviceStatusPage) CollectEspConfigurationLocked()
         {
             bool? skipUser = null;
             bool? skipDevice = null;
@@ -798,7 +815,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.DeviceInfo
         /// Collects active network interface details (name, type, link speed, MAC) using
         /// System.Net.NetworkInformation — instant, no process spawn.
         /// If the active interface is WiFi, fires off a separate async task to collect
-        /// WiFi signal info via netsh (fire-and-forget, never blocks).
+        /// WiFi signal info via the native WLAN API (fire-and-forget, never blocks).
         /// </summary>
         private void CollectActiveNetworkInterfaceInfo()
         {
@@ -841,7 +858,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.DeviceInfo
                 // Fire-and-forget WiFi signal collection — separate event, never blocks
                 if (isWifi)
                 {
-                    Task.Run(() => CollectWiFiSignalInfo());
+                    Guid? nicGuid = Guid.TryParse(activeNic.Id, out var parsedGuid) ? parsedGuid : (Guid?)null;
+                    Task.Run(() => CollectWiFiSignalInfo(nicGuid));
                 }
             }
             catch (Exception ex)
@@ -882,78 +900,37 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.DeviceInfo
         }
 
         /// <summary>
-        /// Collects WiFi signal info via netsh wlan show interfaces and emits a separate
-        /// wifi_signal_info event. Designed to run as fire-and-forget via Task.Run —
-        /// never blocks other collection. VMs without WiFi service simply get no event.
+        /// Collects WiFi signal info via the native WLAN API (<see cref="WifiInfoReader"/>,
+        /// language-neutral, no process spawn) and emits a separate wifi_signal_info event.
+        /// Designed to run as fire-and-forget via Task.Run — never blocks other collection.
+        /// VMs without WiFi service simply get no event.
         /// </summary>
-        private void CollectWiFiSignalInfo()
+        private void CollectWiFiSignalInfo(Guid? interfaceGuid)
         {
             try
             {
-                var psi = new ProcessStartInfo
-                {
-                    FileName = SystemPaths.Netsh,
-                    Arguments = "wlan show interfaces",
-                    UseShellExecute = false,
-                    RedirectStandardOutput = true,
-                    RedirectStandardError = true,
-                    CreateNoWindow = true
-                };
-
-                string output;
-                using (var process = Process.Start(psi))
-                {
-                    output = process.StandardOutput.ReadToEnd();
-                    process.WaitForExit(5000);
-                }
-
-                if (string.IsNullOrEmpty(output))
+                var wifi = WifiInfoReader.TryGetCurrentConnection(interfaceGuid);
+                if (wifi == null)
                     return;
 
-                var data = new Dictionary<string, object>();
-
-                foreach (var line in output.Split(new[] { '\r', '\n' }, StringSplitOptions.RemoveEmptyEntries))
+                var data = new Dictionary<string, object>
                 {
-                    var trimmed = line.Trim();
-                    var colonIndex = trimmed.IndexOf(':');
-                    if (colonIndex < 0) continue;
+                    ["wifiSignalPercent"] = wifi.SignalPercent
+                };
+                if (wifi.Ssid != null)
+                    data["wifiSsid"] = wifi.Ssid;
+                if (wifi.RadioType != null)
+                    data["wifiRadioType"] = wifi.RadioType;
+                if (wifi.Channel.HasValue)
+                    data["wifiChannel"] = wifi.Channel.Value;
 
-                    var key = trimmed.Substring(0, colonIndex).Trim();
-                    var value = trimmed.Substring(colonIndex + 1).Trim();
+                var message = wifi.Ssid != null
+                    ? $"WiFi: {wifi.Ssid}, Signal: {wifi.SignalPercent}%"
+                    : $"WiFi signal info, Signal: {wifi.SignalPercent}%";
+                if (wifi.RadioType != null)
+                    message += $" ({wifi.RadioType})";
 
-                    if (key.Equals("SSID", StringComparison.OrdinalIgnoreCase) &&
-                        !key.Equals("BSSID", StringComparison.OrdinalIgnoreCase))
-                    {
-                        data["wifiSsid"] = value;
-                    }
-                    else if (key.Equals("Signal", StringComparison.OrdinalIgnoreCase))
-                    {
-                        if (int.TryParse(value.TrimEnd('%'), out var signal))
-                            data["wifiSignalPercent"] = signal;
-                    }
-                    else if (key.Equals("Radio type", StringComparison.OrdinalIgnoreCase))
-                    {
-                        data["wifiRadioType"] = value;
-                    }
-                    else if (key.Equals("Channel", StringComparison.OrdinalIgnoreCase))
-                    {
-                        if (int.TryParse(value, out var channel))
-                            data["wifiChannel"] = channel;
-                    }
-                }
-
-                if (data.Count > 0)
-                {
-                    var message = data.ContainsKey("wifiSsid")
-                        ? $"WiFi: {data["wifiSsid"]}"
-                        : "WiFi signal info";
-                    if (data.ContainsKey("wifiSignalPercent"))
-                        message += $", Signal: {data["wifiSignalPercent"]}%";
-                    if (data.ContainsKey("wifiRadioType"))
-                        message += $" ({data["wifiRadioType"]})";
-
-                    EmitDeviceInfoEvent(Constants.EventTypes.WifiSignalInfo, message, data);
-                }
+                EmitDeviceInfoEvent(Constants.EventTypes.WifiSignalInfo, message, data);
             }
             catch (Exception ex)
             {

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Telemetry/DeviceInfo/DeviceInfoCollector.cs

@@ -114,6 +114,23 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.DeviceInfo
             CollectTpmStatus();
         }
 
+        /// <summary>
+        /// Re-reads the ESP configuration (skip flags + ESPTrackingInfo blocking lists) and
+        /// re-emits <c>esp_config_detected</c> when the payload changed. Audit Q2: the earlier
+        /// emissions (agent start, DeviceSetup detection) fire before the registry lists are
+        /// complete — the ESP writes one timestamped subkey per CSP status write, so the
+        /// device-scope list grows progressively and the user-scope <c>S-&lt;SID&gt;</c> lists
+        /// appear only after sign-in. Called by <see cref="Orchestration.DeviceInfoHost"/> as
+        /// phase-driven one-shots (apps sub-phase opened, AccountSetup detected) — never on a
+        /// timer. The StartupEventGate keeps an unchanged payload silent, so a re-collect that
+        /// found nothing new emits nothing.
+        /// </summary>
+        public void RefreshEspConfiguration(string trigger)
+        {
+            _logger.Info($"EnrollmentTracker: re-collecting ESP configuration (trigger={trigger})");
+            CollectEspConfiguration();
+        }
+
         /// <summary>
         /// Re-collects device info that may change during enrollment (e.g. BitLocker enabled via policy).
         /// Called at enrollment complete / FinalizingSetup transition to capture final state.

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Telemetry/Periodic/AgentSelfMetricsCollector.cs

@@ -77,11 +77,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.Periodic
 
         protected override void Collect()
         {
-            // Full path writes 18 keys: agent_version + process metrics (5: cpu, ws, private,
-            // threads, handles) + spool stats (4) + network delta (8: requests, failures,
-            // bytes_up/down, avg_latency, total_up/down/requests). cap=18 → HashHelpers.GetPrime(18)=23
-            // buckets → no resize on the 18th key (cap=16 would land at 17 buckets and resize).
-            var data = new Dictionary<string, object>(capacity: 18, StringComparer.Ordinal)
+            // Full path writes 19 keys: agent_version + process metrics (5: cpu, ws, private,
+            // threads, handles) + spool stats (4) + network delta (9: requests, failures,
+            // bytes_up/down, avg_latency, total_up/down/requests/latency). cap=19 →
+            // HashHelpers.GetPrime(19)=23 buckets → no resize on the 19th key.
+            var data = new Dictionary<string, object>(capacity: 19, StringComparer.Ordinal)
             {
                 { "agent_version", _agentVersion }
             };
@@ -164,6 +164,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.Periodic
                 data["net_total_bytes_up"] = currentNet.TotalBytesUp;
                 data["net_total_bytes_down"] = currentNet.TotalBytesDown;
                 data["net_total_requests"] = currentNet.RequestCount;
+                // Cumulative latency sum: total/requests = session-wide average HTTP
+                // round-trip. The backend projects that average onto the Session row
+                // (AvgApiLatencyMs) — cumulative counters make the write idempotent.
+                data["net_total_latency_ms"] = currentNet.TotalLatencyMs;
             }
             catch (Exception ex)
             {

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Telemetry/Periodic/NetworkChangeDetector.cs

@@ -8,8 +8,8 @@ using System.Net.Sockets;
 using System.Threading;
 using System.Threading.Tasks;
 using AutopilotMonitor.Agent.V2.Core.Logging;
+using AutopilotMonitor.Agent.V2.Core.Monitoring.Interop;
 using AutopilotMonitor.Agent.V2.Core.Orchestration;
-using AutopilotMonitor.Agent.V2.Core.Security;
 using AutopilotMonitor.Shared;
 using AutopilotMonitor.Shared.Models;
 
@@ -37,7 +37,6 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.Periodic
         private const int DebounceDelayMs = 3000;
         private const int ConnectivityCheckDelayMs = 5000;
         private const int ConnectivityCheckTimeoutMs = 8000;
-        private const int WifiCollectTimeoutMs = 3000;
         private const string Source = "NetworkChangeDetector";
 
         /// <summary>
@@ -374,7 +373,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.Periodic
                 // Collect WiFi SSID synchronously (needed for before/after comparison)
                 if (isWifi)
                 {
-                    var wifiInfo = CollectWiFiInfo();
+                    Guid? nicGuid = Guid.TryParse(activeNic.Id, out var parsedGuid) ? parsedGuid : (Guid?)null;
+                    var wifiInfo = CollectWiFiInfo(nicGuid);
                     snapshot.WifiSsid = wifiInfo.ssid;
                     snapshot.WifiSignalPercent = wifiInfo.signal;
                     snapshot.WifiRadioType = wifiInfo.radioType;
@@ -390,73 +390,24 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Telemetry.Periodic
         }
 
         /// <summary>
-        /// Collects WiFi SSID, signal, and radio type via netsh wlan show interfaces.
-        /// Synchronous with timeout — needed before emitting the event.
+        /// Collects WiFi SSID, signal, and radio type via the native WLAN API
+        /// (<see cref="WifiInfoReader"/> — language-neutral, no process spawn).
+        /// Synchronous — needed before emitting the event.
         /// </summary>
-        private (string ssid, int? signal, string radioType) CollectWiFiInfo()
+        private (string ssid, int? signal, string radioType) CollectWiFiInfo(Guid? interfaceGuid)
         {
-            string ssid = null;
-            int? signal = null;
-            string radioType = null;
-
             try
             {
-                var psi = new ProcessStartInfo
-                {
-                    FileName = SystemPaths.Netsh,
-                    Arguments = "wlan show interfaces",
-                    UseShellExecute = false,
-                    RedirectStandardOutput = true,
-                    RedirectStandardError = true,
-                    CreateNoWindow = true
-                };
-
-                string output;
-                using (var process = Process.Start(psi))
-                {
-                    output = process.StandardOutput.ReadToEnd();
-                    if (!process.WaitForExit(WifiCollectTimeoutMs))
-                    {
-                        try { process.Kill(); } catch { }
-                        _logger.Debug("NetworkChangeDetector: netsh timed out");
-                        return (null, null, null);
-                    }
-                }
-
-                if (string.IsNullOrEmpty(output))
-                    return (null, null, null);
-
-                foreach (var line in output.Split(new[] { '\r', '\n' }, StringSplitOptions.RemoveEmptyEntries))
-                {
-                    var trimmed = line.Trim();
-                    var colonIndex = trimmed.IndexOf(':');
-                    if (colonIndex < 0) continue;
-
-                    var key = trimmed.Substring(0, colonIndex).Trim();
-                    var value = trimmed.Substring(colonIndex + 1).Trim();
-
-                    if (key.Equals("SSID", StringComparison.OrdinalIgnoreCase) &&
-                        !key.Equals("BSSID", StringComparison.OrdinalIgnoreCase))
-                    {
-                        ssid = value;
-                    }
-                    else if (key.Equals("Signal", StringComparison.OrdinalIgnoreCase))
-                    {
-                        if (int.TryParse(value.TrimEnd('%'), out var sig))
-                            signal = sig;
-                    }
-                    else if (key.Equals("Radio type", StringComparison.OrdinalIgnoreCase))
-                    {
-                        radioType = value;
-                    }
-                }
+                var wifi = WifiInfoReader.TryGetCurrentConnection(interfaceGuid);
+                if (wifi != null)
+                    return (wifi.Ssid, wifi.SignalPercent, wifi.RadioType);
             }
             catch (Exception ex)
             {
                 _logger.Debug($"NetworkChangeDetector: WiFi info collection failed: {ex.Message}");
             }
 
-            return (ssid, signal, radioType);
+            return (null, null, null);
         }
 
         // -------------------------------------------------------------------

src/Agent/AutopilotMonitor.Agent.V2.Core/Orchestration/DefaultComponentFactory.cs

@@ -37,8 +37,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
     ///     <see cref="StallProbeCollectorAdapter"/>. Owns its 60-s idle-check timer (the
     ///     collector itself has no timer — it's a pure probe invoked from outside).</item>
     ///   <item><b>EspPolicyProviderStallHost</b> — always-on 60-s wall-clock tick for the
-    ///     <c>esp_policy_provider_stalled</c> tripwire (registered EnrollmentStatusTracking CSP
-    ///     provider continuously incomplete ≥ 15 min — the no-timeout ESP wait).</item>
+    ///     two-arm <c>esp_policy_provider_stalled</c> tripwire (EnrollmentStatusTracking CSP
+    ///     provider continuously incomplete ≥ 15 min, OR Setup\Apps providers registered
+    ///     without the expected <c>Sidecar</c>/IME provider — the no-timeout ESP wait).</item>
     /// </list>
     /// Optional peripheral hosts (driven by <see cref="CollectorConfiguration"/> toggles):
     /// <list type="bullet">
@@ -307,11 +308,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
             }
 
             // ESP policy-provider stall tripwire — always-on kernel host (no config gate, like
-            // disk_space_low): a registered EnrollmentStatusTracking CSP provider that never
-            // completes (e.g. co-management "ConfigMgr" without "Sidecar" on a pre-provisioned
-            // device) parks the user ESP at "Apps (Identifying)" WITHOUT any OS timeout. Wall-clock
-            // dwell, deliberately outside the StallProbe gate: the idle clock there resets on any
-            // session activity, and the stall must accrue regardless of it.
+            // disk_space_low): a provider that never completes, or co-management "ConfigMgr"
+            // registered without the expected "Sidecar"/IME provider (issue #106 — even with
+            // TrackingPoliciesCreated=1 the ESP ignores foreign provider names), parks the user
+            // ESP at "Apps (Identifying)" WITHOUT any OS timeout. Wall-clock dwell, deliberately
+            // outside the StallProbe gate: the idle clock there resets on any session activity,
+            // and the stall must accrue regardless of it.
             hosts.Add(new EspPolicyProviderStallHost(
                 sessionId: sessionId,
                 tenantId: tenantId,

src/Agent/AutopilotMonitor.Agent.V2.Core/Orchestration/Hosts/DeviceInfoHost.cs

@@ -36,6 +36,13 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
     ///   <item><c>CollectAtEnd</c> — once, on <c>FinalizingSetup</c> or desktop arrival
     ///   (whichever comes first; desktop arrival also covers no-ESP / WDP v2). Re-fetches
     ///   BitLocker (commonly enabled via policy DURING enrollment) + the active NIC.</item>
+    ///   <item><c>RefreshEspConfiguration</c> — audit Q2, one-shot per trigger: the ESP
+    ///   blocking lists in the registry grow progressively (device scope) and the user-scope
+    ///   lists appear only after sign-in, so the early emissions are structurally partial.
+    ///   Re-collected when the apps sub-phase opens (first app activity in DeviceSetup /
+    ///   AccountSetup — the adapter's <c>phase_transition</c> declaration) and when the
+    ///   AccountSetup ESP phase itself is detected. Phase-driven one-shots, never a timer
+    ///   (no-heartbeat policy).</item>
     /// </list>
     /// <para>
     /// Duplicate-event cost is zero: every re-collected event runs through the collector's
@@ -46,6 +53,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
     {
         public string Name => "DeviceInfoCollector";
 
+        // ESP-config refresh trigger labels (audit Q2) — also the values passed to
+        // DeviceInfoCollector.RefreshEspConfiguration for the agent-log line.
+        internal const string EspConfigTriggerAppsDevice = "apps_phase_device";
+        internal const string EspConfigTriggerAppsUser = "apps_phase_user";
+        internal const string EspConfigTriggerAccountSetup = "account_setup_detected";
+
         private readonly Monitoring.Telemetry.DeviceInfo.DeviceInfoCollector _collector;
         private readonly AgentLogger _logger;
 
@@ -58,6 +71,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
         private int _endCollected;
         private int _disposed;
 
+        // Audit Q2 — fire-once latches per ESP-config refresh trigger. Guarded by _espConfigLock
+        // instead of three Interlocked ints so the membership test + add stays one atomic step.
+        private readonly object _espConfigLock = new object();
+        private readonly HashSet<string> _espConfigRefreshesFired = new HashSet<string>(StringComparer.Ordinal);
+
         public DeviceInfoHost(
             string sessionId,
             string tenantId,
@@ -100,6 +118,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
 
         private void OnSignalPosted(DecisionSignalKind kind, IReadOnlyDictionary<string, string>? payload)
         {
+            HandleEspConfigRefreshTrigger(kind, payload);
+
             var startTrigger = IsEnrollmentStartTrigger(kind, payload);
             var endTrigger = IsEndTrigger(kind, payload);
             if (!startTrigger && !endTrigger) return;
@@ -111,10 +131,6 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
             var runEndCollect = endTrigger && Interlocked.Exchange(ref _endCollected, 1) == 0;
             if (!runStartRefresh && !runEndCollect) return;
 
-            // Both one-shots done → nothing left to observe.
-            if (Volatile.Read(ref _enrollmentStartCollected) == 1 && Volatile.Read(ref _endCollected) == 1)
-                Unsubscribe();
-
             var trigger = kind == DecisionSignalKind.DesktopArrived ? "desktop_arrived" : "esp_phase_changed";
             _logger.Info($"DeviceInfoHost: trigger '{trigger}' — scheduling re-collect (enrollmentStart={runStartRefresh}, end={runEndCollect}).");
 
@@ -134,6 +150,32 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
             });
         }
 
+        /// <summary>
+        /// Audit Q2 — ESP-config refresh one-shots. Each trigger label fires at most once per
+        /// agent run; the collector's StartupEventGate keeps an unchanged payload silent, so
+        /// the worst case is one silent registry re-read per trigger. The subscription stays
+        /// armed for the host's lifetime (unlike the start/end pair there is no "all triggers
+        /// seen" point that is guaranteed to arrive — SkipUser enrollments never enter
+        /// AccountSetup); the per-signal cost is a few dictionary lookups.
+        /// </summary>
+        private void HandleEspConfigRefreshTrigger(DecisionSignalKind kind, IReadOnlyDictionary<string, string>? payload)
+        {
+            var trigger = ClassifyEspConfigRefreshTrigger(kind, payload);
+            if (trigger == null) return;
+
+            lock (_espConfigLock)
+            {
+                if (!_espConfigRefreshesFired.Add(trigger)) return;
+            }
+
+            _logger.Info($"DeviceInfoHost: ESP-config refresh trigger '{trigger}' — scheduling re-collect.");
+            Task.Run(() =>
+            {
+                try { _collector.RefreshEspConfiguration(trigger); }
+                catch (Exception ex) { _logger.Warning($"DeviceInfoHost: ESP-config refresh threw: {ex.Message}"); }
+            });
+        }
+
         /// <summary>First DeviceSetup phase signal — MDM enrollment has populated the registry surface.</summary>
         internal static bool IsEnrollmentStartTrigger(DecisionSignalKind kind, IReadOnlyDictionary<string, string>? payload)
         {
@@ -156,6 +198,45 @@ namespace AutopilotMonitor.Agent.V2.Core.Orchestration
                 && string.Equals(phase, nameof(EnrollmentPhase.FinalizingSetup), StringComparison.Ordinal);
         }
 
+        /// <summary>
+        /// Audit Q2 — maps a posted signal to an ESP-config refresh trigger label, or
+        /// <c>null</c> when the signal is not one. Two trigger families:
+        /// <list type="bullet">
+        ///   <item>The adapter's apps sub-phase declaration (<c>phase_transition</c>
+        ///   InformationalEvent with phase AppsDevice/AppsUser — fired on the first app
+        ///   activity in an ESP phase): by then IME is actively processing the blocking set,
+        ///   so the device-scope ESPTrackingInfo lists are as complete as they get.</item>
+        ///   <item>The AccountSetup ESP phase (UserSetup detection): sign-in has happened, so
+        ///   the user-scope <c>S-&lt;SID&gt;</c> lists exist now — production showed 88 of 89
+        ///   sessions with an EMPTY user list at the last pre-fix emission.</item>
+        /// </list>
+        /// </summary>
+        internal static string? ClassifyEspConfigRefreshTrigger(DecisionSignalKind kind, IReadOnlyDictionary<string, string>? payload)
+        {
+            if (payload == null) return null;
+
+            if (kind == DecisionSignalKind.InformationalEvent
+                && payload.TryGetValue(SignalPayloadKeys.EventType, out var eventType)
+                && string.Equals(eventType, AutopilotMonitor.Shared.Constants.EventTypes.PhaseTransition, StringComparison.Ordinal)
+                && payload.TryGetValue(Telemetry.Events.EventTimelineEmitter.PhaseParamKey, out var declaredPhase))
+            {
+                if (string.Equals(declaredPhase, nameof(EnrollmentPhase.AppsDevice), StringComparison.Ordinal))
+                    return EspConfigTriggerAppsDevice;
+                if (string.Equals(declaredPhase, nameof(EnrollmentPhase.AppsUser), StringComparison.Ordinal))
+                    return EspConfigTriggerAppsUser;
+                return null;
+            }
+
+            if (kind == DecisionSignalKind.EspPhaseChanged
+                && payload.TryGetValue(SignalPayloadKeys.EspPhase, out var espPhase)
+                && string.Equals(espPhase, nameof(EnrollmentPhase.AccountSetup), StringComparison.Ordinal))
+            {
+                return EspConfigTriggerAccountSetup;
+            }
+
+            return null;
+        }
+
         public void Stop() => Unsubscribe();
 
         public void Dispose()

src/Agent/AutopilotMonitor.Agent.V2.Core/Security/HardwareInfo.cs

@@ -93,7 +93,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
                     {
                         using (obj)
                         {
-                            manufacturer = obj["Manufacturer"]?.ToString() ?? Unknown;
+                            manufacturer = NormalizeWmiValue(obj["Manufacturer"]?.ToString());
                             if (manufacturer.IndexOf("lenovo", StringComparison.OrdinalIgnoreCase) >= 0)
                             {
                                 using (var lenovoSearcher = new ManagementObjectSearcher("SELECT Version FROM Win32_ComputerSystemProduct"))
@@ -103,7 +103,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
                                     {
                                         using (lenovoObj)
                                         {
-                                            model = lenovoObj["Version"]?.ToString() ?? Unknown;
+                                            model = NormalizeWmiValue(lenovoObj["Version"]?.ToString());
                                         }
                                         break;
                                     }
@@ -111,7 +111,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
                             }
                             else
                             {
-                                model = obj["Model"]?.ToString() ?? Unknown;
+                                model = NormalizeWmiValue(obj["Model"]?.ToString());
                             }
                         }
                     }
@@ -132,7 +132,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
                     {
                         using (obj)
                         {
-                            serialNumber = obj["SerialNumber"]?.ToString() ?? Unknown;
+                            serialNumber = NormalizeWmiValue(obj["SerialNumber"]?.ToString());
                         }
                     }
                 }
@@ -154,5 +154,17 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
         private static bool IsResolved(string value)
             => !string.IsNullOrWhiteSpace(value)
                && !string.Equals(value, Unknown, StringComparison.OrdinalIgnoreCase);
+
+        /// <summary>
+        /// Trims a raw WMI string value at the source (audit Q5: BIOS serials are used as device
+        /// identity keys downstream — F2 device history joins on them — so padding must never
+        /// leave the agent). Whitespace-only and empty values collapse to the "Unknown" sentinel,
+        /// matching the null fallback so the retry logic treats them as unresolved.
+        /// </summary>
+        internal static string NormalizeWmiValue(string raw)
+        {
+            if (string.IsNullOrWhiteSpace(raw)) return Unknown;
+            return raw.Trim();
+        }
     }
 }

src/Agent/AutopilotMonitor.Agent.V2.Core/Security/SystemPaths.cs

@@ -35,9 +35,6 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
         /// <summary>C:\Windows\System32\tzutil.exe</summary>
         public static readonly string TzUtil = Path.Combine(System32, "tzutil.exe");
 
-        /// <summary>C:\Windows\System32\netsh.exe</summary>
-        public static readonly string Netsh = Path.Combine(System32, "netsh.exe");
-
         /// <summary>C:\Windows\System32\schtasks.exe</summary>
         public static readonly string Schtasks = Path.Combine(System32, "schtasks.exe");
 

src/Agent/AutopilotMonitor.Agent.V2.Core/SignalAdapters/ImeLogTrackerAdapter.cs

@@ -132,7 +132,9 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
 
         // Plan §5 Fix 4c — per-app install-lifecycle timing captured from state transitions.
         // StartedAtUtc on first Downloading/Installing/InProgress transition (set-once);
-        // CompletedAtUtc on first terminal transition (Installed/Skipped/Postponed/Error).
+        // CompletedAtUtc on terminal transition (Installed/Skipped/Postponed/Error), re-armed
+        // to null on a terminal→active transition (IME retry, audit Q1) so the final terminal
+        // stamp reflects the last outcome instead of freezing at the first one.
         // Surfaced via <see cref="AppTimings"/> to peripheral consumers (FinalStatusBuilder,
         // app_tracking_summary emission in EnrollmentTerminationHandler).
         private readonly Dictionary<string, (DateTime? Started, DateTime? Completed)> _appTimings =
@@ -691,10 +693,20 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
         }
 
         /// <summary>
-        /// Plan §5 Fix 4c — set-once per-app install-lifecycle timing. Called for every
-        /// state transition, no-op for duplicate stamps. Returns the up-to-date
+        /// Plan §5 Fix 4c — per-app install-lifecycle timing. Called for every state
+        /// transition, no-op for duplicate stamps. Returns the up-to-date
         /// <see cref="AppInstallTiming"/> so the emitter can inline the values on the
         /// current event's payload without a second dictionary lookup.
+        /// <para>
+        /// Audit Q1: a terminal→active transition (the IME retry path — <see
+        /// cref="AppPackageState.UpdateState"/> permits e.g. Error→Downloading and
+        /// Skipped→Downloading) re-arms <c>Completed</c> to <c>null</c>, so the NEXT terminal
+        /// transition stamps a fresh CompletedAt. StartedAt keeps its first value — the final
+        /// terminal payload then covers first-start → last-terminal (total occupancy across
+        /// attempts) instead of freezing at the first failure. Observed pre-fix: a user-scope
+        /// app Skipped early in DeviceSetup, retried after sign-in, emitted its final
+        /// Installed event with the stale early CompletedAt (durationSeconds=-2342).
+        /// </para>
         /// </summary>
         private AppInstallTiming UpdateAppTiming(string appId, AppInstallationState newState, DateTime now)
         {
@@ -702,13 +714,20 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
             {
                 _appTimings.TryGetValue(appId, out var current);
 
+                // Terminal→active (retry): the earlier terminal stamp no longer describes this
+                // app's outcome — re-arm so the next terminal transition stamps fresh.
+                if (current.Completed != null && IsLifecycleActive(newState))
+                {
+                    current.Completed = null;
+                }
+
                 // First lifecycle-active transition → record StartedAt.
                 if (current.Started == null && IsLifecycleActive(newState))
                 {
                     current.Started = now;
                 }
 
-                // First terminal transition → record CompletedAt.
+                // Terminal transition → record CompletedAt (fresh after a re-arm).
                 if (current.Completed == null && IsCompletedState(newState))
                 {
                     current.Completed = now;

src/Agent/AutopilotMonitor.Agent.V2.Core/Termination/EnrollmentTerminationHandler.cs

@@ -623,12 +623,15 @@ namespace AutopilotMonitor.Agent.V2.Core.Termination
         }
 
         /// <summary>
-        /// Emits a one-shot <c>agent_late_start</c> Warning when the agent had low observation
+        /// Emits a one-shot <c>agent_late_start</c> event when the agent had low observation
         /// coverage: it started a long time after device boot AND lived only briefly before this
         /// terminal outcome (gate in <see cref="ObservationCoverage"/>). Such a session looks like a
         /// normal multi-minute failure in the list (StartedAt is back-dated to the earliest replayed
-        /// IME-log event) but the agent actually observed only the end-state — so the operator should
-        /// read the diagnosis as a post-mortem reconstruction, not live evidence. No state mutation.
+        /// IME-log event) but the agent actually observed only the end-state. Severity and phrasing
+        /// are outcome-calibrated via <see cref="ObservationCoverage.DescribeLateStart"/>: a failure
+        /// stays a Warning with the post-mortem/hung-script note, a success is an Info about
+        /// tail-only coverage (fleet 2026-07-28: tenant 659c3a90 collected 12 Warnings on healthy
+        /// enrollments whose bootstrap merely ran late). No state mutation.
         /// </summary>
         private void MaybeEmitAgentLateStart(EnrollmentTerminatedEventArgs args)
         {
@@ -642,21 +645,24 @@ namespace AutopilotMonitor.Agent.V2.Core.Termination
                         out var bootToStartSeconds, out var uptimeSeconds))
                     return;
 
+                ObservationCoverage.DescribeLateStart(
+                    args.Outcome, _oobeStateAtAgentStart, bootToStartSeconds, uptimeSeconds,
+                    out var severity, out var message, out var note);
+
                 _logger.Info(
                     $"EnrollmentTerminationHandler: low observation coverage — agent started " +
                     $"{bootToStartSeconds / 60.0:F1} min after boot and lived {uptimeSeconds:F0}s; " +
-                    $"emitting agent_late_start.");
+                    $"emitting agent_late_start ({severity}, outcome {args.Outcome}).");
 
                 _post.Emit(new EnrollmentEvent
                 {
                     SessionId = _configuration.SessionId,
                     TenantId = _configuration.TenantId,
                     EventType = Constants.EventTypes.AgentLateStart,
-                    Severity = EventSeverity.Warning,
+                    Severity = severity,
                     Source = "EnrollmentTerminationHandler",
                     Phase = EnrollmentPhase.Unknown,
-                    Message = $"Agent started {bootToStartSeconds / 60.0:F0} min after boot and observed only " +
-                              $"{uptimeSeconds:F0}s before terminating ({args.Outcome}) — low coverage of the enrollment window.",
+                    Message = message,
                     Data = new Dictionary<string, object>
                     {
                         { "bootToAgentStartSeconds", Math.Round(bootToStartSeconds, 1) },
@@ -665,7 +671,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Termination
                         { "agentStartUtc", _agentStartTimeUtc.ToString("o") },
                         { "oobeStateAtAgentStart", _oobeStateAtAgentStart },
                         { "outcome", args.Outcome.ToString() },
-                        { "note", "The agent's bootstrap (an Intune platform script) ran only near the end of the enrollment, so the agent observed the already-decided end-state rather than the failure window. Treat the diagnosis as a post-mortem; check for a platform/remediation script that hung ahead of the bootstrap (see script_timeout_suspected)." },
+                        { "note", note },
                     },
                     ImmediateUpload = true,
                 });

src/Agent/AutopilotMonitor.Agent.V2.Core/Termination/ObservationCoverage.cs

@@ -1,6 +1,8 @@
 #nullable enable
 using System;
 using System.Runtime.InteropServices;
+using AutopilotMonitor.Agent.V2.Core.Orchestration;
+using AutopilotMonitor.Shared.Models;
 
 namespace AutopilotMonitor.Agent.V2.Core.Termination
 {
@@ -76,5 +78,57 @@ namespace AutopilotMonitor.Agent.V2.Core.Termination
             uptimeSeconds = Math.Max(0, uptime.TotalSeconds);
             return bootToStart >= LateStartBootThreshold && uptime <= LowCoverageUptime;
         }
+
+        /// <summary>
+        /// Severity + operator-facing phrasing for the <c>agent_late_start</c> event, calibrated
+        /// by fleet data (2026-07-28): the gate alone is outcome-blind and attached a
+        /// "post-mortem / check for a hung script" Warning to sessions that in fact succeeded —
+        /// there the late bootstrap is deployment hygiene, not an alarm. The OOBE state refines
+        /// the phrasing but must never gate: the canonical hung-script failure terminates on the
+        /// still-shown ESP screen, i.e. with OOBE reading <c>in_progress</c>.
+        /// </summary>
+        /// <param name="outcome">Terminal outcome the enrollment reached.</param>
+        /// <param name="oobeStateAtAgentStart"><see cref="Monitoring.Interop.OobeStateReader"/>
+        /// sample taken at THIS process's start. Only <c>completed</c> / <c>in_progress</c> are
+        /// interpreted; <c>not_started</c> flips in after mid-OOBE reboots and <c>completed</c> is
+        /// routine for post-desktop restarts, so no claim is phrased from the other values.</param>
+        /// <param name="bootToStartSeconds">From <see cref="IsLowObservationCoverage"/>.</param>
+        /// <param name="uptimeSeconds">From <see cref="IsLowObservationCoverage"/>.</param>
+        internal static void DescribeLateStart(
+            EnrollmentTerminationOutcome outcome,
+            string oobeStateAtAgentStart,
+            double bootToStartSeconds,
+            double uptimeSeconds,
+            out EventSeverity severity,
+            out string message,
+            out string note)
+        {
+            var succeeded = outcome == EnrollmentTerminationOutcome.Succeeded;
+            severity = succeeded ? EventSeverity.Info : EventSeverity.Warning;
+
+            message = $"Agent started {bootToStartSeconds / 60.0:F0} min after boot and observed only " +
+                      $"{uptimeSeconds:F0}s before terminating ({outcome}) — low coverage of the enrollment window.";
+            switch (oobeStateAtAgentStart)
+            {
+                case "completed":
+                    message += " OOBE was already completed when the agent started — the diagnosis is a " +
+                               "post-mortem reconstruction, not live observation.";
+                    break;
+                case "in_progress":
+                    message += $" OOBE was still in progress at agent start — the agent did observe the " +
+                               $"final {uptimeSeconds:F0}s live.";
+                    break;
+            }
+
+            note = succeeded
+                ? "The agent's bootstrap (an Intune platform script) ran late in the enrollment, so only the " +
+                  "final stretch was observed live; earlier timeline entries are reconstructed from IME log " +
+                  "replay. The enrollment itself succeeded. If this recurs across sessions, review which " +
+                  "platform/remediation scripts run ahead of the bootstrap (see script_timeout_suspected)."
+                : "The agent's bootstrap (an Intune platform script) ran only near the end of the enrollment, " +
+                  "so the agent observed the already-decided end-state rather than the failure window. Treat " +
+                  "the diagnosis as a post-mortem; check for a platform/remediation script that hung ahead of " +
+                  "the bootstrap (see script_timeout_suspected).";
+        }
     }
 }

src/Backend/AutopilotMonitor.Functions.Tests/AppInstallIdentityAndEspBlockingTests.cs

@@ -0,0 +1,333 @@
+using System.Text.Json;
+using Azure.Data.Tables;
+using AutopilotMonitor.Functions.Functions.Apps;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.DataAccess;
+using AutopilotMonitor.Shared.Models;
+using Moq;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// F1 PR1 (insights spec §F1, source-data audit Q2/Q3): app identity on the name-keyed
+/// summary row and the ESP-blocking join.
+///
+/// Ingest side: <c>appId</c> is adopted from the app events; a second appId under the same
+/// name flags the row (<see cref="AppInstallSummary.AppIdCollision"/>) — in-batch and
+/// cross-batch. Store side: sentinel-gated columns + sticky collision. Join side:
+/// <see cref="EspBlockingSets"/> is positive evidence only, and
+/// <see cref="TableStorageService.ShouldStampEspBlocking"/> never stamps collision rows,
+/// id-less rows, or non-members. Aggregate side: collision rows leave every per-app fleet
+/// aggregate with a disclosed exclusion count.
+/// </summary>
+public class AppInstallIdentityAndEspBlockingTests
+{
+    private static readonly DateTime T0 = new(2026, 7, 26, 10, 0, 0, DateTimeKind.Utc);
+
+    private const string IdA = "aaaaaaaa-1111-2222-3333-444444444444";
+    private const string IdB = "bbbbbbbb-1111-2222-3333-444444444444";
+
+    // ── ingest fold: appId adoption + in-batch collision ────────────────────
+
+    private static Dictionary<string, AppInstallAggregationState> Aggregate(params EnrollmentEvent[] events)
+    {
+        var summaries = new Dictionary<string, AppInstallAggregationState>(StringComparer.OrdinalIgnoreCase);
+        foreach (var evt in events)
+            EventIngestProcessor.AggregateAppInstallEvent(evt, "tenant", "session", summaries);
+        return summaries;
+    }
+
+    private static EnrollmentEvent AppEvent(string eventType, string? appId = null, string appName = "App")
+    {
+        var evt = new EnrollmentEvent
+        {
+            EventType = eventType,
+            Timestamp = T0,
+            Data = new Dictionary<string, object> { ["appName"] = appName },
+        };
+        if (appId != null) evt.Data["appId"] = appId;
+        return evt;
+    }
+
+    [Fact]
+    public void Ingest_AdoptsAppId_FromFirstObservingEvent()
+    {
+        var s = Aggregate(
+            AppEvent("app_install_started", IdA),
+            AppEvent("app_install_completed", IdA))["App"].Summary;
+
+        Assert.Equal(IdA, s.AppId);
+        Assert.False(s.AppIdCollision);
+    }
+
+    [Fact]
+    public void Ingest_SameName_DifferentAppId_FlagsCollision_FirstSeenIdWins()
+    {
+        // Device- + user-scope assignment of the same display name in one batch (audit Q3:
+        // observed in production as two "Company Portal" appIds in one session).
+        var s = Aggregate(
+            AppEvent("app_install_started", IdA),
+            AppEvent("app_install_started", IdB))["App"].Summary;
+
+        Assert.True(s.AppIdCollision);
+        Assert.Equal(IdA, s.AppId);
+    }
+
+    [Fact]
+    public void Ingest_AppIdCaseDrift_IsNotACollision()
+    {
+        var s = Aggregate(
+            AppEvent("app_install_started", IdA),
+            AppEvent("app_install_completed", IdA.ToUpperInvariant()))["App"].Summary;
+
+        Assert.False(s.AppIdCollision);
+    }
+
+    [Fact]
+    public void Ingest_EventsWithoutAppId_LeaveSentinelEmpty()
+    {
+        var s = Aggregate(AppEvent("app_install_completed"))["App"].Summary;
+        Assert.Equal(string.Empty, s.AppId);
+        Assert.False(s.AppIdCollision);
+    }
+
+    // ── reconcile: cross-batch identity ─────────────────────────────────────
+
+    [Fact]
+    public void Reconcile_AdoptsStoredAppId_WhenIncomingBatchHasNone()
+    {
+        var existing = new TableEntity("tenant", "session_App") { ["AppId"] = IdA };
+        var summary = new AppInstallSummary { AppName = "App", StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.Equal(IdA, summary.AppId);
+        Assert.False(summary.AppIdCollision);
+    }
+
+    [Fact]
+    public void Reconcile_CrossBatchDifferentAppId_FlagsCollision_StoredIdWins()
+    {
+        var existing = new TableEntity("tenant", "session_App") { ["AppId"] = IdA };
+        var summary = new AppInstallSummary { AppName = "App", AppId = IdB, StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.True(summary.AppIdCollision);
+        Assert.Equal(IdA, summary.AppId); // first-seen (stored) identity wins, matching the fold rule
+    }
+
+    [Fact]
+    public void Reconcile_CollisionIsSticky_AndEspBlockingVerdictIsAdopted()
+    {
+        var existing = new TableEntity("tenant", "session_App")
+        {
+            ["AppId"] = IdA,
+            ["AppIdCollision"] = true,
+            ["EspBlocking"] = true,
+        };
+        var summary = new AppInstallSummary { AppName = "App", AppId = IdA, StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.True(summary.AppIdCollision);       // one-way sticky
+        Assert.True(summary.EspBlocking);          // terminal-path verdict survives late batches
+    }
+
+    // ── entity builder: sentinel gating ─────────────────────────────────────
+
+    [Fact]
+    public void EntityBuilder_IdentityColumns_AreSentinelGated()
+    {
+        var empty = TableStorageService.BuildAppInstallSummaryEntity(
+            new AppInstallSummary { AppName = "App" }, "rk");
+        // Absent columns → Merge-mode preserves prior values / stays unknown.
+        Assert.False(empty.ContainsKey("AppId"));
+        Assert.False(empty.ContainsKey("EspBlocking"));
+        Assert.False(empty.ContainsKey("AppIdCollision"));
+
+        var full = TableStorageService.BuildAppInstallSummaryEntity(
+            new AppInstallSummary { AppName = "App", AppId = IdA, EspBlocking = true, AppIdCollision = true }, "rk");
+        Assert.Equal(IdA, full.GetString("AppId"));
+        Assert.True(full.GetBoolean("EspBlocking"));
+        Assert.True(full.GetBoolean("AppIdCollision"));
+    }
+
+    // ── EspBlockingSets: positive-evidence parsing ──────────────────────────
+
+    private static Dictionary<string, object> Lists(
+        object? win32 = null, object? userWin32 = null, object? msi = null, object? pfn = null,
+        long? win32Count = null)
+    {
+        var d = new Dictionary<string, object> { ["source"] = "registry_firstsync" };
+        if (win32 != null) d["espTrackedWin32AppIds"] = win32;
+        if (userWin32 != null) d["espTrackedUserWin32AppIds"] = userWin32;
+        if (msi != null) d["espTrackedMsiProductCodes"] = msi;
+        if (pfn != null) d["espTrackedModernAppPfns"] = pfn;
+        if (win32Count.HasValue) d["espTrackedWin32Count"] = win32Count.Value;
+        return d;
+    }
+
+    [Fact]
+    public void Sets_PayloadWithoutListKeys_YieldsNull_NoListsIsNotEmptyLists()
+    {
+        Assert.Null(EspBlockingSets.FromEventData(null));
+        Assert.Null(EspBlockingSets.FromEventData(Lists())); // probe found no Diagnostics key
+    }
+
+    [Fact]
+    public void Sets_MembershipIsCaseInsensitive_AcrossAllFourLists()
+    {
+        // Storage rehydration shape: List<object> (EventDataNormalizer).
+        var sets = EspBlockingSets.FromEventData(Lists(
+            win32: new List<object> { IdA },
+            msi: new List<object> { "{11111111-2222-3333-4444-555555555555}" },
+            pfn: new List<object> { "Contoso.App_abc123" }))!;
+
+        Assert.True(sets.Contains(IdA.ToUpperInvariant()));
+        Assert.True(sets.Contains("{11111111-2222-3333-4444-555555555555}"));
+        Assert.True(sets.Contains("Contoso.App_abc123"));
+        Assert.False(sets.Contains(IdB));           // absent ⇒ unknown at the caller, never false here
+        Assert.False(sets.Contains(null));
+        Assert.Equal(3, sets.ListedCount);
+        Assert.False(sets.IsTruncated);
+    }
+
+    [Fact]
+    public void Sets_EmittedCountAboveListLength_MeansTruncated()
+    {
+        var sets = EspBlockingSets.FromEventData(Lists(
+            win32: new List<object> { IdA }, win32Count: 60))!;
+
+        Assert.True(sets.IsTruncated);
+    }
+
+    [Fact]
+    public void Sets_UserScopeSubset_IsExposedForCoverageLabeling()
+    {
+        var sets = EspBlockingSets.FromEventData(Lists(
+            win32: new List<object> { IdA, IdB },
+            userWin32: new List<object> { IdB }))!;
+
+        Assert.Contains(IdB, sets.UserWin32AppIds);
+        Assert.DoesNotContain(IdA, sets.UserWin32AppIds);
+    }
+
+    // ── terminal-path stamping predicate ────────────────────────────────────
+
+    private static TableEntity Row(string? appId = null, bool? collision = null, bool? espBlocking = null)
+    {
+        var row = new TableEntity("tenant", "session_App");
+        if (appId != null) row["AppId"] = appId;
+        if (collision.HasValue) row["AppIdCollision"] = collision.Value;
+        if (espBlocking.HasValue) row["EspBlocking"] = espBlocking.Value;
+        return row;
+    }
+
+    [Fact]
+    public void Stamp_OnlyPositiveMembers_NeverIdlessCollidedOrAlreadyStamped()
+    {
+        var sets = EspBlockingSets.FromEventData(Lists(win32: new List<object> { IdA }))!;
+
+        Assert.True(TableStorageService.ShouldStampEspBlocking(Row(IdA), sets));
+        Assert.False(TableStorageService.ShouldStampEspBlocking(Row(), sets));                    // no identity → unknown
+        Assert.False(TableStorageService.ShouldStampEspBlocking(Row(IdB), sets));                 // absent ⇒ unknown, never false
+        Assert.False(TableStorageService.ShouldStampEspBlocking(Row(IdA, collision: true), sets)); // ambiguous identity
+        Assert.False(TableStorageService.ShouldStampEspBlocking(Row(IdA, espBlocking: true), sets)); // idempotent
+    }
+
+    // ── per-app fleet aggregates: collision rows leave, disclosed ───────────
+
+    private static AppInstallSummary Summary(
+        string appName, string status = "Succeeded", bool collision = false, int duration = 30)
+        => new()
+        {
+            TenantId = "00000000-0000-0000-0000-000000000fa1",
+            SessionId = "s-1",
+            AppName = appName,
+            Status = status,
+            TerminalState = status == "Succeeded" ? "Installed" : "Error",
+            DurationSeconds = duration,
+            StartedAt = DateTime.UtcNow.AddHours(-2),
+            AppIdCollision = collision,
+        };
+
+    [Fact]
+    public void AppMetricsPayload_ExcludesCollisionRowsFromGroups_DisclosesCount()
+    {
+        var summaries = new List<AppInstallSummary>
+        {
+            Summary("Clean App"),
+            Summary("Clean App"),
+            Summary("Company Portal", collision: true),
+            Summary("Company Portal", status: "Failed", collision: true),
+        };
+
+        var root = JsonSerializer.SerializeToElement(MetricsMath.BuildAppMetricsPayload(summaries));
+
+        Assert.Equal(1, root.GetProperty("totalApps").GetInt32());          // collided name forms no group
+        Assert.Equal(4, root.GetProperty("totalInstalls").GetInt32());      // raw row count unchanged
+        Assert.Equal(2, root.GetProperty("totalCollisionExcluded").GetInt32());
+        Assert.Empty(root.GetProperty("topFailingApps").EnumerateArray());  // the collided Failed row ranks nowhere
+    }
+
+    [Fact]
+    public void AppsList_ExcludesCollisionRows_DisclosesCount()
+    {
+        var summaries = new List<AppInstallSummary>
+        {
+            Summary("Clean App"),
+            Summary("Company Portal", collision: true),
+        };
+
+        var root = JsonSerializer.SerializeToElement(
+            AppsAnalyticsHelper.BuildAppsListResponse(summaries, days: 30));
+
+        var apps = root.GetProperty("apps").EnumerateArray().ToList();
+        Assert.Single(apps);
+        Assert.Equal("Clean App", apps[0].GetProperty("appName").GetString());
+        Assert.Equal(2, root.GetProperty("totalInstalls").GetInt32());
+        Assert.Equal(1, root.GetProperty("collisionExcluded").GetInt32());
+    }
+
+    [Fact]
+    public async Task AppAnalytics_ExcludesCollisionRows_DisclosesCount()
+    {
+        var repo = new Mock<ISessionRepository>();
+        repo.Setup(r => r.GetSessionAsync(It.IsAny<string>(), It.IsAny<string>()))
+            .ReturnsAsync((SessionSummary?)null);
+
+        var summaries = new List<AppInstallSummary>
+        {
+            Summary("Company Portal"),
+            Summary("Company Portal", status: "Failed", collision: true), // the second identity's failure
+        };
+
+        var root = JsonSerializer.SerializeToElement(await AppsAnalyticsHelper.BuildAnalyticsResponseAsync(
+            summaries, repo.Object, "Company Portal", days: 30));
+
+        Assert.Equal(1, root.GetProperty("collisionExcluded").GetInt32());
+        var stats = root.GetProperty("summary");
+        Assert.Equal(1, stats.GetProperty("totalInstalls").GetInt32());
+        Assert.Equal(0, stats.GetProperty("failed").GetInt32());            // ambiguous failure not attributed
+        Assert.Equal(0d, stats.GetProperty("failureRate").GetDouble());
+    }
+
+    [Fact]
+    public async Task AppAnalytics_AllRowsCollided_ReturnsEmptyShape_WithDisclosure()
+    {
+        var repo = new Mock<ISessionRepository>();
+        repo.Setup(r => r.GetSessionAsync(It.IsAny<string>(), It.IsAny<string>()))
+            .ReturnsAsync((SessionSummary?)null);
+
+        var summaries = new List<AppInstallSummary> { Summary("Company Portal", collision: true) };
+
+        var root = JsonSerializer.SerializeToElement(await AppsAnalyticsHelper.BuildAnalyticsResponseAsync(
+            summaries, repo.Object, "Company Portal", days: 30));
+
+        Assert.Equal(1, root.GetProperty("collisionExcluded").GetInt32());
+        Assert.Equal(0, root.GetProperty("summary").GetProperty("totalInstalls").GetInt32());
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/AppInstallTerminalStateAndReconcileTests.cs

@@ -0,0 +1,201 @@
+using Azure.Data.Tables;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// PR0 (2026-07-26, source-data audit §0.5 of tasks/insights-expansion-spec.md).
+///
+/// Ingest side: the agent emits <c>app_install_completed</c> for EVERY terminal transition —
+/// including Skipped/Postponed no-ops (V1 wire parity; production: 78 % of zero-duration rows
+/// were skips). The payload's <c>state</c> field must be persisted as
+/// <see cref="AppInstallSummary.TerminalState"/> so metrics can exclude skips from durations
+/// and rates.
+///
+/// Store side: <see cref="TableStorageService.ReconcileAppInstallSummaryWithExisting"/> pins the
+/// cross-batch contract — earliest StartedAt wins, terminal status is sticky, and the Q4
+/// out-of-order case (terminal batch first, started batch later) recomputes the duration.
+/// </summary>
+public class AppInstallTerminalStateAndReconcileTests
+{
+    private static readonly DateTime T0 = new(2026, 7, 26, 10, 0, 0, DateTimeKind.Utc);
+
+    // ── ingest: TerminalState fold ──────────────────────────────────────────
+
+    private static Dictionary<string, AppInstallAggregationState> Aggregate(params EnrollmentEvent[] events)
+    {
+        var summaries = new Dictionary<string, AppInstallAggregationState>(StringComparer.OrdinalIgnoreCase);
+        foreach (var evt in events)
+            EventIngestProcessor.AggregateAppInstallEvent(evt, "tenant", "session", summaries);
+        return summaries;
+    }
+
+    private static EnrollmentEvent AppEvent(string eventType, DateTime ts, string? state = null, string appName = "App")
+    {
+        var evt = new EnrollmentEvent
+        {
+            EventType = eventType,
+            Timestamp = ts,
+            Data = new Dictionary<string, object> { ["appName"] = appName },
+        };
+        if (state != null) evt.Data["state"] = state;
+        return evt;
+    }
+
+    [Theory]
+    [InlineData("Installed")]
+    [InlineData("Skipped")]
+    [InlineData("Postponed")]
+    public void Completed_PersistsPayloadState(string state)
+    {
+        var s = Aggregate(AppEvent("app_install_completed", T0, state))["App"].Summary;
+
+        Assert.Equal("Succeeded", s.Status);          // backward-compatible status
+        Assert.Equal(state, s.TerminalState);          // new precision column
+    }
+
+    [Fact]
+    public void Completed_WithoutState_LeavesSentinelEmpty_NeverGuessed()
+    {
+        var s = Aggregate(AppEvent("app_install_completed", T0))["App"].Summary;
+        Assert.Equal(string.Empty, s.TerminalState);
+    }
+
+    [Fact]
+    public void Completed_WithNonTerminalOrGarbageState_LeavesSentinelEmpty()
+    {
+        // A payload state outside the terminal set (bug or drift) must not be persisted.
+        var s = Aggregate(AppEvent("app_install_completed", T0, "Downloading"))["App"].Summary;
+        Assert.Equal(string.Empty, s.TerminalState);
+    }
+
+    [Fact]
+    public void Failed_SetsError()
+    {
+        var s = Aggregate(AppEvent("app_install_failed", T0))["App"].Summary;
+        Assert.Equal("Failed", s.Status);
+        Assert.Equal("Error", s.TerminalState);
+    }
+
+    [Fact]
+    public void FailedThenCompleted_RetrySucceeds_LastTerminalWins()
+    {
+        // IME retry path (audit Q1): Error → retry → Installed. The later completed event is
+        // the authoritative outcome for Status AND TerminalState.
+        var s = Aggregate(
+            AppEvent("app_install_failed", T0),
+            AppEvent("app_install_completed", T0.AddMinutes(10), "Installed"))["App"].Summary;
+
+        Assert.Equal("Succeeded", s.Status);
+        Assert.Equal("Installed", s.TerminalState);
+    }
+
+    [Fact]
+    public void SkippedEvent_SetsSkipped_ButNeverOverridesStrongerTerminal()
+    {
+        var fresh = Aggregate(AppEvent("app_install_skipped", T0))["App"].Summary;
+        Assert.Equal("Succeeded", fresh.Status);
+        Assert.Equal("Skipped", fresh.TerminalState);
+
+        var afterFailure = Aggregate(
+            AppEvent("app_install_failed", T0),
+            AppEvent("app_install_skipped", T0.AddSeconds(5)))["App"].Summary;
+        Assert.Equal("Error", afterFailure.TerminalState);
+    }
+
+    // ── store: cross-batch reconcile ────────────────────────────────────────
+
+    [Fact]
+    public void Reconcile_Q4_OutOfOrder_RecomputesDurationFromStoredCompletedAt()
+    {
+        // Terminal batch landed first: row has CompletedAt but no usable duration.
+        var existing = new TableEntity("tenant", "session_App")
+        {
+            ["Status"] = "Succeeded",
+            ["StartedAt"] = new DateTimeOffset(T0.AddMinutes(5)),   // late (completed-event) stamp
+            ["CompletedAt"] = new DateTimeOffset(T0.AddMinutes(5)),
+        };
+        // Started batch arrives now with the true (earlier) start.
+        var summary = new AppInstallSummary
+        {
+            AppName = "App", SessionId = "session", TenantId = "tenant",
+            Status = "InProgress", StartedAt = T0,
+        };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.Equal(T0, summary.StartedAt);                        // earlier start kept
+        Assert.Equal(T0.AddMinutes(5), summary.CompletedAt);        // endpoint adopted from the row
+        Assert.Equal(300, summary.DurationSeconds);                 // recomputed — was permanently unset before PR0
+        Assert.Equal("Succeeded", summary.Status);                  // sticky terminal (below)
+    }
+
+    [Fact]
+    public void Reconcile_TerminalStatusIsSticky_AgainstLateInProgressBatch()
+    {
+        var existing = new TableEntity("tenant", "session_App") { ["Status"] = "Failed" };
+        var summary = new AppInstallSummary { AppName = "App", Status = "InProgress", StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.Equal("Failed", summary.Status);
+    }
+
+    [Fact]
+    public void Reconcile_TerminalToTerminal_StaysWithIncomingEvidence()
+    {
+        // Failed row + succeeded batch (successful retry) → the new terminal wins.
+        var existing = new TableEntity("tenant", "session_App") { ["Status"] = "Failed" };
+        var summary = new AppInstallSummary { AppName = "App", Status = "Succeeded", StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.Equal("Succeeded", summary.Status);
+    }
+
+    [Fact]
+    public void Reconcile_Q4Guard_DoesNotFireWithoutStoredCompletedAt()
+    {
+        var existing = new TableEntity("tenant", "session_App") { ["Status"] = "InProgress" };
+        var summary = new AppInstallSummary { AppName = "App", Status = "InProgress", StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.Null(summary.CompletedAt);
+        Assert.Equal(0, summary.DurationSeconds);
+    }
+
+    [Fact]
+    public void Reconcile_Q4Guard_IgnoresStoredCompletedAtBeforeIncomingStart()
+    {
+        // Negative interval (clock mess) must not produce a negative duration.
+        var existing = new TableEntity("tenant", "session_App")
+        {
+            ["Status"] = "Succeeded",
+            ["CompletedAt"] = new DateTimeOffset(T0.AddMinutes(-10)),
+        };
+        var summary = new AppInstallSummary { AppName = "App", Status = "InProgress", StartedAt = T0 };
+
+        TableStorageService.ReconcileAppInstallSummaryWithExisting(summary, existing);
+
+        Assert.Null(summary.CompletedAt);
+        Assert.Equal(0, summary.DurationSeconds);
+    }
+
+    // ── entity builder: sentinel gating for the new column ─────────────────
+
+    [Fact]
+    public void EntityBuilder_TerminalState_IsSentinelGated()
+    {
+        var withState = new AppInstallSummary { AppName = "App", TerminalState = "Skipped" };
+        var entity = TableStorageService.BuildAppInstallSummaryEntity(withState, "rk");
+        Assert.Equal("Skipped", entity.GetString("TerminalState"));
+
+        var withoutState = new AppInstallSummary { AppName = "App" };
+        var empty = TableStorageService.BuildAppInstallSummaryEntity(withoutState, "rk");
+        // Column absent → Merge-mode preserves a prior batch's terminal state.
+        Assert.False(empty.ContainsKey("TerminalState"));
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/AppsAnalyticsFailureRateTests.cs

@@ -0,0 +1,247 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text.Json;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.Functions.Apps;
+using AutopilotMonitor.Shared.DataAccess;
+using AutopilotMonitor.Shared.Models;
+using Moq;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Pins the terminal-only failure-rate convention on the Apps dashboard aggregations
+/// (<see cref="AppsAnalyticsHelper"/>): every failureRate is Failed / (Failed + Succeeded) —
+/// finished installs only. "InProgress" rows (still installing, or orphaned by a session that
+/// died mid-install) sit in the install counts but never dilute a rate, matching the enrollment
+/// success-rate convention. Responses are anonymous projections, so tests assert on the
+/// JSON wire shape like the sibling suites.
+/// </summary>
+public class AppsAnalyticsFailureRateTests
+{
+    private const string TenantId = "00000000-0000-0000-0000-000000000fa1";
+
+    private static AppInstallSummary A(
+        string status,
+        string appName = "Contoso App",
+        DateTime? startedAt = null,
+        string sessionId = "s-1",
+        string appVersion = "")
+        => new()
+        {
+            TenantId = TenantId,
+            SessionId = sessionId,
+            AppName = appName,
+            AppVersion = appVersion,
+            Status = status,
+            StartedAt = startedAt ?? DateTime.UtcNow.AddHours(-2),
+            DurationSeconds = status == "Succeeded" ? 30 : 0,
+        };
+
+    // ── /apps/list ──────────────────────────────────────────────────────────
+
+    [Fact]
+    public void List_FailureRate_IsOverFinishedInstalls_InProgressExcluded()
+    {
+        var summaries = new List<AppInstallSummary>
+        {
+            A("Failed"),
+            A("Succeeded"),
+            A("InProgress"),
+            A("InProgress"),
+        };
+
+        var root = JsonSerializer.SerializeToElement(
+            AppsAnalyticsHelper.BuildAppsListResponse(summaries, days: 30));
+
+        var app = root.GetProperty("apps")[0];
+        Assert.Equal(4, app.GetProperty("totalInstalls").GetInt32());
+        // 1 / (1 + 1) = 50% (over finished), NOT 1 / 4 = 25% (over all installs).
+        Assert.Equal(50d, app.GetProperty("failureRate").GetDouble());
+    }
+
+    [Fact]
+    public void List_FailureRate_IsZero_WhenNothingFinishedYet()
+    {
+        var summaries = new List<AppInstallSummary> { A("InProgress"), A("InProgress") };
+
+        var root = JsonSerializer.SerializeToElement(
+            AppsAnalyticsHelper.BuildAppsListResponse(summaries, days: 30));
+
+        Assert.Equal(0d, root.GetProperty("apps")[0].GetProperty("failureRate").GetDouble());
+    }
+
+    [Fact]
+    public void List_Trend_UsesFinishedRates_AndGatesOnFinishedSamples()
+    {
+        var now = DateTime.UtcNow;
+        var firstHalf = now.AddDays(-20);  // days: 30 → midpoint at now - 15d
+        var secondHalf = now.AddDays(-5);
+
+        var summaries = new List<AppInstallSummary>();
+        // First half: 1 failed / 5 finished = 20%.
+        summaries.Add(A("Failed", startedAt: firstHalf));
+        for (var i = 0; i < 4; i++) summaries.Add(A("Succeeded", startedAt: firstHalf));
+        // Second half: 3 failed / 5 finished = 60% — plus 10 in-flight installs that
+        // previously dragged the half's rate down to 20% and masked the regression.
+        for (var i = 0; i < 3; i++) summaries.Add(A("Failed", startedAt: secondHalf));
+        for (var i = 0; i < 2; i++) summaries.Add(A("Succeeded", startedAt: secondHalf));
+        for (var i = 0; i < 10; i++) summaries.Add(A("InProgress", startedAt: secondHalf));
+
+        var root = JsonSerializer.SerializeToElement(
+            AppsAnalyticsHelper.BuildAppsListResponse(summaries, days: 30));
+
+        var app = root.GetProperty("apps")[0];
+        Assert.Equal("worsening", app.GetProperty("trend").GetString());
+        Assert.Equal(40d, app.GetProperty("trendDelta").GetDouble());
+    }
+
+    [Fact]
+    public void List_Trend_StaysStable_WhenAHalfLacksFiveFinishedInstalls()
+    {
+        var now = DateTime.UtcNow;
+        var summaries = new List<AppInstallSummary>();
+        // First half: 5 finished. Second half: 4 finished + many in-flight — the old
+        // install-count gate would have passed; the finished-sample gate must not.
+        for (var i = 0; i < 5; i++) summaries.Add(A("Succeeded", startedAt: now.AddDays(-20)));
+        for (var i = 0; i < 4; i++) summaries.Add(A("Failed", startedAt: now.AddDays(-5)));
+        for (var i = 0; i < 10; i++) summaries.Add(A("InProgress", startedAt: now.AddDays(-5)));
+
+        var root = JsonSerializer.SerializeToElement(
+            AppsAnalyticsHelper.BuildAppsListResponse(summaries, days: 30));
+
+        var app = root.GetProperty("apps")[0];
+        Assert.Equal("stable", app.GetProperty("trend").GetString());
+        Assert.Equal(JsonValueKind.Null, app.GetProperty("trendDelta").ValueKind);
+    }
+
+    // ── /apps/{appName}/analytics ───────────────────────────────────────────
+
+    private static async Task<JsonElement> BuildAnalyticsAsync(
+        List<AppInstallSummary> summaries, ISessionRepository? sessionRepo = null)
+    {
+        if (sessionRepo == null)
+        {
+            var mock = new Mock<ISessionRepository>();
+            mock.Setup(r => r.GetSessionAsync(It.IsAny<string>(), It.IsAny<string>()))
+                .ReturnsAsync((SessionSummary?)null);
+            sessionRepo = mock.Object;
+        }
+        var result = await AppsAnalyticsHelper.BuildAnalyticsResponseAsync(
+            summaries, sessionRepo, "Contoso App", days: 30);
+        return JsonSerializer.SerializeToElement(result);
+    }
+
+    [Fact]
+    public async Task Analytics_SummaryVersionAndTimeSeriesRates_AreOverFinishedInstalls()
+    {
+        var summaries = new List<AppInstallSummary>
+        {
+            A("Failed", appVersion: "1.0"),
+            A("Succeeded", appVersion: "1.0"),
+            A("InProgress", appVersion: "1.0"),
+            A("InProgress", appVersion: "1.0"),
+        };
+
+        var root = await BuildAnalyticsAsync(summaries);
+
+        // Summary: 1 / (1 + 1) = 50%, NOT 1 / 4 = 25%.
+        var summary = root.GetProperty("summary");
+        Assert.Equal(4, summary.GetProperty("totalInstalls").GetInt32());
+        Assert.Equal(50d, summary.GetProperty("failureRate").GetDouble());
+
+        var version = root.GetProperty("versionBreakdown")[0];
+        Assert.Equal(4, version.GetProperty("installs").GetInt32());
+        Assert.Equal(50d, version.GetProperty("failureRate").GetDouble());
+
+        // All four installs land in today's bucket → same terminal-only rate there.
+        var buckets = root.GetProperty("timeSeries").EnumerateArray().ToList();
+        var todayBucket = buckets.Single(b => b.GetProperty("installs").GetInt32() > 0);
+        Assert.Equal(50d, todayBucket.GetProperty("failureRate").GetDouble());
+    }
+
+    // ── PR0 (2026-07-26): skip / unmeasured classification ──────────────────
+
+    [Fact]
+    public void List_Skips_LeaveRateAndDurations_AndAreReportedSeparately()
+    {
+        var skip = A("Succeeded");
+        skip.TerminalState = "Skipped";
+        skip.DurationSeconds = 0;
+        var unmeasured = A("Succeeded");
+        unmeasured.TerminalState = "Installed";
+        unmeasured.DurationSeconds = 0;
+
+        var summaries = new List<AppInstallSummary> { A("Failed"), A("Succeeded"), skip, unmeasured };
+
+        var root = JsonSerializer.SerializeToElement(
+            AppsAnalyticsHelper.BuildAppsListResponse(summaries, days: 30));
+        var app = root.GetProperty("apps")[0];
+
+        Assert.Equal(4, app.GetProperty("totalInstalls").GetInt32());
+        Assert.Equal(2, app.GetProperty("succeeded").GetInt32());          // real install + unmeasured
+        Assert.Equal(1, app.GetProperty("skipped").GetInt32());
+        Assert.Equal(1, app.GetProperty("unmeasured").GetInt32());
+        // 1 / (1 + 2) — the skip is not an attempt.
+        Assert.Equal(33.3d, app.GetProperty("failureRate").GetDouble());
+        // avg over the single measured row (30s) — the zero must not drag it to 15.
+        Assert.Equal(30d, app.GetProperty("avgDurationSeconds").GetDouble());
+    }
+
+    [Fact]
+    public async Task Analytics_Summary_CarriesSkippedAndUnmeasured_AndP95OverMeasuredOnly()
+    {
+        var skip = A("Succeeded");
+        skip.TerminalState = "Skipped";
+        skip.DurationSeconds = 0;
+        var unmeasured = A("Succeeded");
+        unmeasured.TerminalState = "Installed";
+        unmeasured.DurationSeconds = 0;
+        var measured = A("Succeeded");
+        measured.TerminalState = "Installed";
+        measured.DurationSeconds = 100;
+
+        var root = await BuildAnalyticsAsync(new List<AppInstallSummary> { skip, unmeasured, measured });
+        var summary = root.GetProperty("summary");
+
+        Assert.Equal(2, summary.GetProperty("succeeded").GetInt32());
+        Assert.Equal(1, summary.GetProperty("skipped").GetInt32());
+        Assert.Equal(1, summary.GetProperty("unmeasured").GetInt32());
+        Assert.Equal(100d, summary.GetProperty("avgDurationSeconds").GetDouble());
+        Assert.Equal(100, summary.GetProperty("p95DurationSeconds").GetInt32());
+        Assert.Equal(0d, summary.GetProperty("failureRate").GetDouble());
+    }
+
+    [Fact]
+    public async Task Analytics_DeviceModelBreakdown_RatesAndSampleFloor_UseFinishedInstalls()
+    {
+        var sessionRepo = new Mock<ISessionRepository>();
+        sessionRepo.Setup(r => r.GetSessionAsync(TenantId, It.IsAny<string>()))
+            .ReturnsAsync((string _, string sessionId) => new SessionSummary
+            {
+                TenantId = TenantId,
+                SessionId = sessionId,
+                Manufacturer = "Contoso",
+                Model = sessionId.StartsWith("s-a") ? "ModelA" : "ModelB",
+            });
+
+        var summaries = new List<AppInstallSummary>();
+        // ModelA: 5 finished (1 failed → 20%) + 3 in-flight → ranked, rate undiluted.
+        summaries.Add(A("Failed", sessionId: "s-a-0"));
+        for (var i = 1; i <= 4; i++) summaries.Add(A("Succeeded", sessionId: $"s-a-{i}"));
+        for (var i = 5; i <= 7; i++) summaries.Add(A("InProgress", sessionId: $"s-a-{i}"));
+        // ModelB: 4 finished + 6 in-flight — 10 installs, but below the 5-finished floor.
+        for (var i = 0; i <= 3; i++) summaries.Add(A("Failed", sessionId: $"s-b-{i}"));
+        for (var i = 4; i <= 9; i++) summaries.Add(A("InProgress", sessionId: $"s-b-{i}"));
+
+        var root = await BuildAnalyticsAsync(summaries, sessionRepo.Object);
+
+        var breakdown = root.GetProperty("deviceModelBreakdown");
+        var model = Assert.Single(breakdown.EnumerateArray());
+        Assert.Equal("ModelA", model.GetProperty("model").GetString());
+        Assert.Equal(8, model.GetProperty("installs").GetInt32());
+        Assert.Equal(20d, model.GetProperty("failureRate").GetDouble());
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/AppsProjectionEquivalenceTests.cs

@@ -34,6 +34,14 @@ public class AppsProjectionEquivalenceTests
             ["AppType"] = "Win32",
             ["AppVersion"] = "2.1.0",
             ["Status"] = "Failed",
+            // PR0/PR1 classification columns. The fixture MUST populate every column a Build*
+            // method reads: a column absent here passes the equivalence assert trivially (both
+            // sides map to default) — exactly how the missing TerminalState projection slipped
+            // through and zeroed totalSkipped in production (2026-07-27).
+            ["TerminalState"] = "Skipped",
+            ["AppId"] = "9e8d7c6b-5a49-4832-9182-736451234567",
+            ["EspBlocking"] = true,
+            ["AppIdCollision"] = true,
             ["StartedAt"] = new DateTimeOffset(DateTime.UtcNow.AddHours(-6)),
             ["CompletedAt"] = new DateTimeOffset(DateTime.UtcNow.AddHours(-5)),
             ["DurationSeconds"] = 340,
@@ -83,6 +91,9 @@ public class AppsProjectionEquivalenceTests
 
         Assert.Equal(fromFull.AppName, fromProjected.AppName);
         Assert.Equal(fromFull.Status, fromProjected.Status);
+        // IsSkipTerminalState / skip classification input — see the fixture comment.
+        Assert.Equal(fromFull.TerminalState, fromProjected.TerminalState);
+        Assert.Equal(fromFull.AppIdCollision, fromProjected.AppIdCollision);
         Assert.Equal(fromFull.StartedAt, fromProjected.StartedAt);
         Assert.Equal(fromFull.DurationSeconds, fromProjected.DurationSeconds);
         Assert.Equal(fromFull.DownloadBytes, fromProjected.DownloadBytes);
@@ -122,6 +133,10 @@ public class AppsProjectionEquivalenceTests
         Assert.Equal(fromFull.FailureMessage, fromProjected.FailureMessage);
         Assert.Equal(fromFull.ExitCode, fromProjected.ExitCode);
         Assert.Equal(fromFull.DetectionResult, fromProjected.DetectionResult);
+        Assert.Equal(fromFull.TerminalState, fromProjected.TerminalState);
+        Assert.Equal(fromFull.AppId, fromProjected.AppId);
+        Assert.Equal(fromFull.EspBlocking, fromProjected.EspBlocking);
+        Assert.Equal(fromFull.AppIdCollision, fromProjected.AppIdCollision);
     }
 
     [Fact]

src/Backend/AutopilotMonitor.Functions.Tests/AzureMonitorAlertWebhookFunctionTests.cs

@@ -0,0 +1,172 @@
+using AutopilotMonitor.Functions.Functions.Infrastructure;
+using AutopilotMonitor.Shared.DataAccess;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Pins the pure helpers of <see cref="AzureMonitorAlertWebhookFunction"/> — the anonymous,
+/// secret-gated bridge that turns Azure Monitor Common Alert Schema notifications into ops
+/// events. Load-bearing bits: the secret gate is fail-closed (an unset app setting rejects
+/// everything), severity mapping can never escalate on schema drift (unknown → Info), and
+/// parsing is defensive (only the alert rule name is mandatory).
+/// </summary>
+public class AzureMonitorAlertWebhookFunctionTests
+{
+    // ── SecretMatches ──────────────────────────────────────────────────────
+
+    [Fact]
+    public void SecretMatches_accepts_equal_secrets()
+    {
+        Assert.True(AzureMonitorAlertWebhookFunction.SecretMatches("s3cret-value", "s3cret-value"));
+    }
+
+    [Theory]
+    [InlineData("wrong", "right")]
+    [InlineData("right-but-longer", "right")]
+    [InlineData("", "configured")]
+    [InlineData(null, "configured")]
+    public void SecretMatches_rejects_mismatch_or_missing_provided(string? provided, string configured)
+    {
+        Assert.False(AzureMonitorAlertWebhookFunction.SecretMatches(provided, configured));
+    }
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    public void SecretMatches_fails_closed_when_not_configured(string? configured)
+    {
+        // While the app setting is unset, NOTHING matches — not even an empty provided value.
+        Assert.False(AzureMonitorAlertWebhookFunction.SecretMatches("anything", configured));
+        Assert.False(AzureMonitorAlertWebhookFunction.SecretMatches("", configured));
+        Assert.False(AzureMonitorAlertWebhookFunction.SecretMatches(null, configured));
+    }
+
+    // ── MapSeverity ────────────────────────────────────────────────────────
+
+    [Theory]
+    [InlineData("Sev0", OpsEventSeverity.Critical)]
+    [InlineData("Sev1", OpsEventSeverity.Error)]
+    [InlineData("Sev2", OpsEventSeverity.Warning)]
+    [InlineData("Sev3", OpsEventSeverity.Info)]
+    [InlineData("Sev4", OpsEventSeverity.Info)]
+    public void MapSeverity_maps_azure_scale_to_ops_scale(string azure, string expected)
+    {
+        Assert.Equal(expected, AzureMonitorAlertWebhookFunction.MapSeverity(azure));
+    }
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    [InlineData("Sev5")]
+    [InlineData("Critical")] // someone posting the ops scale directly must not pass through
+    public void MapSeverity_falls_back_to_info_on_unknown_values(string? azure)
+    {
+        Assert.Equal(OpsEventSeverity.Info, AzureMonitorAlertWebhookFunction.MapSeverity(azure));
+    }
+
+    // ── TryParse ───────────────────────────────────────────────────────────
+
+    private const string FullPayload = """
+        {
+          "schemaId": "azureMonitorCommonAlertSchema",
+          "data": {
+            "essentials": {
+              "alertId": "/subscriptions/xxx/providers/Microsoft.AlertsManagement/alerts/abc",
+              "alertRule": "Portal RSC prefetch tail",
+              "severity": "Sev2",
+              "signalType": "Log",
+              "monitorCondition": "Fired",
+              "monitoringService": "Log Alerts V2",
+              "alertTargetIDs": [
+                "/subscriptions/xxx/resourcegroups/rg-autopilotmonitor-prd-gwc/providers/microsoft.insights/components/autopilotmonitorweb"
+              ],
+              "description": "RSC prefetches exceeding 10s",
+              "firedDateTime": "2026-07-27T15:30:00.000Z"
+            },
+            "alertContext": {
+              "condition": {
+                "windowSize": "PT1H",
+                "allOf": [
+                  {
+                    "searchQuery": "dependencies | where name contains '_rsc='",
+                    "metricValue": 18.0,
+                    "operator": "GreaterThan",
+                    "threshold": "0"
+                  }
+                ]
+              }
+            }
+          }
+        }
+        """;
+
+    [Fact]
+    public void TryParse_extracts_essentials_and_metric_value()
+    {
+        var alert = AzureMonitorAlertWebhookFunction.TryParse(FullPayload);
+
+        Assert.NotNull(alert);
+        Assert.Equal("Portal RSC prefetch tail", alert!.AlertRule);
+        Assert.Equal("Sev2", alert.Severity);
+        Assert.Equal("Fired", alert.MonitorCondition);
+        Assert.Equal("RSC prefetches exceeding 10s", alert.Description);
+        Assert.Equal("Log Alerts V2", alert.MonitoringService);
+        // No targetResourceName in the payload → resource name segment of the first target ID.
+        Assert.Equal("autopilotmonitorweb", alert.TargetResource);
+        Assert.Equal(18.0, alert.MetricValue);
+    }
+
+    [Fact]
+    public void TryParse_prefers_explicit_target_resource_name()
+    {
+        const string payload = """
+            {"data":{"essentials":{"alertRule":"r","targetResourceName":"explicit-name",
+            "alertTargetIDs":["/sub/x/other-name"]}}}
+            """;
+        var alert = AzureMonitorAlertWebhookFunction.TryParse(payload);
+        Assert.Equal("explicit-name", alert!.TargetResource);
+    }
+
+    [Fact]
+    public void TryParse_minimal_payload_needs_only_the_alert_rule()
+    {
+        var alert = AzureMonitorAlertWebhookFunction.TryParse(
+            """{"data":{"essentials":{"alertRule":"minimal"}}}""");
+
+        Assert.NotNull(alert);
+        Assert.Equal("minimal", alert!.AlertRule);
+        Assert.Null(alert.Severity);
+        Assert.Equal("Fired", alert.MonitorCondition); // default when absent
+        Assert.Null(alert.Description);
+        Assert.Null(alert.MetricValue);
+    }
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    [InlineData("   ")]
+    [InlineData("not json at all")]
+    [InlineData("{}")]                                          // no data
+    [InlineData("""{"data":{}}""")]                             // no essentials
+    [InlineData("""{"data":{"essentials":{}}}""")]              // no alertRule
+    [InlineData("""{"data":{"essentials":{"alertRule":""}}}""")] // empty alertRule
+    [InlineData("""{"data":{"essentials":{"alertRule":123}}}""")] // wrong type
+    public void TryParse_rejects_unusable_payloads(string? body)
+    {
+        Assert.Null(AzureMonitorAlertWebhookFunction.TryParse(body!));
+    }
+
+    [Fact]
+    public void TryParse_tolerates_malformed_alert_context()
+    {
+        // alertContext with unexpected shapes must not break essentials extraction.
+        const string payload = """
+            {"data":{"essentials":{"alertRule":"r","severity":"Sev0"},
+            "alertContext":{"condition":{"allOf":"not-an-array"}}}}
+            """;
+        var alert = AzureMonitorAlertWebhookFunction.TryParse(payload);
+        Assert.NotNull(alert);
+        Assert.Equal("Sev0", alert!.Severity);
+        Assert.Null(alert.MetricValue);
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/ConfigRedactionTests.cs

@@ -183,4 +183,13 @@ public class ConfigRedactionTests
     public void CanViewSecrets_PureDelegatedReader_False()
         => Assert.False(GetTenantConfigurationFunction.CanViewSecrets(
             new RequestContext { IsDelegatedReader = true, TenantId = TenantA, TargetTenantId = TenantB }));
+
+    [Theory]
+    [InlineData(Constants.TenantRoles.Operator)]
+    [InlineData(Constants.TenantRoles.Viewer)]
+    public void CanViewSecrets_OwnTenantOperatorOrViewer_False(string role)
+        // Member-tier read-only Settings view (config GET is MemberRead): an own-tenant
+        // Operator/Viewer arrives with IsTenantAdmin=false ⇒ always the redacted copy.
+        => Assert.False(GetTenantConfigurationFunction.CanViewSecrets(
+            new RequestContext { UserRole = role, TenantId = TenantA, TargetTenantId = TenantA }));
 }

src/Backend/AutopilotMonitor.Functions.Tests/CrossTenantAccessTests.cs

@@ -324,4 +324,42 @@ public class CrossTenantAccessTests
         Assert.NotNull(entry);
         Assert.Equal(TenantScoping.QueryParam, entry.TenantScoping);
     }
+
+    // ── Cross-tenant rule WRITES: global/rules/{gather,analyze}/{ruleId} ──
+    // Bug (2026-07-24): a Global Admin editing a foreign tenant's rule via the JWT-scoped
+    // rules/{type}/{ruleId} route silently upserted into the GA's OWN tenant (200 → false "saved");
+    // the reload re-read the foreign tenant and showed the unchanged rule. Fix: dedicated cross-tenant
+    // write routes. They MUST be GlobalAdminOnly (a write tier → no delegated-read rescue, so a
+    // tenant-admin / delegated caller can never cross into another tenant's rules) + QueryParam (the
+    // tenant is named by ?tenantId=, not the {ruleId} segment). A downgrade to a tenant tier, or a slip
+    // to TenantScoping.None (which would ignore ?tenantId= and re-introduce the home-tenant misroute),
+    // is caught here.
+
+    [Theory]
+    [InlineData("PUT",    "global/rules/gather/{ruleId}")]
+    [InlineData("DELETE", "global/rules/gather/{ruleId}")]
+    [InlineData("PUT",    "global/rules/analyze/{ruleId}")]
+    [InlineData("DELETE", "global/rules/analyze/{ruleId}")]
+    public void GlobalRuleWriteRoutes_AreGlobalAdminOnly_QueryParam(string httpMethod, string routeTemplate)
+    {
+        var entry = EndpointAccessPolicyCatalog.Entries
+            .FirstOrDefault(e => e.HttpMethod == httpMethod.ToUpperInvariant() && e.RouteTemplate == routeTemplate);
+
+        Assert.NotNull(entry);
+        Assert.Equal(EndpointPolicy.GlobalAdminOnly, entry!.Policy);
+        Assert.Equal(TenantScoping.QueryParam, entry.TenantScoping);
+    }
+
+    [Theory]
+    [InlineData("PUT",    "/api/global/rules/gather/rule-1")]
+    [InlineData("DELETE", "/api/global/rules/gather/rule-1")]
+    [InlineData("PUT",    "/api/global/rules/analyze/rule-1")]
+    [InlineData("DELETE", "/api/global/rules/analyze/rule-1")]
+    public void GlobalRuleWrite_CrossTenant_GlobalAdmin_IsAllowed(string httpMethod, string path)
+    {
+        var (isBlocked, targetTenantId) = SimulateQueryParamCheck(httpMethod, path, TenantA, hasGlobalScope: true, queryTenantId: TenantB);
+
+        Assert.False(isBlocked, $"Global Admin should reach the target tenant on the cross-tenant write route: {httpMethod} {path}");
+        Assert.Equal(TenantB, targetTenantId);
+    }
 }

src/Backend/AutopilotMonitor.Functions.Tests/DeletionManifestBuilderTests.cs

@@ -75,8 +75,9 @@ public class DeletionManifestBuilderTests
         Assert.Equal(2,  c["classifierVerdictsByIdLevel"]);
         Assert.Equal(11, c["signalsByKind"]);
 
-        // 15 cascade tables + tombstone = 16 steps; no inventory steps when no contributions row.
-        Assert.Equal(16, manifest.Steps.Count);
+        // 16 cascade tables (incl. SessionTimeBreakdowns, F1 PR2) + tombstone = 17 steps;
+        // no inventory steps when no contributions row.
+        Assert.Equal(17, manifest.Steps.Count);
         Assert.Equal(DeletionStepClass.Final, manifest.Steps.Last().Class);
     }
 
@@ -102,7 +103,8 @@ public class DeletionManifestBuilderTests
             .WithSessionsByStage(1, propValue: "stage_marker")
             .WithDeadEndsByReason(1, propValue: "deadend_marker")
             .WithClassifierVerdictsByIdLevel(1, propValue: "classifier_marker")
-            .WithSignalsByKind(1, propValue: "signalkind_marker"));
+            .WithSignalsByKind(1, propValue: "signalkind_marker")
+            .WithSessionTimeBreakdown(propValue: "breakdown_marker"));
 
         var manifest = await NewBuilder(reader).BuildAsync(
             TenantId, SessionId, "admin_delete",
@@ -145,8 +147,8 @@ public class DeletionManifestBuilderTests
             new DeletionRetentionContext { TenantRetentionDays = 90 });
 
         var nonTombstoneSteps = manifest.Steps.Where(s => s.Class != DeletionStepClass.Final).ToList();
-        // Without a contributions row, neither inventory step is emitted → exactly 15 cascade-table steps.
-        Assert.Equal(15, nonTombstoneSteps.Count);
+        // Without a contributions row, neither inventory step is emitted → exactly 16 cascade-table steps.
+        Assert.Equal(16, nonTombstoneSteps.Count);
         foreach (var step in nonTombstoneSteps)
         {
             Assert.Equal(0, step.RowCount);
@@ -175,8 +177,8 @@ public class DeletionManifestBuilderTests
             new DeletionActor { Type = "admin", Actor = "alice@example.com" },
             new DeletionRetentionContext { TenantRetentionDays = 90 });
 
-        // Expect: 15 cascade + 2 inventory + 1 tombstone = 18 steps.
-        Assert.Equal(18, manifest.Steps.Count);
+        // Expect: 16 cascade + 2 inventory + 1 tombstone = 19 steps.
+        Assert.Equal(19, manifest.Steps.Count);
 
         var aggregate = manifest.Steps.Single(s => s.Class == DeletionStepClass.Aggregate);
         Assert.Equal(DeletionStepNames.SoftwareInventoryDecrement, aggregate.Step);
@@ -640,6 +642,7 @@ public class DeletionManifestBuilderTests
         private TableEntity? _vulnerabilityReport;
         private TableEntity? _deviceSnapshot;
         private TableEntity? _eventSessionIndex;
+        private TableEntity? _sessionTimeBreakdown;
         private TableEntity? _sessionInventoryContributions;
         private string _sessionsIndexRowKey = "INDEX_" + SessionId;
 
@@ -719,6 +722,16 @@ public class DeletionManifestBuilderTests
             return this;
         }
 
+        public ReaderSeed WithSessionTimeBreakdown(string propValue = "tab")
+        {
+            _sessionTimeBreakdown = new TableEntity(TenantId, SessionId)
+            {
+                ["Marker"] = propValue,
+            };
+            _sessionTimeBreakdown.ETag = new ETag("0xTAB");
+            return this;
+        }
+
         public ReaderSeed WithEventTypeIndex(int n, string propValue = "eti")
         {
             // PK = {tenantId}_{eventType}, RK ends with _{sessionId}.
@@ -796,6 +809,8 @@ public class DeletionManifestBuilderTests
                   .ReturnsAsync(_deviceSnapshot);
             reader.Setup(r => r.GetEntityOrNullAsync(Constants.TableNames.EventSessionIndex, TenantId, SessionId, It.IsAny<CancellationToken>()))
                   .ReturnsAsync(_eventSessionIndex);
+            reader.Setup(r => r.GetEntityOrNullAsync(Constants.TableNames.SessionTimeBreakdowns, TenantId, SessionId, It.IsAny<CancellationToken>()))
+                  .ReturnsAsync(_sessionTimeBreakdown);
             reader.Setup(r => r.GetEntityOrNullAsync(Constants.TableNames.SessionInventoryContributions, TenantId, SessionId, It.IsAny<CancellationToken>()))
                   .ReturnsAsync(_sessionInventoryContributions);
 

src/Backend/AutopilotMonitor.Functions.Tests/DeviceJourneyAndFtrTests.cs

@@ -0,0 +1,809 @@
+using AutopilotMonitor.Functions.Functions.Metrics;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// F2 PR4 (insights spec §F2): device-key normalization (junk serials), terminal-only chain
+/// maintenance (upsert, cap, deletion-ref pruning), journey grouping (first-success end,
+/// 30-day gap, Incomplete attempts, WG-Pending stays open), the pure daily FTR aggregation
+/// core (<see cref="MaintenanceService.BuildDeviceJourneyAggregates"/>) and the persistence
+/// round-trips (table-storage-serialization rule — every property must survive Store→Map).
+/// </summary>
+public class DeviceJourneyAndFtrTests
+{
+    private static readonly DateTime T0 = new(2026, 7, 20, 10, 0, 0, DateTimeKind.Utc);
+    private const string TenantA = "aaaaaaaa-0000-0000-0000-000000000001";
+    private const string TenantB = "bbbbbbbb-0000-0000-0000-000000000002";
+
+    private static SessionSummary Session(
+        string sessionId,
+        SessionStatus status = SessionStatus.Succeeded,
+        DateTime? startedAt = null,
+        DateTime? completedAt = null,
+        int? durationSeconds = 1800,
+        string? serial = "PF4X1ABC",
+        string tenantId = TenantA,
+        bool wg = false,
+        string enrollmentType = "v1",
+        string? adminMarkedAction = null)
+        => new()
+        {
+            TenantId = tenantId,
+            SessionId = sessionId,
+            SerialNumber = serial!,
+            Status = status,
+            StartedAt = startedAt ?? T0,
+            CompletedAt = completedAt ?? (startedAt ?? T0).AddSeconds(durationSeconds ?? 0),
+            DurationSeconds = durationSeconds,
+            IsPreProvisioned = wg,
+            EnrollmentType = enrollmentType,
+            AdminMarkedAction = adminMarkedAction,
+            Manufacturer = "Contoso",
+            Model = "Laptop 5",
+        };
+
+    private static DeviceSessionRef Ref(
+        string sessionId, SessionStatus status, DateTime startedAt, DateTime? completedAt = null)
+        => new()
+        {
+            SessionId = sessionId,
+            Status = status.ToString(),
+            StartedAt = startedAt,
+            CompletedAt = completedAt ?? startedAt.AddMinutes(30),
+        };
+
+    // ── serial normalization / junk exclusion ───────────────────────────────
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    [InlineData("   ")]
+    [InlineData("abc")]                      // shorter than 4 chars
+    [InlineData("0")]
+    [InlineData("None")]
+    [InlineData("INVALID")]
+    [InlineData("Unknown")]                  // the agent's WMI-failure sentinel (audit Q5)
+    [InlineData("unknown")]
+    [InlineData("System Serial Number")]
+    [InlineData("TO BE FILLED BY O.E.M.")]
+    [InlineData("Default String")]
+    [InlineData("  Default string  ")]       // junk match happens after trim + case-fold
+    public void NormalizeSerial_JunkAndPlaceholders_YieldNull(string? raw)
+    {
+        Assert.Null(DeviceJourneyCalculator.NormalizeSerial(raw));
+    }
+
+    [Theory]
+    [InlineData("PF4X1ABC", "pf4x1abc")]
+    [InlineData("  PF4X1ABC  ", "pf4x1abc")] // trim
+    [InlineData("pf4x1abc", "pf4x1abc")]
+    [InlineData("1234", "1234")]             // exactly 4 chars is a valid identity
+    public void NormalizeSerial_RealSerials_TrimAndCaseFold(string raw, string expected)
+    {
+        Assert.Equal(expected, DeviceJourneyCalculator.NormalizeSerial(raw));
+    }
+
+    // ── terminal set / ref building ─────────────────────────────────────────
+
+    [Theory]
+    [InlineData(SessionStatus.Succeeded, true)]
+    [InlineData(SessionStatus.Failed, true)]
+    [InlineData(SessionStatus.Incomplete, true)]
+    [InlineData(SessionStatus.InProgress, false)]
+    [InlineData(SessionStatus.Pending, false)]      // WG part 1 sealed — an OPEN session
+    [InlineData(SessionStatus.Stalled, false)]
+    [InlineData(SessionStatus.AwaitingUser, false)]
+    [InlineData(SessionStatus.Unknown, false)]
+    public void IsTerminal_MatchesTheF2TerminalSet(SessionStatus status, bool expected)
+    {
+        Assert.Equal(expected, DeviceJourneyCalculator.IsTerminal(status));
+    }
+
+    [Fact]
+    public void BuildSessionRef_MapsAllFields_DurationIsAuthoritativeDurationSeconds()
+    {
+        // CompletedAt deliberately diverges from StartedAt + DurationSeconds (25 % of terminal
+        // sessions in production) — the ref must carry DurationSeconds verbatim.
+        var session = Session("s-1", startedAt: T0, completedAt: T0.AddHours(5), durationSeconds: 1800,
+            wg: true, enrollmentType: "v1", adminMarkedAction: "Succeeded");
+
+        var reference = DeviceJourneyCalculator.BuildSessionRef(session)!;
+
+        Assert.Equal("s-1", reference.SessionId);
+        Assert.Equal(T0, reference.StartedAt);
+        Assert.Equal(T0.AddHours(5), reference.CompletedAt);
+        Assert.Equal("Succeeded", reference.Status);
+        Assert.Equal("v1", reference.EnrollmentType);
+        Assert.True(reference.IsPreProvisioned);
+        Assert.Equal(1800, reference.DurationSeconds);
+        Assert.True(reference.AdminMarked);
+    }
+
+    [Fact]
+    public void BuildSessionRef_WhiteGlovePending_YieldsNull_NeverAnAttempt()
+    {
+        var pending = Session("s-1", status: SessionStatus.Pending, wg: true, durationSeconds: null);
+        Assert.Null(DeviceJourneyCalculator.BuildSessionRef(pending));
+    }
+
+    [Fact]
+    public void BuildSessionRef_Incomplete_KeepsNullDuration()
+    {
+        var incomplete = Session("s-1", status: SessionStatus.Incomplete, durationSeconds: null);
+        var reference = DeviceJourneyCalculator.BuildSessionRef(incomplete)!;
+        Assert.Equal("Incomplete", reference.Status);
+        Assert.Null(reference.DurationSeconds);
+        Assert.False(reference.AdminMarked);
+    }
+
+    // ── chain maintenance ───────────────────────────────────────────────────
+
+    [Fact]
+    public void MergeChain_UpsertsBySessionId_ReclassificationReplacesTheEntry()
+    {
+        var existing = new List<DeviceSessionRef> { Ref("s-1", SessionStatus.Incomplete, T0) };
+        var update = Ref("s-1", SessionStatus.Succeeded, T0);
+
+        var merged = DeviceJourneyCalculator.MergeChain(existing, new[] { update });
+
+        var entry = Assert.Single(merged);
+        Assert.Equal("Succeeded", entry.Status);
+    }
+
+    [Fact]
+    public void MergeChain_OrdersByStartedAt_TiebreakSessionId_AndCapsToMostRecent()
+    {
+        var refs = new List<DeviceSessionRef>();
+        for (var i = 0; i < 25; i++)
+            refs.Add(Ref($"s-{i:D2}", SessionStatus.Failed, T0.AddDays(i)));
+        // Same StartedAt as s-24 — the ordinal sessionId tiebreak must keep the order stable.
+        refs.Add(Ref("s-00-twin", SessionStatus.Failed, T0.AddDays(24)));
+
+        var merged = DeviceJourneyCalculator.MergeChain(null, refs);
+
+        Assert.Equal(20, merged.Count);                       // cap 20 most recent
+        Assert.Equal("s-06", merged[0].SessionId);            // s-00..s-05 dropped (oldest)
+        Assert.Equal(T0.AddDays(24), merged[^1].StartedAt);
+        Assert.Equal("s-24", merged[^1].SessionId);           // "s-00-twin" < "s-24" ordinal → sorts first
+        Assert.Equal("s-00-twin", merged[^2].SessionId);
+    }
+
+    [Fact]
+    public void MergeChain_DropsNonTerminalRefs_Defensively()
+    {
+        var corrupt = new List<DeviceSessionRef>
+        {
+            Ref("s-1", SessionStatus.Succeeded, T0),
+            new() { SessionId = "s-2", Status = "Pending", StartedAt = T0.AddDays(1) },
+            new() { SessionId = "s-3", Status = "garbage", StartedAt = T0.AddDays(2) },
+        };
+
+        var merged = DeviceJourneyCalculator.MergeChain(corrupt, Array.Empty<DeviceSessionRef>());
+
+        Assert.Equal("s-1", Assert.Single(merged).SessionId);
+    }
+
+    [Fact]
+    public void RemoveSessionRefs_DropsDeletedSessions_EmptyChainSignalsRowDeletion()
+    {
+        var chain = new List<DeviceSessionRef>
+        {
+            Ref("s-1", SessionStatus.Failed, T0),
+            Ref("s-2", SessionStatus.Succeeded, T0.AddDays(1)),
+        };
+
+        var partial = DeviceJourneyCalculator.RemoveSessionRefs(
+            chain, new HashSet<string>(StringComparer.Ordinal) { "s-1" });
+        Assert.Equal("s-2", Assert.Single(partial).SessionId);
+
+        var empty = DeviceJourneyCalculator.RemoveSessionRefs(
+            chain, new HashSet<string>(StringComparer.Ordinal) { "s-1", "s-2" });
+        Assert.Empty(empty);
+    }
+
+    // ── journey grouping ────────────────────────────────────────────────────
+
+    [Fact]
+    public void GroupJourneys_SingleSuccess_IsOneCompletedFirstTimeRightJourney()
+    {
+        var journeys = DeviceJourneyCalculator.GroupJourneys(
+            new[] { Ref("s-1", SessionStatus.Succeeded, T0) });
+
+        var journey = Assert.Single(journeys);
+        Assert.True(journey.Completed);
+        Assert.Single(journey.Attempts);
+        Assert.Equal("s-1", journey.CompletingRef!.SessionId);
+    }
+
+    [Fact]
+    public void GroupJourneys_FailAndIncompleteAreAttempts_JourneyEndsWithFirstSuccess()
+    {
+        var journeys = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0),
+            Ref("s-2", SessionStatus.Incomplete, T0.AddHours(2)),
+            Ref("s-3", SessionStatus.Succeeded, T0.AddHours(4)),
+        });
+
+        var journey = Assert.Single(journeys);
+        Assert.True(journey.Completed);
+        Assert.Equal(3, journey.Attempts.Count); // Incomplete is a non-successful attempt
+    }
+
+    [Fact]
+    public void GroupJourneys_SessionAfterSuccess_StartsANewJourney_Redeployment()
+    {
+        var journeys = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Succeeded, T0),
+            Ref("s-2", SessionStatus.Failed, T0.AddDays(2)),
+        });
+
+        Assert.Equal(2, journeys.Count);
+        Assert.True(journeys[0].Completed);
+        Assert.False(journeys[1].Completed);   // open — no failed verdict, just not done yet
+        Assert.Null(journeys[1].CompletingRef);
+    }
+
+    [Fact]
+    public void GroupJourneys_FailuresOnly_JourneyStaysOpen_NeverCompleted()
+    {
+        var journeys = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0),
+            Ref("s-2", SessionStatus.Failed, T0.AddDays(1)),
+        });
+
+        var journey = Assert.Single(journeys);
+        Assert.False(journey.Completed);
+        Assert.Equal(2, journey.Attempts.Count);
+    }
+
+    [Fact]
+    public void GroupJourneys_GapOver30Days_StartsANewJourney_AbandonedStaysUncompleted()
+    {
+        var journeys = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0, completedAt: T0.AddHours(1)),
+            Ref("s-2", SessionStatus.Failed, T0.AddHours(1).AddDays(31)),
+            Ref("s-3", SessionStatus.Succeeded, T0.AddHours(1).AddDays(31).AddHours(3)),
+        });
+
+        Assert.Equal(2, journeys.Count);
+        Assert.False(journeys[0].Completed);            // shelved/abandoned — never counts for FTR
+        Assert.Single(journeys[0].Attempts);
+        Assert.True(journeys[1].Completed);
+        Assert.Equal(2, journeys[1].Attempts.Count);
+    }
+
+    [Fact]
+    public void GroupJourneys_GapBoundary_Exactly30DaysStaysOneJourney_StrictlyMoreSplits()
+    {
+        var prevEnd = T0.AddHours(1);
+
+        var exactly30 = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0, completedAt: prevEnd),
+            Ref("s-2", SessionStatus.Failed, prevEnd.AddDays(30)),
+        });
+        Assert.Single(exactly30); // spec: "> 30 days" — the boundary itself does not split
+
+        var over30 = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0, completedAt: prevEnd),
+            Ref("s-2", SessionStatus.Failed, prevEnd.AddDays(30).AddHours(1)),
+        });
+        Assert.Equal(2, over30.Count);
+    }
+
+    [Fact]
+    public void GroupJourneys_GapMeasuredFromCompletedAt_NotStartedAt()
+    {
+        // Previous attempt STARTED 40 days before the next one but went terminal only 10 days
+        // before it — the gap is 10 days, one journey. StartedAt-based measuring would split.
+        var journeys = DeviceJourneyCalculator.GroupJourneys(new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0.AddDays(-40), completedAt: T0.AddDays(-10)),
+            Ref("s-2", SessionStatus.Succeeded, T0),
+        });
+
+        var journey = Assert.Single(journeys);
+        Assert.Equal(2, journey.Attempts.Count);
+    }
+
+    [Fact]
+    public void GroupJourneys_WhiteGloveIsOneSessionRow_OneAttempt_NoStitching()
+    {
+        // WG part 1 + part 2 share one session row (second-sweep finding): a completed WG
+        // enrollment is exactly one ref and one first-time-right attempt.
+        var wg = Session("s-1", wg: true, durationSeconds: 5400, completedAt: T0.AddDays(6));
+        var reference = DeviceJourneyCalculator.BuildSessionRef(wg)!;
+
+        var journeys = DeviceJourneyCalculator.GroupJourneys(new[] { reference });
+
+        var journey = Assert.Single(journeys);
+        Assert.True(journey.Completed);
+        Assert.Single(journey.Attempts);
+        Assert.Equal(5400, journey.Attempts[0].DurationSeconds); // pause-free combined duration, verbatim
+    }
+
+    [Fact]
+    public void Derive_CurrentJourneyAttempts_IsTheLastJourneys_OpenOrCompleted()
+    {
+        var (journeyCount, currentAttempts) = DeviceJourneyCalculator.Derive(new[]
+        {
+            Ref("s-1", SessionStatus.Succeeded, T0),
+            Ref("s-2", SessionStatus.Failed, T0.AddDays(1)),
+            Ref("s-3", SessionStatus.Failed, T0.AddDays(2)),
+        });
+
+        Assert.Equal(2, journeyCount);
+        Assert.Equal(2, currentAttempts); // the open redeployment journey has 2 attempts so far
+
+        Assert.Equal((0, 0), DeviceJourneyCalculator.Derive(Array.Empty<DeviceSessionRef>()));
+    }
+
+    // ── history row assembly (display precedence) ───────────────────────────
+
+    [Fact]
+    public void BuildDeviceHistoryRow_DisplayFields_FollowTheNewestChainEntry()
+    {
+        var newest = Session("s-2", serial: " PF4X1ABC ", startedAt: T0.AddDays(1));
+        var chain = new List<DeviceSessionRef>
+        {
+            Ref("s-1", SessionStatus.Failed, T0),
+            Ref("s-2", SessionStatus.Succeeded, T0.AddDays(1)),
+        };
+
+        var row = TableStorageService.BuildDeviceHistoryRow(TenantA, "pf4x1abc", chain, existing: null, newest);
+
+        Assert.Equal("PF4X1ABC", row.SerialNumber); // trimmed original casing
+        Assert.Equal("Contoso", row.Manufacturer);
+        Assert.Equal("Laptop 5", row.Model);
+        Assert.Equal(2, row.CurrentJourneyAttempts);
+        Assert.Equal(1, row.JourneyCount);
+        Assert.Equal(DeviceJourneyCalculator.CurrentVersion, row.JourneyVersion);
+    }
+
+    [Fact]
+    public void BuildDeviceHistoryRow_BackfillingAnOlderSession_KeepsExistingDisplayFields()
+    {
+        var older = Session("s-1", serial: "pf4x1abc", startedAt: T0);
+        older.Model = "Laptop 4"; // stale model from the older enrollment
+        var chain = new List<DeviceSessionRef>
+        {
+            Ref("s-1", SessionStatus.Failed, T0),
+            Ref("s-2", SessionStatus.Succeeded, T0.AddDays(1)),
+        };
+        var existing = new DeviceHistory
+        {
+            TenantId = TenantA, SerialKey = "pf4x1abc",
+            SerialNumber = "PF4X1ABC", Manufacturer = "Contoso", Model = "Laptop 5",
+        };
+
+        var row = TableStorageService.BuildDeviceHistoryRow(TenantA, "pf4x1abc", chain, existing, older);
+
+        Assert.Equal("PF4X1ABC", row.SerialNumber); // not regressed to the older session's data
+        Assert.Equal("Laptop 5", row.Model);
+    }
+
+    // ── persistence round-trips ─────────────────────────────────────────────
+
+    [Fact]
+    public void DeviceHistoryEntity_RoundTripsAllFields_IncludingChainJson()
+    {
+        var history = new DeviceHistory
+        {
+            TenantId = TenantA,
+            SerialKey = "pf4x1abc",
+            SerialNumber = "PF4X1ABC",
+            Manufacturer = "Contoso",
+            Model = "Laptop 5",
+            CurrentJourneyAttempts = 2,
+            JourneyCount = 3,
+            JourneyVersion = 1,
+            LastUpdated = T0,
+            Chain = new List<DeviceSessionRef>
+            {
+                new()
+                {
+                    SessionId = "s-1", StartedAt = T0, CompletedAt = T0.AddHours(5),
+                    Status = "Succeeded", EnrollmentType = "v2", IsPreProvisioned = true,
+                    DurationSeconds = 1800, AdminMarked = true,
+                },
+                new()
+                {
+                    SessionId = "s-2", StartedAt = T0.AddDays(1), CompletedAt = null,
+                    Status = "Incomplete", EnrollmentType = "v1", DurationSeconds = null,
+                },
+            },
+        };
+
+        var mapped = TableStorageService.MapToDeviceHistory(
+            TableStorageService.BuildDeviceHistoryEntity(history));
+
+        Assert.Equal(TenantA, mapped.TenantId);
+        Assert.Equal("pf4x1abc", mapped.SerialKey);
+        Assert.Equal("PF4X1ABC", mapped.SerialNumber);
+        Assert.Equal("Contoso", mapped.Manufacturer);
+        Assert.Equal("Laptop 5", mapped.Model);
+        Assert.Equal(2, mapped.CurrentJourneyAttempts);
+        Assert.Equal(3, mapped.JourneyCount);
+        Assert.Equal(1, mapped.JourneyVersion);
+        Assert.Equal(T0, mapped.LastUpdated);
+
+        Assert.Equal(2, mapped.Chain.Count);
+        var first = mapped.Chain[0];
+        Assert.Equal("s-1", first.SessionId);
+        Assert.Equal(T0, first.StartedAt);
+        Assert.Equal(T0.AddHours(5), first.CompletedAt);
+        Assert.Equal("Succeeded", first.Status);
+        Assert.Equal("v2", first.EnrollmentType);
+        Assert.True(first.IsPreProvisioned);
+        Assert.Equal(1800, first.DurationSeconds);
+        Assert.True(first.AdminMarked);
+        var second = mapped.Chain[1];
+        Assert.Null(second.CompletedAt);      // Incomplete: no terminal timestamp requirement
+        Assert.Null(second.DurationSeconds);  // Incomplete: deliberately no duration — stays null
+        Assert.False(second.AdminMarked);
+    }
+
+    [Fact]
+    public void DeviceHistoryEntity_RowKeyEncoding_MakesForbiddenKeyCharsSafe_AndRoundTrips()
+    {
+        // '/', '\', '#', '?' are forbidden in Table Storage keys — a defensive encoding keeps a
+        // pathological serial writable, and the mapper restores the exact normalized serial.
+        var weird = "ab/cd\\ef#1?x";
+        var history = new DeviceHistory { TenantId = TenantA, SerialKey = weird };
+
+        var entity = TableStorageService.BuildDeviceHistoryEntity(history);
+        Assert.All(new[] { "/", "\\", "#", "?" }, c => Assert.DoesNotContain(c, entity.RowKey));
+
+        Assert.Equal(weird, TableStorageService.MapToDeviceHistory(entity).SerialKey);
+
+        // Ordinary serials stay human-readable.
+        Assert.Equal("pf4x1abc", TableStorageService.SerialRowKey("pf4x1abc"));
+    }
+
+    [Fact]
+    public void DeviceJourneyAggregateEntity_RoundTripsAllFields_RowKeyIsTheDate()
+    {
+        var aggregate = new DeviceJourneyDailyAggregate
+        {
+            TenantId = TenantA,
+            Date = "2026-07-20",
+            JourneyVersion = 1,
+            CompletedJourneyCount = 23,
+            FirstTimeRightCount = 19,
+            ExcludedSessionCount = 2,
+            ComputedAt = T0,
+            AttemptHistogram = new List<DeviceJourneyAttemptBucket>
+            {
+                new() { Attempts = 1, JourneyCount = 19 },
+                new() { Attempts = 2, JourneyCount = 3 },
+                new() { Attempts = 4, JourneyCount = 1 },
+            },
+        };
+
+        var entity = TableStorageService.BuildDeviceJourneyAggregateEntity(aggregate);
+        Assert.Equal("2026-07-20", entity.RowKey);
+
+        var mapped = TableStorageService.MapToDeviceJourneyAggregate(entity);
+        Assert.Equal(TenantA, mapped.TenantId);
+        Assert.Equal("2026-07-20", mapped.Date);
+        Assert.Equal(1, mapped.JourneyVersion);
+        Assert.Equal(23, mapped.CompletedJourneyCount);
+        Assert.Equal(19, mapped.FirstTimeRightCount);
+        Assert.Equal(2, mapped.ExcludedSessionCount);
+        Assert.Equal(T0, mapped.ComputedAt);
+        Assert.Equal(3, mapped.AttemptHistogram.Count);
+        Assert.Equal(1, mapped.AttemptHistogram[0].Attempts);
+        Assert.Equal(19, mapped.AttemptHistogram[0].JourneyCount);
+        Assert.Equal(4, mapped.AttemptHistogram[2].Attempts);
+    }
+
+    // ── daily FTR aggregation core ──────────────────────────────────────────
+
+    private static readonly DateTime WindowStart = T0.Date.AddDays(-30);
+    private static readonly DateTime WindowEnd = T0.Date.AddDays(1);
+
+    private static DeviceHistory History(string tenantId, params DeviceSessionRef[] chain)
+        => new() { TenantId = tenantId, SerialKey = Guid.NewGuid().ToString("N"), Chain = chain.ToList() };
+
+    [Fact]
+    public void FtrAggregates_BucketByCompletingSessionsStartDate_AndMirrorGlobal()
+    {
+        var histories = new List<DeviceHistory>
+        {
+            // Tenant A device 1: fail → success (2 attempts, completes on T0 date).
+            History(TenantA,
+                Ref("a1-1", SessionStatus.Failed, T0.AddDays(-1)),
+                Ref("a1-2", SessionStatus.Succeeded, T0)),
+            // Tenant A device 2: first-time right on the same date.
+            History(TenantA, Ref("a2-1", SessionStatus.Succeeded, T0.AddHours(2))),
+            // Tenant B device: first-time right on a different date.
+            History(TenantB, Ref("b1-1", SessionStatus.Succeeded, T0.AddDays(-3))),
+        };
+
+        var rows = MaintenanceService.BuildDeviceJourneyAggregates(
+            histories, Array.Empty<SessionSummary>(), WindowStart, WindowEnd, T0);
+
+        // Tenant A (1 date) + Tenant B (1 date) + global (2 dates) = 4 rows.
+        Assert.Equal(4, rows.Count);
+
+        var tenantARow = rows.Single(r => r.TenantId == TenantA);
+        Assert.Equal(T0.ToString("yyyy-MM-dd"), tenantARow.Date);
+        Assert.Equal(2, tenantARow.CompletedJourneyCount);
+        Assert.Equal(1, tenantARow.FirstTimeRightCount);
+        Assert.Equal(DeviceJourneyCalculator.CurrentVersion, tenantARow.JourneyVersion);
+        Assert.Equal(2, tenantARow.AttemptHistogram.Count);
+        Assert.Equal(1, tenantARow.AttemptHistogram[0].Attempts);
+        Assert.Equal(1, tenantARow.AttemptHistogram[0].JourneyCount);
+        Assert.Equal(2, tenantARow.AttemptHistogram[1].Attempts);
+
+        var globalOnT0 = rows.Single(r => r.TenantId == "global" && r.Date == T0.ToString("yyyy-MM-dd"));
+        Assert.Equal(2, globalOnT0.CompletedJourneyCount); // tenant B's journey sits on its own date
+    }
+
+    [Fact]
+    public void FtrAggregates_OpenAndAbandonedJourneys_NeverCount()
+    {
+        var histories = new List<DeviceHistory>
+        {
+            // Open journey: failures only (e.g. a WG device still waiting for its user session
+            // has, at most, prior failed attempts in the chain — the Pending row itself never
+            // becomes a ref).
+            History(TenantA, Ref("a1-1", SessionStatus.Failed, T0)),
+            // Abandoned journey (31d gap, no success) followed by a completed journey.
+            History(TenantA,
+                Ref("a2-1", SessionStatus.Failed, T0.AddDays(-20), completedAt: T0.AddDays(-20).AddHours(1)),
+                Ref("a2-2", SessionStatus.Succeeded, T0.AddDays(11).AddHours(2))),
+        };
+
+        var rows = MaintenanceService.BuildDeviceJourneyAggregates(
+            histories, Array.Empty<SessionSummary>(), WindowStart, T0.Date.AddDays(12), T0);
+
+        // Only the completed journey produced rows: tenant + global on one date.
+        Assert.Equal(2, rows.Count);
+        var tenantRow = rows.Single(r => r.TenantId == TenantA);
+        Assert.Equal(1, tenantRow.CompletedJourneyCount);
+        Assert.Equal(1, tenantRow.FirstTimeRightCount); // the abandoned attempt belongs to a DIFFERENT journey
+        Assert.Equal(T0.AddDays(11).ToString("yyyy-MM-dd"), tenantRow.Date);
+    }
+
+    [Fact]
+    public void FtrAggregates_JourneysCompletedOutsideTheWindow_AreSkipped()
+    {
+        var histories = new List<DeviceHistory>
+        {
+            History(TenantA, Ref("a1-1", SessionStatus.Succeeded, WindowStart.AddDays(-1))),
+        };
+
+        var rows = MaintenanceService.BuildDeviceJourneyAggregates(
+            histories, Array.Empty<SessionSummary>(), WindowStart, WindowEnd, T0);
+
+        Assert.Empty(rows);
+    }
+
+    [Fact]
+    public void FtrAggregates_JunkSerialExclusions_ProduceDisclosureRows()
+    {
+        var excluded = new List<SessionSummary>
+        {
+            Session("x-1", serial: "Unknown", startedAt: T0),
+            Session("x-2", serial: "Unknown", startedAt: T0.AddHours(1)),
+            Session("x-3", serial: "Unknown", startedAt: WindowStart.AddDays(-2)), // outside window — ignored
+        };
+
+        var rows = MaintenanceService.BuildDeviceJourneyAggregates(
+            Array.Empty<DeviceHistory>(), excluded, WindowStart, WindowEnd, T0);
+
+        Assert.Equal(2, rows.Count); // tenant + global disclosure rows, no journey claim
+        var tenantRow = rows.Single(r => r.TenantId == TenantA);
+        Assert.Equal(0, tenantRow.CompletedJourneyCount);
+        Assert.Equal(0, tenantRow.FirstTimeRightCount);
+        Assert.Empty(tenantRow.AttemptHistogram);
+        Assert.Equal(2, tenantRow.ExcludedSessionCount);
+        Assert.Equal(2, rows.Single(r => r.TenantId == "global").ExcludedSessionCount);
+    }
+
+    [Fact]
+    public void FtrAggregates_TwoCompletedJourneysOfOneDevice_SameDay_BothCount()
+    {
+        // Redeploy on the same day: success closes journey 1, the next success closes journey 2.
+        var histories = new List<DeviceHistory>
+        {
+            History(TenantA,
+                Ref("a1-1", SessionStatus.Succeeded, T0),
+                Ref("a1-2", SessionStatus.Succeeded, T0.AddHours(3))),
+        };
+
+        var rows = MaintenanceService.BuildDeviceJourneyAggregates(
+            histories, Array.Empty<SessionSummary>(), WindowStart, WindowEnd, T0);
+
+        var tenantRow = rows.Single(r => r.TenantId == TenantA);
+        Assert.Equal(2, tenantRow.CompletedJourneyCount);
+        Assert.Equal(2, tenantRow.FirstTimeRightCount);
+    }
+
+    // ── attempt number (session banner, PR5) ────────────────────────────────
+
+    [Fact]
+    public void ComputeAttemptNumber_TerminalSessionInChain_ReturnsPositionWithinItsJourney()
+    {
+        var chain = new[]
+        {
+            Ref("s-1", SessionStatus.Failed, T0),
+            Ref("s-2", SessionStatus.Succeeded, T0.AddHours(2)),
+            Ref("s-3", SessionStatus.Failed, T0.AddDays(5)), // redeployment → new journey
+        };
+
+        Assert.Equal(1, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-1", T0));
+        Assert.Equal(2, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-2", T0.AddHours(2)));
+        Assert.Equal(1, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-3", T0.AddDays(5)));
+    }
+
+    [Fact]
+    public void ComputeAttemptNumber_LiveSession_ContinuesTheOpenJourney()
+    {
+        var chain = new[] { Ref("s-1", SessionStatus.Failed, T0) };
+        // s-2 is still InProgress — not a chain ref; the virtual-attempt rule places it.
+        Assert.Equal(2, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-2", T0.AddDays(1)));
+    }
+
+    [Fact]
+    public void ComputeAttemptNumber_LiveSession_AfterCompletedJourney_IsAttemptOne()
+    {
+        var chain = new[] { Ref("s-1", SessionStatus.Succeeded, T0) };
+        Assert.Equal(1, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-2", T0.AddDays(2)));
+    }
+
+    [Fact]
+    public void ComputeAttemptNumber_LiveSession_AfterThirtyDayGap_IsAttemptOne()
+    {
+        var prevEnd = T0.AddHours(1);
+        var chain = new[] { Ref("s-1", SessionStatus.Failed, T0, completedAt: prevEnd) };
+        Assert.Equal(1, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-2", prevEnd.AddDays(31)));
+        // Inside the gap boundary the open journey continues instead.
+        Assert.Equal(2, DeviceJourneyCalculator.ComputeAttemptNumber(chain, "s-3", prevEnd.AddDays(29)));
+    }
+
+    [Fact]
+    public void ComputeAttemptNumber_EmptyChain_YieldsNull_NoGuessedPosition()
+    {
+        Assert.Null(DeviceJourneyCalculator.ComputeAttemptNumber(
+            Array.Empty<DeviceSessionRef>(), "s-1", T0));
+    }
+
+    // ── fleet response helpers (PR5) ────────────────────────────────────────
+
+    [Fact]
+    public void SumAggregates_SumsAdditiveCounts_AndMergesHistograms()
+    {
+        var daily = new List<DeviceJourneyDailyAggregate>
+        {
+            new()
+            {
+                CompletedJourneyCount = 2, FirstTimeRightCount = 1, ExcludedSessionCount = 1,
+                AttemptHistogram = new List<DeviceJourneyAttemptBucket>
+                {
+                    new() { Attempts = 1, JourneyCount = 1 },
+                    new() { Attempts = 2, JourneyCount = 1 },
+                },
+            },
+            new()
+            {
+                CompletedJourneyCount = 1, FirstTimeRightCount = 1, ExcludedSessionCount = 0,
+                AttemptHistogram = new List<DeviceJourneyAttemptBucket> { new() { Attempts = 1, JourneyCount = 1 } },
+            },
+        };
+
+        var totals = DeviceJourneyMetricsResponse.SumAggregates(daily);
+
+        Assert.Equal(3, totals.CompletedJourneys);
+        Assert.Equal(2, totals.FirstTimeRight);
+        Assert.Equal(66.7, totals.FtrRatePct);
+        Assert.Equal(1, totals.ExcludedSessions);
+        Assert.Equal(2, totals.AttemptHistogram.Count);
+        Assert.Equal(2, totals.AttemptHistogram.Single(b => b.Attempts == 1).JourneyCount);
+        Assert.Equal(1, totals.AttemptHistogram.Single(b => b.Attempts == 2).JourneyCount);
+    }
+
+    [Fact]
+    public void SumAggregates_NoCompletedJourneys_RateIsNull_NeverZero()
+    {
+        var totals = DeviceJourneyMetricsResponse.SumAggregates(new List<DeviceJourneyDailyAggregate>
+        {
+            new() { CompletedJourneyCount = 0, FirstTimeRightCount = 0, ExcludedSessionCount = 3 },
+        });
+
+        Assert.Equal(0, totals.CompletedJourneys);
+        Assert.Null(totals.FtrRatePct);
+        Assert.Equal(3, totals.ExcludedSessions);
+    }
+
+    [Theory]
+    [InlineData(null, 30)]     // default window
+    [InlineData("garbage", 30)]
+    [InlineData("7", 7)]
+    [InlineData("0", 1)]       // clamped low
+    [InlineData("999", 180)]   // clamped to aggregate retention
+    public void ClampDays_DefaultsAndClamps(string? raw, int expected)
+    {
+        Assert.Equal(expected, DeviceJourneyMetricsResponse.ClampDays(raw));
+    }
+
+    [Fact]
+    public void InclusiveWindowStart_YieldsExactlyNCalendarDays()
+    {
+        // Codex review: both range ends are inclusive, so "today - days" returned N+1 keys —
+        // days=1 summed yesterday AND today.
+        var today = new DateTime(2026, 7, 27);
+        Assert.Equal(today, DeviceJourneyMetricsResponse.InclusiveWindowStart(today, 1));
+        Assert.Equal(today.AddDays(-29), DeviceJourneyMetricsResponse.InclusiveWindowStart(today, 30));
+    }
+
+    private static DeviceHistory RepeaterHistory(
+        string serial, int attempts, params DeviceSessionRef[] chain)
+        => new()
+        {
+            TenantId = TenantA,
+            SerialKey = serial,
+            SerialNumber = serial.ToUpperInvariant(),
+            Chain = chain.ToList(),
+            CurrentJourneyAttempts = attempts,
+            JourneyCount = 1,
+        };
+
+    [Fact]
+    public void SelectRepeatDevices_FiltersGate_WindowsOnNewestRef_AndPicksNewestFailure()
+    {
+        var histories = new List<DeviceHistory>
+        {
+            // Single attempt — not a repeat device.
+            RepeaterHistory("clean-1", attempts: 1, Ref("c1", SessionStatus.Succeeded, T0)),
+            // Repeat device, active in window: newest failed ref carries the reason lookup.
+            RepeaterHistory("hot-1", attempts: 3,
+                Ref("h1", SessionStatus.Failed, T0.AddDays(-2)),
+                Ref("h2", SessionStatus.Failed, T0.AddDays(-1)),
+                Ref("h3", SessionStatus.Incomplete, T0)),
+            // Repeat device, but its pain is OLDER than the window — current lists only.
+            RepeaterHistory("old-1", attempts: 4,
+                Ref("o1", SessionStatus.Failed, T0.AddDays(-40))),
+            // Lower attempt count → sorts after hot-1.
+            RepeaterHistory("warm-1", attempts: 2,
+                Ref("w1", SessionStatus.Failed, T0.AddDays(-3)),
+                Ref("w2", SessionStatus.Succeeded, T0.AddDays(-2))),
+        };
+
+        var selected = DeviceJourneyMetricsResponse.SelectRepeatDevices(histories, T0.AddDays(-30));
+
+        Assert.Equal(2, selected.Count);
+        Assert.Equal("hot-1", selected[0].History.SerialKey);   // attempts desc
+        Assert.Equal("h3", selected[0].Newest.SessionId);
+        Assert.Equal("h2", selected[0].NewestFailed!.SessionId); // last FAILED, not last ref
+        Assert.Equal("warm-1", selected[1].History.SerialKey);
+    }
+
+    [Fact]
+    public void SelectRepeatDevices_CapsAtTen_ByAttemptsThenRecency()
+    {
+        var histories = new List<DeviceHistory>();
+        for (var i = 0; i < 12; i++)
+        {
+            histories.Add(RepeaterHistory($"dev-{i:D2}", attempts: 2 + i,
+                Ref($"r-{i}", SessionStatus.Failed, T0.AddDays(-i))));
+        }
+
+        var selected = DeviceJourneyMetricsResponse.SelectRepeatDevices(histories, T0.AddDays(-30));
+
+        Assert.Equal(10, selected.Count);
+        Assert.Equal("dev-11", selected[0].History.SerialKey); // highest attempts first
+        Assert.DoesNotContain(selected, c => c.History.SerialKey == "dev-00");
+        Assert.DoesNotContain(selected, c => c.History.SerialKey == "dev-01");
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/DiagnosticsBlobCascadeDeleterTests.cs

@@ -154,6 +154,80 @@ public class DiagnosticsBlobCascadeDeleterTests
         Assert.Equal(0, deleter.HostedDeleteCalls);
     }
 
+    // ── REAL DeleteAsync, Hosted branch — prefix sweep + belt-and-braces name delete ──
+    // The RecordingDeleter above replaces DeleteAsync wholesale, so these run the REAL
+    // base implementation against a fake storage service to pin the sweep behaviour.
+
+    private const string SessionGuid = "aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa";
+
+    [Fact]
+    public async Task RealDeleteAsync_Hosted_SweepsSessionPrefix_AndDeletesManifestBlob()
+    {
+        var storage = new RecordingHostedStorage();
+        var deleter = RealDeleter(storage);
+        var manifest = ManifestWith(blobName: $"{TenantA}/{Filename}", destination: "Hosted");
+        manifest.SessionId = SessionGuid;
+
+        var outcome = await deleter.DeleteAsync(manifest);
+
+        Assert.Equal(DiagnosticsBlobDeleteOutcome.HostedDeleted, outcome);
+        Assert.Equal((TenantA, SessionGuid), storage.SweptSession);
+        Assert.Equal($"{TenantA}/{Filename}", storage.DeletedBlobPath);
+    }
+
+    [Fact]
+    public async Task RealDeleteAsync_Hosted_NonGuidSessionId_SkipsSweep_StillDeletesManifestBlob()
+    {
+        // Legacy manifests may carry a malformed/empty SessionId — the sweep must degrade
+        // to the single-blob delete, never throw into the cascade poison path.
+        var storage = new RecordingHostedStorage();
+        var deleter = RealDeleter(storage);
+        var manifest = ManifestWith(blobName: $"{TenantA}/{Filename}", destination: "Hosted");
+        manifest.SessionId = "not-a-guid";
+
+        var outcome = await deleter.DeleteAsync(manifest);
+
+        Assert.Equal(DiagnosticsBlobDeleteOutcome.HostedDeleted, outcome);
+        Assert.Null(storage.SweptSession);
+        Assert.Equal($"{TenantA}/{Filename}", storage.DeletedBlobPath);
+    }
+
+    private static DiagnosticsBlobCascadeDeleter RealDeleter(RecordingHostedStorage storage)
+        // TenantConfigurationService is only consulted on the CustomerSas branch; the
+        // Hosted-branch tests never reach it, so null keeps the harness Azure-free.
+        => new(
+            storage,
+            tenantConfig: null!,
+            Microsoft.Extensions.Logging.Abstractions.NullLogger<DiagnosticsBlobCascadeDeleter>.Instance);
+
+    private sealed class RecordingHostedStorage : HostedDiagnosticsBlobService
+    {
+        public (string TenantId, string SessionId)? SweptSession { get; private set; }
+        public string? DeletedBlobPath { get; private set; }
+
+        public RecordingHostedStorage()
+            : base(
+                new Azure.Storage.Blobs.BlobServiceClient("UseDevelopmentStorage=true"),
+                Microsoft.Extensions.Logging.Abstractions.NullLogger<HostedDiagnosticsBlobService>.Instance,
+                usesManagedIdentity: false)
+        {
+        }
+
+        public override Task<int> DeleteBySessionPrefixAsync(
+            string tenantId, string sessionId, CancellationToken cancellationToken = default)
+        {
+            SweptSession = (tenantId, sessionId);
+            return Task.FromResult(1);
+        }
+
+        public override Task<bool> DeleteIfExistsAsync(
+            string blobPath, CancellationToken cancellationToken = default)
+        {
+            DeletedBlobPath = blobPath;
+            return Task.FromResult(true);
+        }
+    }
+
     // ── BuildBlobUrl helper (CustomerSas URL construction) ─────────────────────────
 
     [Fact]

src/Backend/AutopilotMonitor.Functions.Tests/EndpointPolicyCatalogCompletenessTests.cs

@@ -160,9 +160,9 @@ public class EndpointPolicyCatalogCompletenessTests
 
         Assert.NotNull(getConfig);
         Assert.NotNull(putConfig);
-        // GET config read is admin-tier read (own-tenant Admin / GA / read-only GlobalReader);
-        // PUT config write stays Admin/GA only.
-        Assert.Equal(EndpointPolicy.TenantAdminOrGlobalReader, getConfig.Policy);
+        // GET config read is member-tier (Operators/Viewers get the redacted read-only Settings
+        // view; the handler redacts for every non-admin caller); PUT config write stays Admin/GA only.
+        Assert.Equal(EndpointPolicy.MemberRead, getConfig.Policy);
         Assert.Equal(EndpointPolicy.TenantAdminOrGA, putConfig.Policy);
 
         var getRules = EndpointAccessPolicyCatalog.FindPolicy("GET", "/api/rules/gather");
@@ -338,7 +338,7 @@ public class EndpointPolicyCatalogCompletenessTests
     [InlineData("POST", "/api/agent/telemetry",          true)]   // DeviceOrBootstrapAuth
     [InlineData("GET",  "/api/bootstrap/validate/ABC123", true)]  // PublicAnonymous (param route)
     [InlineData("GET",  "/api/sessions",                 false)]  // MemberRead
-    [InlineData("GET",  "/api/config/00000000-0000-0000-0000-000000000001", false)] // TenantAdminOrGlobalReader
+    [InlineData("GET",  "/api/config/00000000-0000-0000-0000-000000000001", false)] // MemberRead
     [InlineData("GET",  "/api/global/sessions",          false)]  // GlobalReadOrAdmin
     [InlineData("GET",  "/api/nonexistent",              false)]  // unregistered → fail-closed
     public void SkipsJwtValidation_MatchesPolicyTier(string method, string path, bool expectedExempt)

src/Backend/AutopilotMonitor.Functions.Tests/FleetHealthPayloadTests.cs

@@ -10,9 +10,10 @@ namespace AutopilotMonitor.Functions.Tests;
 /// <summary>
 /// Covers the server-side Fleet Health aggregation in
 /// <see cref="MetricsMath.BuildFleetHealthPayload"/>, the single source of truth shared by the
-/// tenant (metrics/fleet-health) and global (global/metrics/fleet-health) functions. These assert
-/// the semantics ported verbatim from the previous client-side aggregation — notably success rate
-/// over ALL sessions and average duration over non-in-progress sessions (failures included).
+/// tenant (metrics/fleet-health) and global (global/metrics/fleet-health) functions. Success and
+/// failure rates follow the SLA convention — over finished enrollments (Succeeded + Failed) only,
+/// so in-flight sessions never dilute them. Average duration stays over non-in-progress sessions
+/// that carry a duration (failures included).
 /// </summary>
 public class FleetHealthPayloadTests
 {
@@ -55,7 +56,7 @@ public class FleetHealthPayloadTests
     }
 
     [Fact]
-    public void SuccessRate_IsOverAllSessions_NotJustTerminal()
+    public void SuccessRate_IsOverFinishedSessions_InFlightAndIncompleteExcluded()
     {
         var sessions = new List<SessionSummary>
         {
@@ -63,16 +64,34 @@ public class FleetHealthPayloadTests
             S(SessionStatus.Succeeded),
             S(SessionStatus.Failed),
             S(SessionStatus.InProgress),
+            S(SessionStatus.Pending),
+            S(SessionStatus.Incomplete), // terminal but non-failure — stays out of the rate
         };
 
         var payload = MetricsMath.BuildFleetHealthPayload(sessions, days: 7);
 
-        Assert.Equal(4, payload.Stats.Total);
+        Assert.Equal(6, payload.Stats.Total);
         Assert.Equal(2, payload.Stats.Succeeded);
         Assert.Equal(1, payload.Stats.Failed);
         Assert.Equal(1, payload.Stats.InProgress);
-        // 2 / 4 = 50% (over total), NOT 2 / 3 = 66.7% (over terminal).
-        Assert.Equal(50d, payload.Stats.SuccessRate);
+        Assert.Equal(1, payload.Stats.Incomplete);
+        // 2 / (2 + 1) = 66.7% (over finished), NOT 2 / 6 = 33.3% (over total).
+        Assert.Equal(66.7d, payload.Stats.SuccessRate);
+    }
+
+    [Fact]
+    public void SuccessRate_IsZero_WhenNothingFinishedYet()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S(SessionStatus.InProgress),
+            S(SessionStatus.Pending),
+        };
+
+        var payload = MetricsMath.BuildFleetHealthPayload(sessions, days: 7);
+
+        // No finished enrollments → no rate; clients render "—" off Succeeded + Failed == 0.
+        Assert.Equal(0d, payload.Stats.SuccessRate);
     }
 
     [Fact]
@@ -111,6 +130,27 @@ public class FleetHealthPayloadTests
         Assert.DoesNotContain(payload.ModelHealth, m => m.Model == "Contoso Model1");
     }
 
+    [Fact]
+    public void ModelHealth_CountsFailedSeparately_InFlightInTotalOnly()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S(SessionStatus.Succeeded, model: "X"),
+            S(SessionStatus.Failed, model: "X"),
+            S(SessionStatus.InProgress, model: "X"),
+        };
+
+        var payload = MetricsMath.BuildFleetHealthPayload(sessions, days: 7);
+
+        var m = Assert.Single(payload.ModelHealth);
+        // Total keeps the full device count; the client derives the rate from
+        // Succeeded / (Succeeded + Failed), so the in-flight session must land
+        // in Total but in neither outcome bucket.
+        Assert.Equal(3, m.Total);
+        Assert.Equal(1, m.Succeeded);
+        Assert.Equal(1, m.Failed);
+    }
+
     [Fact]
     public void ModelKey_FallsBackToUnknown_WhenManufacturerAndModelBlank()
     {
@@ -147,10 +187,13 @@ public class FleetHealthPayloadTests
     {
         var sessions = new List<SessionSummary>
         {
-            // Model A: 1 failed of 2 → 50%
+            // Model A: 1 failed / 2 finished → 50%; the 2 in-flight sessions widen Total
+            // but must not dilute the rate (over-total it would read 25%).
             S(SessionStatus.Failed, model: "A"),
             S(SessionStatus.Succeeded, model: "A"),
-            // Model B: 2 failed of 2 → 100%
+            S(SessionStatus.InProgress, model: "A"),
+            S(SessionStatus.Pending, model: "A"),
+            // Model B: 2 failed of 2 finished → 100%
             S(SessionStatus.Failed, model: "B"),
             S(SessionStatus.Failed, model: "B"),
             // Model C: no failures → excluded entirely
@@ -164,35 +207,37 @@ public class FleetHealthPayloadTests
         Assert.Equal(2, payload.TopFailingModels[0].Failed);
         Assert.Equal(100, payload.TopFailingModels[0].FailureRate);
         Assert.Equal("Contoso A", payload.TopFailingModels[1].Model);
+        Assert.Equal(4, payload.TopFailingModels[1].Total);
         Assert.Equal(50, payload.TopFailingModels[1].FailureRate);
         Assert.DoesNotContain(payload.TopFailingModels, m => m.Model == "Contoso C");
     }
 
     [Fact]
-    public void FailureReasons_TruncateLongReasons_AndTakeTopFive()
+    public void FailureReasons_GroupByPrefix_KeepFullestText_AndTakeTopFive()
     {
         var sessions = new List<SessionSummary>();
-        // 6 distinct reasons with descending frequency 6..1 → only top 5 kept.
-        for (int freq = 6; freq >= 1; freq--)
+        // 5 distinct short reasons with descending frequency 5..1.
+        for (int freq = 5; freq >= 1; freq--)
             for (int i = 0; i < freq; i++)
                 sessions.Add(S(SessionStatus.Failed, failureReason: $"Reason {freq}"));
-        // A null reason folds into "Unknown".
-        sessions.Add(S(SessionStatus.Failed, failureReason: null));
-        // An over-long reason is collapsed to a 50-char prefix + ellipsis.
-        var longReason = new string('x', 60);
-        sessions.Add(S(SessionStatus.Failed, failureReason: longReason));
+
+        // Two variants that share the first 50 characters collapse into one group;
+        // the longest variant is kept for display so the UI can show the full text.
+        var prefix = new string('x', 50);
+        var shortVariant = prefix + " short";
+        var longVariant = prefix + " a much longer tail that must survive intact";
+        for (int i = 0; i < 6; i++)
+            sessions.Add(S(SessionStatus.Failed, failureReason: shortVariant));
+        sessions.Add(S(SessionStatus.Failed, failureReason: longVariant));
 
         var payload = MetricsMath.BuildFleetHealthPayload(sessions, days: 30);
 
         Assert.Equal(5, payload.FailureReasons.Count);
-        Assert.Equal("Reason 6", payload.FailureReasons[0].Reason);
-        Assert.Equal(6, payload.FailureReasons[0].Count);
-        // The collapsed long reason exists in the tally (50 chars + "...").
-        var collapsed = new string('x', 50) + "...";
-        Assert.True(
-            sessions.Count(s => s.FailureReason == longReason) == 1,
-            "sanity: exactly one long-reason session");
-        Assert.Equal(53, collapsed.Length);
+        // The grouped cluster (6 + 1 = 7) outranks every "Reason N".
+        Assert.Equal(7, payload.FailureReasons[0].Count);
+        // The full, longest text survives end-to-end — no 50-char truncation.
+        Assert.Equal(longVariant, payload.FailureReasons[0].Reason);
+        Assert.DoesNotContain("...", payload.FailureReasons[0].Reason);
     }
 
     [Fact]

src/Backend/AutopilotMonitor.Functions.Tests/GeoMetricsProjectionEquivalenceTests.cs

@@ -50,6 +50,8 @@ public class GeoMetricsProjectionEquivalenceTests
             ["GeoCity"] = "Frankfurt",
             ["GeoLoc"] = "50.11,8.68",
             ["IsPreProvisioned"] = false,
+            ["AvgApiLatencyMs"] = 217.4,
+            ["ApiRequestCount"] = 270,
             // Noise the geo aggregation never reads.
             ["SerialNumber"] = "SN-FULL",
             ["Manufacturer"] = "Contoso",
@@ -76,6 +78,8 @@ public class GeoMetricsProjectionEquivalenceTests
         Assert.Equal(fromFull.GeoRegion, fromProjected.GeoRegion);
         Assert.Equal(fromFull.GeoCity, fromProjected.GeoCity);
         Assert.Equal(fromFull.GeoLoc, fromProjected.GeoLoc);
+        Assert.Equal(fromFull.AvgApiLatencyMs, fromProjected.AvgApiLatencyMs);
+        Assert.Equal(fromFull.ApiRequestCount, fromProjected.ApiRequestCount);
 
         // The grouping key must be identical too — it is composed from the Geo* fields.
         Assert.Equal(

src/Backend/AutopilotMonitor.Functions.Tests/GeographicMetricsApiLatencyTests.cs

@@ -0,0 +1,124 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using AutopilotMonitor.Functions.Functions.Metrics;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Pins the API-latency aggregation of
+/// <see cref="GetGeographicMetricsFunction.ComputeGeographicMetrics"/> (2026-07-28): both the
+/// per-location and the global figure are REQUEST-WEIGHTED averages —
+/// Σ(sessionAvg·requests)/Σ(requests) — not averages of session averages, so a long session
+/// with many uploads counts proportionally. Sessions without latency data (agents predating
+/// <c>AvgApiLatencyMs</c>) drop out of the weighting entirely.
+/// </summary>
+public class GeographicMetricsApiLatencyTests
+{
+    private static SessionSummary S(string country, string city, double? avgLatencyMs, int? requestCount)
+        => new()
+        {
+            TenantId = "00000000-0000-0000-0000-000000000fe0",
+            SessionId = Guid.NewGuid().ToString(),
+            Status = SessionStatus.Succeeded,
+            StartedAt = DateTime.UtcNow.AddHours(-2),
+            GeoCountry = country,
+            GeoRegion = country,
+            GeoCity = city,
+            GeoLoc = "0,0",
+            AvgApiLatencyMs = avgLatencyMs,
+            ApiRequestCount = requestCount,
+        };
+
+    [Fact]
+    public void LocationLatency_IsRequestWeighted_NotAverageOfAverages()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S("ID", "Jakarta", 1000, 300),
+            S("ID", "Jakarta", 500, 100),
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "country");
+
+        var loc = Assert.Single(result.Locations);
+        // (1000·300 + 500·100) / 400 = 875 — an unweighted mean would say 750.
+        Assert.Equal(875d, loc.AvgApiLatencyMs);
+        Assert.Equal(2, loc.ApiLatencySessionCount);
+    }
+
+    [Fact]
+    public void SessionsWithoutLatencyData_DropOutOfWeighting_ButStayInSessionCount()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S("DE", "Frankfurt", 200, 100),
+            S("DE", "Frankfurt", avgLatencyMs: null, requestCount: null), // pre-feature agent
+            S("DE", "Frankfurt", avgLatencyMs: 0, requestCount: 50),      // zero = no signal
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "country");
+
+        var loc = Assert.Single(result.Locations);
+        Assert.Equal(3, loc.SessionCount);
+        Assert.Equal(1, loc.ApiLatencySessionCount);
+        Assert.Equal(200d, loc.AvgApiLatencyMs);
+    }
+
+    [Fact]
+    public void NoLatencyDataAnywhere_YieldsZeroes_NotDivisionByZero()
+    {
+        var sessions = new List<SessionSummary> { S("DE", "Frankfurt", null, null) };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "country");
+
+        var loc = Assert.Single(result.Locations);
+        Assert.Equal(0d, loc.AvgApiLatencyMs);
+        Assert.Equal(0, loc.ApiLatencySessionCount);
+        Assert.Equal(0d, loc.ApiLatencyVsGlobalPct);
+        Assert.Equal(0d, result.GlobalAverages.AvgApiLatencyMs);
+    }
+
+    [Fact]
+    public void GlobalLatency_WeightsAcrossLocations_AndVsGlobalPctCompares()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S("DE", "Frankfurt", 200, 100),
+            S("ID", "Jakarta", 800, 100),
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "country");
+
+        // Equal request weights → global = (200 + 800) / 2 = 500.
+        Assert.Equal(500d, result.GlobalAverages.AvgApiLatencyMs);
+
+        var de = result.Locations.Single(l => l.Country == "DE");
+        var id = result.Locations.Single(l => l.Country == "ID");
+        Assert.Equal(-60d, de.ApiLatencyVsGlobalPct); // 200 vs 500
+        Assert.Equal(60d, id.ApiLatencyVsGlobalPct);  // 800 vs 500
+    }
+
+    [Fact]
+    public void LocationWithoutData_KeepsZeroPct_EvenWhenGlobalHasData()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S("DE", "Frankfurt", 200, 100),
+            S("FR", "Paris", null, null),
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "country");
+
+        var fr = result.Locations.Single(l => l.Country == "FR");
+        Assert.Equal(0d, fr.AvgApiLatencyMs);
+        Assert.Equal(0d, fr.ApiLatencyVsGlobalPct); // no data ≠ "0 ms fast"
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/GeographicMetricsSuccessRateTests.cs

@@ -0,0 +1,99 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using AutopilotMonitor.Functions.Functions.Metrics;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Pins the success-rate semantics of the geographic aggregation
+/// (<see cref="GetGeographicMetricsFunction.ComputeGeographicMetrics"/>): the rate is an outcome
+/// quota over finished enrollments (Succeeded + Failed) following the SLA convention. In-flight
+/// sessions (InProgress/Pending/Stalled/AwaitingUser) and Incomplete (terminal, non-failure) sit
+/// in SessionCount but never dilute the rate — the customer-visible symptom this pins against was
+/// a site mid-rollout showing a poor rate despite zero actual failures.
+/// </summary>
+public class GeographicMetricsSuccessRateTests
+{
+    private static SessionSummary S(SessionStatus status, string city = "Frankfurt")
+        => new()
+        {
+            TenantId = "00000000-0000-0000-0000-000000000fe0",
+            SessionId = Guid.NewGuid().ToString(),
+            Status = status,
+            StartedAt = DateTime.UtcNow.AddHours(-2),
+            GeoCountry = "DE",
+            GeoRegion = "Hessen",
+            GeoCity = city,
+            GeoLoc = "50.11,8.68",
+        };
+
+    [Fact]
+    public void SuccessRate_IsOverFinishedSessions_InFlightAndIncompleteExcluded()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S(SessionStatus.Succeeded),
+            S(SessionStatus.Succeeded),
+            S(SessionStatus.Succeeded),
+            S(SessionStatus.Failed),
+            S(SessionStatus.InProgress),
+            S(SessionStatus.Pending),
+            S(SessionStatus.AwaitingUser),
+            S(SessionStatus.Incomplete), // terminal but non-failure — stays out of the rate
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "city");
+
+        var loc = Assert.Single(result.Locations);
+        Assert.Equal(8, loc.SessionCount);
+        Assert.Equal(3, loc.Succeeded);
+        Assert.Equal(1, loc.Failed);
+        // 3 / (3 + 1) = 75% (over finished), NOT 3 / 8 = 37.5% (over all sessions).
+        Assert.Equal(75d, loc.SuccessRate);
+    }
+
+    [Fact]
+    public void SuccessRate_IsZero_WhenNothingFinishedYet()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            S(SessionStatus.InProgress),
+            S(SessionStatus.Pending),
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "city");
+
+        var loc = Assert.Single(result.Locations);
+        // No finished enrollments → no rate; the frontend renders "—" off Succeeded + Failed == 0.
+        Assert.Equal(0d, loc.SuccessRate);
+        Assert.Equal(0, loc.Succeeded + loc.Failed);
+        Assert.Equal(2, loc.SessionCount);
+    }
+
+    [Fact]
+    public void SuccessRate_IsPerLocation_NotGlobal()
+    {
+        var sessions = new List<SessionSummary>
+        {
+            // Frankfurt: 1 / 1 finished → 100% despite the in-flight session.
+            S(SessionStatus.Succeeded, city: "Frankfurt"),
+            S(SessionStatus.InProgress, city: "Frankfurt"),
+            // Berlin: 1 / 2 finished → 50%.
+            S(SessionStatus.Succeeded, city: "Berlin"),
+            S(SessionStatus.Failed, city: "Berlin"),
+        };
+
+        var result = GetGeographicMetricsFunction.ComputeGeographicMetrics(
+            sessions, new List<AppInstallSummary>(), "city");
+
+        var frankfurt = result.Locations.Single(l => l.City == "Frankfurt");
+        var berlin = result.Locations.Single(l => l.City == "Berlin");
+        Assert.Equal(100d, frankfurt.SuccessRate);
+        Assert.Equal(50d, berlin.SuccessRate);
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/GetImeVersionHistoryRedactionTests.cs

@@ -0,0 +1,189 @@
+using System.Text.Json;
+using AutopilotMonitor.Functions.Functions.Metrics;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Guards the cross-tenant redaction on GET metrics/ime-versions.
+///
+/// The route is <c>MemberRead</c> (EndpointAccessPolicyCatalog) and the archive itself is a
+/// single global partition — every tenant member reads the SAME platform-wide rows. The only
+/// thing standing between an ordinary tenant user and another tenant's identifiers is the
+/// projection in <see cref="GetImeVersionHistoryFunction.BuildResponsePayload"/>. That made it
+/// worth pinning: the redaction had no test at all, and a boundary nobody asserts is one a
+/// later refactor can open without a single failure.
+///
+/// The JSON tests are the ones that matter — a property dropped from an anonymous type is only
+/// a real redaction once it is also absent from the wire, and the wire is what the customer's
+/// MCP client receives. camelCase mirrors Program.cs's PropertyNamingPolicy.
+/// </summary>
+public class GetImeVersionHistoryRedactionTests
+{
+    private static readonly DateTime First = new(2026, 7, 8, 3, 18, 13, DateTimeKind.Utc);
+    private static readonly DateTime Last = new(2026, 7, 27, 16, 59, 34, DateTimeKind.Utc);
+
+    private static readonly JsonSerializerOptions CamelCase =
+        new() { PropertyNamingPolicy = JsonNamingPolicy.CamelCase };
+
+    // =========================================================================
+    // Non-global caller (ordinary tenant member) — the redaction
+    // =========================================================================
+
+    [Fact]
+    public void NonGlobalCaller_DropsForeignTenantAndSessionIds()
+    {
+        var json = SerializeFor(hasGlobalScope: false, MakeEntry());
+
+        Assert.DoesNotContain("firstSeenTenantId", json, StringComparison.OrdinalIgnoreCase);
+        Assert.DoesNotContain("firstSeenSessionId", json, StringComparison.OrdinalIgnoreCase);
+    }
+
+    /// <summary>
+    /// The values, not just the property names: a rename of the model property must not let the
+    /// identifier survive under a different key. Asserting on the GUIDs catches that.
+    /// </summary>
+    [Fact]
+    public void NonGlobalCaller_ForeignIdentifierValuesAppearNowhereInThePayload()
+    {
+        var json = SerializeFor(hasGlobalScope: false, MakeEntry());
+
+        Assert.DoesNotContain("57f34dd1-6a42-47f9-9bc0-b921fa6caa30", json, StringComparison.OrdinalIgnoreCase);
+        Assert.DoesNotContain("baae0453-6d19-47b5-aa58-8bed5a572fc8", json, StringComparison.OrdinalIgnoreCase);
+    }
+
+    [Fact]
+    public void NonGlobalCaller_ExposesExactlyTheFourAllowedProperties()
+    {
+        using var doc = JsonDocument.Parse(SerializeFor(hasGlobalScope: false, MakeEntry()));
+
+        var keys = doc.RootElement[0].EnumerateObject().Select(p => p.Name).OrderBy(n => n).ToArray();
+
+        Assert.Equal(new[] { "firstSeenAt", "lastSeenAt", "sessionCount", "version" }, keys);
+    }
+
+    /// <summary>
+    /// SessionCount stays on purpose — it is a platform-wide total, and the route exists so any
+    /// tenant member can track Microsoft's IME rollout. Pinned so that "the counts are global"
+    /// stays a decision someone made rather than something that quietly drifts either way.
+    /// </summary>
+    [Fact]
+    public void NonGlobalCaller_KeepsVersionDatesAndGlobalSessionCount()
+    {
+        using var doc = JsonDocument.Parse(SerializeFor(hasGlobalScope: false, MakeEntry()));
+
+        var row = doc.RootElement[0];
+        Assert.Equal("1.103.101.0", row.GetProperty("version").GetString());
+        Assert.Equal(First, row.GetProperty("firstSeenAt").GetDateTime());
+        Assert.Equal(Last, row.GetProperty("lastSeenAt").GetDateTime());
+        Assert.Equal(3109, row.GetProperty("sessionCount").GetInt32());
+    }
+
+    // =========================================================================
+    // Global caller (Global Admin or Global Reader) — full archive
+    // =========================================================================
+
+    [Fact]
+    public void GlobalCaller_KeepsTenantAndSessionIds()
+    {
+        using var doc = JsonDocument.Parse(SerializeFor(hasGlobalScope: true, MakeEntry()));
+
+        var row = doc.RootElement[0];
+        Assert.Equal("57f34dd1-6a42-47f9-9bc0-b921fa6caa30", row.GetProperty("firstSeenTenantId").GetString());
+        Assert.Equal("baae0453-6d19-47b5-aa58-8bed5a572fc8", row.GetProperty("firstSeenSessionId").GetString());
+    }
+
+    [Fact]
+    public void GlobalCaller_ExposesTheFullEntry()
+    {
+        using var doc = JsonDocument.Parse(SerializeFor(hasGlobalScope: true, MakeEntry()));
+
+        var keys = doc.RootElement[0].EnumerateObject().Select(p => p.Name).OrderBy(n => n).ToArray();
+
+        Assert.Equal(
+            new[] { "firstSeenAt", "firstSeenSessionId", "firstSeenTenantId", "lastSeenAt", "sessionCount", "version" },
+            keys);
+    }
+
+    // =========================================================================
+    // Shape-preserving properties of the projection
+    // =========================================================================
+
+    [Fact]
+    public void BothScopes_EmitAJsonArray_EvenForASingleRow()
+    {
+        foreach (var scope in new[] { true, false })
+        {
+            using var doc = JsonDocument.Parse(SerializeFor(scope, MakeEntry()));
+            Assert.Equal(JsonValueKind.Array, doc.RootElement.ValueKind);
+            Assert.Equal(1, doc.RootElement.GetArrayLength());
+        }
+    }
+
+    [Fact]
+    public void BothScopes_EmptyArchive_SerializesAsEmptyArray()
+    {
+        foreach (var scope in new[] { true, false })
+        {
+            using var doc = JsonDocument.Parse(SerializeFor(scope));
+            Assert.Equal(JsonValueKind.Array, doc.RootElement.ValueKind);
+            Assert.Equal(0, doc.RootElement.GetArrayLength());
+        }
+    }
+
+    /// <summary>
+    /// The repository orders by FirstSeenAt descending; the projection must not reshuffle, or a
+    /// redacted caller would see a different sequence than a Global Admin for the same archive.
+    /// </summary>
+    [Fact]
+    public void NonGlobalCaller_PreservesRowOrderAndCount()
+    {
+        var newest = MakeEntry("1.103.101.0", First);
+        var older = MakeEntry("1.101.111.0", First.AddDays(-48));
+        var oldest = MakeEntry("1.99.101.0", First.AddDays(-131));
+
+        using var doc = JsonDocument.Parse(SerializeFor(false, newest, older, oldest));
+
+        Assert.Equal(3, doc.RootElement.GetArrayLength());
+        Assert.Equal("1.103.101.0", doc.RootElement[0].GetProperty("version").GetString());
+        Assert.Equal("1.101.111.0", doc.RootElement[1].GetProperty("version").GetString());
+        Assert.Equal("1.99.101.0", doc.RootElement[2].GetProperty("version").GetString());
+    }
+
+    /// <summary>
+    /// BuildResponsePayload returns `object`; the branch that keeps everything hands back the
+    /// entries themselves. System.Text.Json resolves an `object` declared type via the runtime
+    /// type, which is what keeps the wire format identical to the pre-extraction inline code —
+    /// assert it rather than trust it, since a regression here would silently emit "{}" rows.
+    /// </summary>
+    [Fact]
+    public void GlobalPayload_SerializedAsObject_StillEmitsPopulatedRows()
+    {
+        var payload = GetImeVersionHistoryFunction.BuildResponsePayload([MakeEntry()], hasGlobalScope: true);
+
+        var json = JsonSerializer.Serialize(payload, CamelCase);
+
+        Assert.Contains("1.103.101.0", json, StringComparison.Ordinal);
+        Assert.NotEqual("[{}]", json);
+    }
+
+    // =========================================================================
+    // Helpers
+    // =========================================================================
+
+    private static string SerializeFor(bool hasGlobalScope, params ImeVersionHistoryEntry[] versions) =>
+        JsonSerializer.Serialize(
+            GetImeVersionHistoryFunction.BuildResponsePayload(versions, hasGlobalScope),
+            CamelCase);
+
+    private static ImeVersionHistoryEntry MakeEntry(string version = "1.103.101.0", DateTime? firstSeenAt = null) =>
+        new()
+        {
+            Version = version,
+            FirstSeenAt = firstSeenAt ?? First,
+            FirstSeenSessionId = "baae0453-6d19-47b5-aa58-8bed5a572fc8",
+            FirstSeenTenantId = "57f34dd1-6a42-47f9-9bc0-b921fa6caa30",
+            LastSeenAt = Last,
+            SessionCount = 3109,
+        };
+}

src/Backend/AutopilotMonitor.Functions.Tests/GetTenantFeatureFlagsPayloadTests.cs

@@ -35,6 +35,7 @@ public class GetTenantFeatureFlagsPayloadTests
         Assert.Equal(new[]
         {
             "bootstrapTokenEnabled",
+            "diagnosticsUploadConfigured",
             "edition",
             "enableIntegrityBypassAnalyzer",
             "enableSoftwareInventoryAnalyzer",
@@ -114,6 +115,35 @@ public class GetTenantFeatureFlagsPayloadTests
         Assert.DoesNotContain("webhookUrl", json, StringComparison.OrdinalIgnoreCase);
     }
 
+    // ── diagnosticsUploadConfigured (drives the session-detail "Collect Logs" button) ──
+
+    [Theory]
+    // mode Off (or unset) → never configured, regardless of destination
+    [InlineData(null, null, null, false)]
+    [InlineData("Off", "https://x.blob.core.windows.net/c?sv=1", "CustomerSas", false)]
+    [InlineData("Off", null, "Hosted", false)]
+    // mode active + Hosted destination → configured without any SAS
+    [InlineData("OnFailure", null, "Hosted", true)]
+    [InlineData("Always", null, "Hosted", true)]
+    // mode active + CustomerSas → configured only when a SAS URL is present
+    [InlineData("OnFailure", "https://x.blob.core.windows.net/c?sv=1", "CustomerSas", true)]
+    [InlineData("OnFailure", null, "CustomerSas", false)]
+    [InlineData("Always", "", null, false)]
+    public void Payload_DiagnosticsUploadConfigured_ModeAndDestinationMatrix(
+        string? mode, string? sasUrl, string? destination, bool expected)
+    {
+        // Null-forgiving: the model declares these non-nullable with defaults, but table
+        // round-trips can produce null at runtime — the payload must tolerate that shape.
+        var element = Serialize(new TenantConfiguration
+        {
+            DiagnosticsUploadMode = mode!,
+            DiagnosticsBlobSasUrl = sasUrl!,
+            DiagnosticsUploadDestination = destination!,
+        });
+
+        Assert.Equal(expected, element.GetProperty("diagnosticsUploadConfigured").GetBoolean());
+    }
+
     // ── Edition / trial surface ─────────────────────────────────────────────
 
     [Fact]

src/Backend/AutopilotMonitor.Functions.Tests/GetTpmPssUnsupportedFunctionTests.cs

@@ -0,0 +1,194 @@
+using AutopilotMonitor.Functions.Functions.Reports;
+using AutopilotMonitor.Shared.DataAccess;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Tests for the TPM-PSS-unsupported aggregation logic extracted from
+/// GetTpmPssUnsupportedFunction. All tests target the pure static BuildAggregatedResult
+/// method — no HTTP or DI needed. Aggregation is by serial number (the TPM incompatibility
+/// is a per-device condition, not a per-model one — firmware levels differ within a model).
+/// </summary>
+public class GetTpmPssUnsupportedFunctionTests
+{
+    private static readonly DateTime T0 = new(2026, 3, 1, 12, 0, 0, DateTimeKind.Utc);
+
+    // =========================================================================
+    // Empty / no matching reports
+    // =========================================================================
+
+    [Fact]
+    public void BuildAggregatedResult_EmptyList_ReturnsEmpty()
+    {
+        var (aggregated, totalRawReports) = GetTpmPssUnsupportedFunction.BuildAggregatedResult([]);
+
+        Assert.Empty(aggregated);
+        Assert.Equal(0, totalRawReports);
+    }
+
+    [Fact]
+    public void BuildAggregatedResult_FiltersTpmPssUnsupportedOnly()
+    {
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", T0),
+            MakeEntry("DeviceNotRegistered", "Lenovo", "ThinkPad T14", "SN2", T0.AddMinutes(1)),
+            MakeEntry("AuthCertificateRejected", "HP", "EliteBook", "SN3", T0.AddMinutes(2)),
+        };
+
+        var (aggregated, totalRawReports) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        Assert.Single(aggregated);
+        Assert.Equal(1, totalRawReports);
+    }
+
+    // =========================================================================
+    // Aggregation by serial number
+    // =========================================================================
+
+    [Fact]
+    public void BuildAggregatedResult_AggregatesBySerialCaseInsensitive()
+    {
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "S4SQ8685", T0),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "s4sq8685", T0.AddMinutes(1)),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", " S4SQ8685 ", T0.AddMinutes(2)),
+        };
+
+        var (aggregated, totalRawReports) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        Assert.Single(aggregated);
+        Assert.Equal(3, totalRawReports);
+        var item = ToDynamic(aggregated[0]);
+        Assert.Equal(3, (int)item.attemptCount);
+    }
+
+    [Fact]
+    public void BuildAggregatedResult_DistinctSerials_ProduceDistinctRows()
+    {
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", T0),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN2", T0.AddMinutes(1)),
+            MakeEntry("TpmPssUnsupported", "Microsoft", "Surface Book", "SN3", T0.AddMinutes(2)),
+        };
+
+        var (aggregated, totalRawReports) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        Assert.Equal(3, aggregated.Count);
+        Assert.Equal(3, totalRawReports);
+    }
+
+    [Fact]
+    public void BuildAggregatedResult_UsesMostRecentManufacturerModel()
+    {
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "OldMfr", "OldModel", "SN1", T0),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", T0.AddMinutes(5)),
+        };
+
+        var (aggregated, _) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        var item = ToDynamic(aggregated[0]);
+        Assert.Equal("Lenovo", (string)item.manufacturer);
+        Assert.Equal("ThinkCentre M720q", (string)item.model);
+        Assert.Equal("SN1", (string)item.serialNumber);
+    }
+
+    // =========================================================================
+    // Timestamps
+    // =========================================================================
+
+    [Fact]
+    public void BuildAggregatedResult_FirstSeenLastSeen_Correct()
+    {
+        var t1 = T0;
+        var t2 = T0.AddHours(1);
+        var t3 = T0.AddHours(2);
+
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", t2),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", t1),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", t3),
+        };
+
+        var (aggregated, _) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        var item = ToDynamic(aggregated[0]);
+        Assert.Equal(t1, (DateTime)item.firstSeen);
+        Assert.Equal(t3, (DateTime)item.lastSeen);
+    }
+
+    [Fact]
+    public void BuildAggregatedResult_OrderedByLastSeenDescending()
+    {
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "SN1", T0),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkPad X13", "SN2", T0.AddHours(2)),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720s", "SN3", T0.AddHours(1)),
+        };
+
+        var (aggregated, _) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        Assert.Equal(3, aggregated.Count);
+        var first = ToDynamic(aggregated[0]);
+        var second = ToDynamic(aggregated[1]);
+        var third = ToDynamic(aggregated[2]);
+
+        Assert.True((DateTime)first.lastSeen > (DateTime)second.lastSeen);
+        Assert.True((DateTime)second.lastSeen > (DateTime)third.lastSeen);
+    }
+
+    // =========================================================================
+    // Null / empty serial number
+    // =========================================================================
+
+    [Fact]
+    public void BuildAggregatedResult_NullOrEmptySerials_GroupedIntoSingleBucket()
+    {
+        var reports = new List<DistressReportEntry>
+        {
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", null, T0),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "", T0.AddMinutes(1)),
+            MakeEntry("TpmPssUnsupported", "Lenovo", "ThinkCentre M720q", "   ", T0.AddMinutes(2)),
+        };
+
+        var (aggregated, _) = GetTpmPssUnsupportedFunction.BuildAggregatedResult(reports);
+
+        Assert.Single(aggregated);
+        var item = ToDynamic(aggregated[0]);
+        Assert.Equal(3, (int)item.attemptCount);
+    }
+
+    // =========================================================================
+    // Helpers
+    // =========================================================================
+
+    private static DistressReportEntry MakeEntry(
+        string errorType, string? manufacturer, string? model, string? serial, DateTime ingestedAt) =>
+        new()
+        {
+            TenantId = "test-tenant",
+            ErrorType = errorType,
+            Manufacturer = manufacturer,
+            Model = model,
+            SerialNumber = serial,
+            IngestedAt = ingestedAt,
+            AgentTimestamp = ingestedAt,
+        };
+
+    private static dynamic ToDynamic(object obj)
+    {
+        var type = obj.GetType();
+        var dict = new System.Dynamic.ExpandoObject() as IDictionary<string, object?>;
+        foreach (var prop in type.GetProperties())
+        {
+            dict[prop.Name] = prop.GetValue(obj);
+        }
+        return dict;
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/HardwareRejectionNotificationTrackerDedupTests.cs

@@ -0,0 +1,360 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text.RegularExpressions;
+using System.Threading;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.DataAccess.TableStorage;
+using AutopilotMonitor.Functions.Services;
+using Azure;
+using Azure.Data.Tables;
+using Microsoft.Extensions.Logging.Abstractions;
+using Moq;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Behavioural coverage for <see cref="TableHardwareRejectionNotificationTracker"/> — the dedup
+/// store behind the two distress-driven bell notifications (hardware rejection per model,
+/// TPM-PSS incompatibility per serial). <see cref="HardwareRejectionNotificationTrackerKeyTests"/>
+/// pins the RowKey shape; this file pins what the tracker DOES with those keys.
+///
+/// CORRECTNESS GUARD: the "one bell ever" promise rests entirely on the 409 branch of
+/// <c>AddEntityAsync</c> — a rewrite to Upsert or a get-then-add would silently start ringing
+/// the bell on every distress report (a device with an incompatible TPM retries forever, so
+/// "every report" means a notification storm, not a second bell).
+///
+/// The retention sweep is covered here too because <c>DeleteOlderThanAsync</c> prunes BOTH key
+/// spaces from the shared table: pruning a <c>tpmpss|</c> row re-arms that device's bell.
+/// </summary>
+public class HardwareRejectionNotificationTrackerDedupTests
+{
+    private const string TenantId = "77777777-7777-7777-7777-777777777777";
+    private const string OtherTenantId = "88888888-8888-8888-8888-888888888888";
+    private const string Serial = "S4SQ8685";
+
+    // ── TPM PSS: first bell, then never again ──────────────────────────────────────
+
+    [Fact]
+    public async Task TpmPss_FirstReport_RegistersRowAndReturnsTrue()
+    {
+        var harness = new Harness();
+
+        var isFirst = await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial);
+
+        Assert.True(isFirst);
+        var entity = Assert.Single(harness.Store.Values);
+        Assert.Equal(TenantId.ToLowerInvariant(), entity.PartitionKey);
+        Assert.Equal("tpmpss|s4sq8685", entity.RowKey);
+        // Serialization guard: the row must carry the fields the cleanup sweep and any future
+        // operator query rely on — losing FirstNotifiedAt would make the row immortal.
+        Assert.Equal(TenantId, entity.GetString("TenantId"));
+        Assert.Equal(Serial, entity.GetString("SerialNumber"));
+        Assert.NotNull(entity.GetDateTime("FirstNotifiedAt"));
+    }
+
+    [Fact]
+    public async Task TpmPss_SecondReport_SameSerial_ReturnsFalse()
+    {
+        var harness = new Harness();
+
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.False(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.Single(harness.Store);
+    }
+
+    [Theory]
+    [InlineData("s4sq8685")]
+    [InlineData("  S4SQ8685  ")]
+    [InlineData("S4Sq8685")]
+    public async Task TpmPss_SecondReport_CasingOrWhitespaceVariant_StillDeduped(string variant)
+    {
+        var harness = new Harness();
+
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.False(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, variant));
+    }
+
+    [Fact]
+    public async Task TpmPss_DifferentSerial_OrDifferentTenant_RingsAgain()
+    {
+        var harness = new Harness();
+
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, "OTHER123"));
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(OtherTenantId, Serial));
+        Assert.Equal(3, harness.Store.Count);
+    }
+
+    [Theory]
+    [InlineData("", Serial)]
+    [InlineData("   ", Serial)]
+    [InlineData(TenantId, "")]
+    [InlineData(TenantId, "   ")]
+    public async Task TpmPss_BlankInput_ReturnsFalse_WithoutTouchingTheTable(string tenantId, string serial)
+    {
+        var harness = new Harness();
+
+        Assert.False(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(tenantId, serial));
+        Assert.Empty(harness.Store);
+        harness.Table.Verify(
+            c => c.AddEntityAsync(It.IsAny<TableEntity>(), It.IsAny<CancellationToken>()),
+            Times.Never);
+    }
+
+    [Fact]
+    public async Task TpmPss_UnexpectedTableFailure_ReturnsFalse()
+    {
+        // Fail-soft in the SUPPRESSING direction: the write may or may not have landed, so the
+        // caller must not fire a bell it might fire again on the next report.
+        var harness = new Harness { AddFailure = new RequestFailedException(503, "ServerBusy") };
+
+        Assert.False(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+    }
+
+    // ── Hardware rejection: same contract, disjoint key space ──────────────────────
+
+    [Fact]
+    public async Task HardwareRejection_SecondReport_SameModel_ReturnsFalse()
+    {
+        var harness = new Harness();
+
+        Assert.True(await harness.Sut.TryRegisterFirstNotificationAsync(TenantId, "Lenovo", "ThinkPad X1"));
+        Assert.False(await harness.Sut.TryRegisterFirstNotificationAsync(TenantId, "LENOVO", " thinkpad x1 "));
+        Assert.Single(harness.Store);
+    }
+
+    [Fact]
+    public async Task HardwareRejection_AndTpmPss_DoNotDedupEachOtherInTheSharedTable()
+    {
+        var harness = new Harness();
+
+        Assert.True(await harness.Sut.TryRegisterFirstNotificationAsync(TenantId, "Lenovo", Serial));
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.Equal(2, harness.Store.Count);
+    }
+
+    // ── F3 rule regression: one bell per episode, keyspaces stay independent ──────
+
+    [Fact]
+    public async Task RuleRegression_SecondRegister_SameRule_ReturnsFalse_OneBellPerEpisode()
+    {
+        var harness = new Harness();
+        var alert = new AutopilotMonitor.Shared.Models.RuleRegressionAlert
+        {
+            TenantId = TenantId, RuleId = "ANALYZE-NET-001",
+            FirstNotifiedAt = DateTime.UtcNow, LastEvaluatedAt = DateTime.UtcNow,
+        };
+
+        Assert.True(await harness.Sut.TryRegisterRuleRegressionAsync(TenantId, alert));
+        Assert.False(await harness.Sut.TryRegisterRuleRegressionAsync(TenantId, alert));
+
+        var entity = Assert.Single(harness.Store.Values);
+        Assert.Equal("ruleregression|analyze-net-001", entity.RowKey);
+        Assert.NotNull(entity.GetDateTime("FirstNotifiedAt")); // the retention sweep depends on it
+    }
+
+    [Fact]
+    public async Task RuleRegression_DeleteReArms_AndDoesNotTouchOtherKeyspaces()
+    {
+        var harness = new Harness();
+        var alert = new AutopilotMonitor.Shared.Models.RuleRegressionAlert
+        {
+            TenantId = TenantId, RuleId = "ANALYZE-NET-001",
+            FirstNotifiedAt = DateTime.UtcNow, LastEvaluatedAt = DateTime.UtcNow,
+        };
+        Assert.True(await harness.Sut.TryRegisterRuleRegressionAsync(TenantId, alert));
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+
+        await harness.Sut.DeleteRuleRegressionAsync(TenantId, "ANALYZE-NET-001");
+
+        // The episode re-arms; the unrelated tpmpss row survives.
+        Assert.Equal("tpmpss|s4sq8685", Assert.Single(harness.Store.Values).RowKey);
+        Assert.True(await harness.Sut.TryRegisterRuleRegressionAsync(TenantId, alert));
+    }
+
+    // ── Retention sweep ───────────────────────────────────────────────────────────
+
+    [Fact]
+    public async Task DeleteOlderThanAsync_PrunesRuleRegressionRows_ReArmingLongBurningEpisodes()
+    {
+        var harness = new Harness();
+        Assert.True(await harness.Sut.TryRegisterRuleRegressionAsync(TenantId,
+            new AutopilotMonitor.Shared.Models.RuleRegressionAlert
+            {
+                TenantId = TenantId, RuleId = "ANALYZE-NET-001",
+                FirstNotifiedAt = DateTime.UtcNow, LastEvaluatedAt = DateTime.UtcNow,
+            }));
+        harness.AgeRows(TimeSpan.FromDays(31));
+
+        // Spec §F3: retention cleanup re-arms — a month-old still-burning regression rings again.
+        Assert.Equal(1, await harness.Sut.DeleteOlderThanAsync(DateTime.UtcNow.AddDays(-30)));
+        Assert.Empty(harness.Store);
+    }
+
+    [Fact]
+    public async Task DeleteOlderThanAsync_PrunesTpmPssRows_AndReArmsThatDevicesBell()
+    {
+        var harness = new Harness();
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        harness.AgeRows(TimeSpan.FromDays(31));
+
+        var deleted = await harness.Sut.DeleteOlderThanAsync(DateTime.UtcNow.AddDays(-30));
+
+        Assert.Equal(1, deleted);
+        Assert.Empty(harness.Store);
+        // A device whose TPM was never fixed keeps reporting; after the retention window it is
+        // worth one more bell rather than permanent silence.
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+    }
+
+    [Fact]
+    public async Task DeleteOlderThanAsync_KeepsRowsInsideTheWindow_DedupSurvives()
+    {
+        var harness = new Harness();
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.True(await harness.Sut.TryRegisterFirstNotificationAsync(TenantId, "Lenovo", "ThinkPad X1"));
+
+        var deleted = await harness.Sut.DeleteOlderThanAsync(DateTime.UtcNow.AddDays(-30));
+
+        Assert.Equal(0, deleted);
+        Assert.Equal(2, harness.Store.Count);
+        Assert.False(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.False(await harness.Sut.TryRegisterFirstNotificationAsync(TenantId, "Lenovo", "ThinkPad X1"));
+    }
+
+    [Fact]
+    public async Task DeleteOlderThanAsync_FiltersOnFirstNotifiedAt_NotTimestamp()
+    {
+        // Rows are insert-once, so FirstNotifiedAt is the creation time. Timestamp would look
+        // equivalent but is rewritten by a table restore/copy, which would silently re-arm every
+        // bell in the tenant at once.
+        var harness = new Harness();
+        var cutoff = new DateTime(2026, 7, 25, 12, 0, 0, DateTimeKind.Utc);
+
+        await harness.Sut.DeleteOlderThanAsync(cutoff);
+
+        var filter = Assert.Single(harness.QueryFilters);
+        Assert.Equal("FirstNotifiedAt lt datetime'2026-07-25T12:00:00Z'", filter);
+    }
+
+    [Fact]
+    public async Task DeleteOlderThanAsync_ContinuesAfterASingleRowFailure()
+    {
+        var harness = new Harness();
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, Serial));
+        Assert.True(await harness.Sut.TryRegisterFirstTpmPssNotificationAsync(TenantId, "OTHER123"));
+        harness.AgeRows(TimeSpan.FromDays(31));
+        harness.FailDeleteForRowKeys.Add("tpmpss|s4sq8685");
+
+        var deleted = await harness.Sut.DeleteOlderThanAsync(DateTime.UtcNow.AddDays(-30));
+
+        Assert.Equal(1, deleted);
+        Assert.Equal("tpmpss|s4sq8685", Assert.Single(harness.Store.Values).RowKey);
+    }
+
+    [Fact]
+    public async Task DeleteOlderThanAsync_QueryFailure_ReturnsZero()
+    {
+        var harness = new Harness { QueryFailure = new RequestFailedException(500, "Boom") };
+
+        Assert.Equal(0, await harness.Sut.DeleteOlderThanAsync(DateTime.UtcNow.AddDays(-30)));
+    }
+
+    // ── Harness ───────────────────────────────────────────────────────────────────
+
+    private sealed class Harness
+    {
+        private static readonly Regex CutoffFilter =
+            new(@"^FirstNotifiedAt lt datetime'(?<cutoff>[^']+)'$", RegexOptions.Compiled);
+
+        public TableHardwareRejectionNotificationTracker Sut { get; }
+        public Mock<TableClient> Table { get; }
+        public Dictionary<(string Pk, string Rk), TableEntity> Store { get; } = new();
+        public List<string> QueryFilters { get; } = new();
+        public HashSet<string> FailDeleteForRowKeys { get; } = new();
+
+        /// <summary>Non-409 failure injected into AddEntityAsync (fail-soft path).</summary>
+        public Exception? AddFailure { get; init; }
+
+        /// <summary>Failure injected into QueryAsync enumeration (sweep fail-soft path).</summary>
+        public Exception? QueryFailure { get; init; }
+
+        public Harness()
+        {
+            Table = new Mock<TableClient>();
+
+            Table.Setup(c => c.AddEntityAsync(It.IsAny<TableEntity>(), It.IsAny<CancellationToken>()))
+                .Returns<TableEntity, CancellationToken>((e, _) =>
+                {
+                    if (AddFailure != null) throw AddFailure;
+                    if (Store.ContainsKey((e.PartitionKey, e.RowKey)))
+                        throw new RequestFailedException(409, "Conflict", "EntityAlreadyExists", null);
+                    Store[(e.PartitionKey, e.RowKey)] = e;
+                    return Task.FromResult(new Mock<Response>().Object);
+                });
+
+            Table.Setup(c => c.QueryAsync<TableEntity>(
+                    It.IsAny<string>(),
+                    It.IsAny<int?>(),
+                    It.IsAny<IEnumerable<string>>(),
+                    It.IsAny<CancellationToken>()))
+                .Returns<string, int?, IEnumerable<string>, CancellationToken>((filter, _, _, _) =>
+                {
+                    if (QueryFailure != null) throw QueryFailure;
+                    QueryFilters.Add(filter);
+                    var rows = Store.Values.Where(e => MatchesCutoff(filter, e)).ToList();
+                    return AsyncPageable<TableEntity>.FromPages(new[]
+                    {
+                        Page<TableEntity>.FromValues(rows, null, new Mock<Response>().Object),
+                    });
+                });
+
+            Table.Setup(c => c.DeleteEntityAsync(
+                    It.IsAny<string>(),
+                    It.IsAny<string>(),
+                    It.IsAny<ETag>(),
+                    It.IsAny<CancellationToken>()))
+                .Returns<string, string, ETag, CancellationToken>((pk, rk, _, _) =>
+                {
+                    if (FailDeleteForRowKeys.Contains(rk))
+                        throw new RequestFailedException(500, "DeleteFailed");
+                    Store.Remove((pk, rk));
+                    return Task.FromResult(new Mock<Response>().Object);
+                });
+
+            var serviceClient = new Mock<TableServiceClient>();
+            serviceClient.Setup(s => s.GetTableClient(It.IsAny<string>())).Returns(Table.Object);
+
+            Sut = new TableHardwareRejectionNotificationTracker(
+                new TableStorageService(serviceClient.Object, NullLogger<TableStorageService>.Instance),
+                NullLogger<TableHardwareRejectionNotificationTracker>.Instance);
+        }
+
+        /// <summary>Backdates every stored row so a retention sweep sees it as expired.</summary>
+        public void AgeRows(TimeSpan age)
+        {
+            foreach (var entity in Store.Values)
+                entity["FirstNotifiedAt"] = DateTime.UtcNow - age;
+        }
+
+        /// <summary>
+        /// Minimal server-side emulation of the OData filter the tracker builds. An unrecognised
+        /// filter matches nothing, so a rewrite that no longer selects expired rows fails loudly
+        /// here instead of leaving dead rows in production.
+        /// </summary>
+        private static bool MatchesCutoff(string filter, TableEntity entity)
+        {
+            var match = CutoffFilter.Match(filter ?? string.Empty);
+            if (!match.Success) return false;
+
+            var cutoff = DateTime.Parse(
+                match.Groups["cutoff"].Value,
+                null,
+                System.Globalization.DateTimeStyles.AdjustToUniversal | System.Globalization.DateTimeStyles.AssumeUniversal);
+            var firstNotified = entity.GetDateTime("FirstNotifiedAt");
+            return firstNotified.HasValue && firstNotified.Value.ToUniversalTime() < cutoff;
+        }
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/HardwareRejectionNotificationTrackerKeyTests.cs

@@ -59,4 +59,35 @@ public class HardwareRejectionNotificationTrackerKeyTests
         var key = TableHardwareRejectionNotificationTracker.BuildRowKey("", "");
         Assert.Equal("|", key);
     }
+
+    // =========================================================================
+    // TPM PSS row keys (same table, "tpmpss|" prefix — dedup per serial)
+    // =========================================================================
+
+    [Fact]
+    public void BuildTpmPssRowKey_LowercasesTrimsAndPrefixes()
+    {
+        var key = TableHardwareRejectionNotificationTracker.BuildTpmPssRowKey("  S4SQ8685 ");
+        Assert.Equal("tpmpss|s4sq8685", key);
+    }
+
+    [Theory]
+    [InlineData("S4SQ8685", "s4sq8685")]
+    [InlineData("s4Sq8685", "S4SQ8685")]
+    public void BuildTpmPssRowKey_IsCaseInsensitive(string serialA, string serialB)
+    {
+        Assert.Equal(
+            TableHardwareRejectionNotificationTracker.BuildTpmPssRowKey(serialA),
+            TableHardwareRejectionNotificationTracker.BuildTpmPssRowKey(serialB));
+    }
+
+    [Fact]
+    public void BuildTpmPssRowKey_DoesNotCollideWithHardwareKeyOfSameText()
+    {
+        // A hardware key is "{mfr}|{model}"; the TPM key's fixed "tpmpss|" prefix keeps the
+        // two key spaces disjoint within the shared table.
+        var hardwareKey = TableHardwareRejectionNotificationTracker.BuildRowKey("Lenovo", "S4SQ8685");
+        var tpmKey = TableHardwareRejectionNotificationTracker.BuildTpmPssRowKey("S4SQ8685");
+        Assert.NotEqual(hardwareKey, tpmKey);
+    }
 }

src/Backend/AutopilotMonitor.Functions.Tests/HostedDiagnosticsBlobServiceTests.cs

@@ -4,9 +4,12 @@ using System.Linq;
 using System.Threading;
 using System.Threading.Tasks;
 using AutopilotMonitor.Functions.Services.Diagnostics;
+using Azure;
 using Azure.Storage.Blobs;
+using Azure.Storage.Blobs.Models;
 using Azure.Storage.Sas;
 using Microsoft.Extensions.Logging.Abstractions;
+using Moq;
 using Xunit;
 
 namespace AutopilotMonitor.Functions.Tests;
@@ -172,6 +175,119 @@ public class HostedDiagnosticsBlobServiceTests
         Assert.StartsWith(TenantB + "/", b.BlobPath);
     }
 
+    // ── DeleteBySessionPrefixAsync — cascade sweep for multi-package sessions ──────
+
+    private const string SessionA = "aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa";
+
+    [Theory]
+    [InlineData("not-a-guid", SessionA)]
+    [InlineData(TenantA, "not-a-guid")]
+    public async Task DeleteBySessionPrefix_RejectsNonGuidInputs(string tenantId, string sessionId)
+    {
+        var svc = new PrefixDeleteFake(containerExists: true);
+        await Assert.ThrowsAsync<ArgumentException>(
+            () => svc.DeleteBySessionPrefixAsync(tenantId, sessionId));
+    }
+
+    [Fact]
+    public async Task DeleteBySessionPrefix_MissingContainer_IsNoOp()
+    {
+        var svc = new PrefixDeleteFake(containerExists: false);
+        var deleted = await svc.DeleteBySessionPrefixAsync(TenantA, SessionA);
+        Assert.Equal(0, deleted);
+        Assert.Null(svc.CapturedPrefix);
+    }
+
+    [Fact]
+    public async Task DeleteBySessionPrefix_UsesSessionScopedPrefix_AndDeletesEveryMatch()
+    {
+        // Two packages for the session: an on-demand (server-requested) one and the terminal
+        // one. The sweep must issue a delete for BOTH even though the Sessions row (and thus
+        // the deletion manifest) only references the last.
+        var svc = new PrefixDeleteFake(
+            containerExists: true,
+            blobNames: new[]
+            {
+                $"{TenantA}/AgentDiagnostics-{SessionA}-20260724T100000-server-requested.zip",
+                $"{TenantA}/AgentDiagnostics-{SessionA}-20260724T113000.zip",
+            });
+
+        var deleted = await svc.DeleteBySessionPrefixAsync(TenantA, SessionA);
+
+        Assert.Equal(2, deleted);
+        Assert.Equal($"{TenantA}/AgentDiagnostics-{SessionA}-", svc.CapturedPrefix);
+        Assert.Equal(2, svc.DeletedBlobNames.Count);
+        Assert.All(svc.DeletedBlobNames, n => Assert.StartsWith(svc.CapturedPrefix!, n));
+    }
+
+    [Fact]
+    public async Task DeleteBySessionPrefix_NoMatches_ReturnsZero()
+    {
+        var svc = new PrefixDeleteFake(containerExists: true, blobNames: Array.Empty<string>());
+        var deleted = await svc.DeleteBySessionPrefixAsync(TenantA, SessionA);
+        Assert.Equal(0, deleted);
+        Assert.Equal($"{TenantA}/AgentDiagnostics-{SessionA}-", svc.CapturedPrefix);
+    }
+
+    /// <summary>
+    /// Overrides <c>GetContainerClient</c> with a Moq'd <see cref="BlobContainerClient"/> whose
+    /// enumeration returns a canned page for the CAPTURED prefix. Prefix filtering itself is
+    /// the Azure SDK's contract — these tests pin what prefix we ask for and that every
+    /// returned item is deleted.
+    /// </summary>
+    private sealed class PrefixDeleteFake : HostedDiagnosticsBlobService
+    {
+        private readonly Mock<BlobContainerClient> _container;
+
+        public string? CapturedPrefix { get; private set; }
+        public List<string> DeletedBlobNames { get; } = new();
+
+        public PrefixDeleteFake(bool containerExists, string[]? blobNames = null)
+            : base(
+                new BlobServiceClient("UseDevelopmentStorage=true"),
+                NullLogger<HostedDiagnosticsBlobService>.Instance,
+                usesManagedIdentity: false)
+        {
+            _container = new Mock<BlobContainerClient>();
+            _container
+                .Setup(c => c.ExistsAsync(It.IsAny<CancellationToken>()))
+                .ReturnsAsync(Response.FromValue(containerExists, Mock.Of<Response>()));
+
+            _container
+                .Setup(c => c.GetBlobsAsync(
+                    It.IsAny<BlobTraits>(), It.IsAny<BlobStates>(), It.IsAny<string?>(), It.IsAny<CancellationToken>()))
+                .Returns((BlobTraits _, BlobStates _, string? prefix, CancellationToken _) =>
+                {
+                    CapturedPrefix = prefix;
+                    var items = (blobNames ?? Array.Empty<string>())
+                        .Select(n => BlobsModelFactory.BlobItem(name: n))
+                        .ToList();
+                    return AsyncPageable<BlobItem>.FromPages(new[]
+                    {
+                        Page<BlobItem>.FromValues(items, continuationToken: null, Mock.Of<Response>()),
+                    });
+                });
+
+            _container
+                .Setup(c => c.GetBlobClient(It.IsAny<string>()))
+                .Returns((string name) =>
+                {
+                    var blob = new Mock<BlobClient>();
+                    blob
+                        .Setup(b => b.DeleteIfExistsAsync(
+                            It.IsAny<DeleteSnapshotsOption>(), It.IsAny<BlobRequestConditions>(), It.IsAny<CancellationToken>()))
+                        .ReturnsAsync(() =>
+                        {
+                            DeletedBlobNames.Add(name);
+                            return Response.FromValue(true, Mock.Of<Response>());
+                        });
+                    return blob.Object;
+                });
+        }
+
+        protected override BlobContainerClient GetContainerClient() => _container.Object;
+    }
+
     // ── Constants ───────────────────────────────────────────────────────────────────
 
     [Fact]

src/Backend/AutopilotMonitor.Functions.Tests/IngestApiLatencyProjectionTests.cs

@@ -0,0 +1,165 @@
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Pins the per-session API-latency projection (2026-07-28): the V2 agent emits cumulative
+/// counters <c>net_total_latency_ms</c> / <c>net_total_requests</c> in every
+/// <c>agent_metrics_snapshot</c>; ingest derives the session-wide average from the LAST
+/// snapshot of the batch and persists it via <c>UpdateSessionNetworkLatencyAsync</c>
+/// (last-write-wins on cumulative counters → replay-idempotent).
+/// Wire-format note: the ingest body is parsed with Newtonsoft into
+/// <c>Dictionary&lt;string, object&gt;</c>, so integer JSON numbers arrive as boxed
+/// <c>long</c> — fixtures mirror that, not the agent's in-memory types.
+/// </summary>
+public class IngestApiLatencyProjectionTests
+{
+    private static EnrollmentEvent Snapshot(object? totalLatencyMs, object? totalRequests)
+    {
+        var evt = new EnrollmentEvent
+        {
+            EventType = "agent_metrics_snapshot",
+            Data = new Dictionary<string, object>
+            {
+                ["agent_cpu_percent"] = 3.62,
+                ["net_requests"] = 50L,
+                ["net_avg_latency_ms"] = 513.0,
+            },
+        };
+        if (totalLatencyMs != null) evt.Data["net_total_latency_ms"] = totalLatencyMs;
+        if (totalRequests != null) evt.Data["net_total_requests"] = totalRequests;
+        return evt;
+    }
+
+    [Fact]
+    public void CumulativeCounters_YieldSessionAverage()
+    {
+        var events = new List<EnrollmentEvent> { Snapshot(51_300L, 100L) };
+
+        Assert.True(EventIngestProcessor.TryComputeSessionApiLatency(events, out var avg, out var count));
+        Assert.Equal(513.0, avg);
+        Assert.Equal(100, count);
+    }
+
+    [Fact]
+    public void AverageIsRounded_ToOneDecimal()
+    {
+        var events = new List<EnrollmentEvent> { Snapshot(1_000L, 3L) };
+
+        Assert.True(EventIngestProcessor.TryComputeSessionApiLatency(events, out var avg, out _));
+        Assert.Equal(333.3, avg);
+    }
+
+    [Fact]
+    public void LastSnapshotOfBatch_Wins()
+    {
+        // Cumulative counters grow monotonically — the last snapshot carries the most
+        // complete session average.
+        var events = new List<EnrollmentEvent>
+        {
+            Snapshot(51_300L, 100L),
+            new EnrollmentEvent { EventType = "download_progress" },
+            Snapshot(120_000L, 200L),
+        };
+
+        Assert.True(EventIngestProcessor.TryComputeSessionApiLatency(events, out var avg, out var count));
+        Assert.Equal(600.0, avg);
+        Assert.Equal(200, count);
+    }
+
+    [Fact]
+    public void OldAgentSnapshot_WithoutTotalLatencyField_DoesNotMatch()
+    {
+        // Agents predating net_total_latency_ms emit net_total_requests only.
+        var events = new List<EnrollmentEvent> { Snapshot(totalLatencyMs: null, totalRequests: 100L) };
+
+        Assert.False(EventIngestProcessor.TryComputeSessionApiLatency(events, out _, out _));
+    }
+
+    [Fact]
+    public void SnapshotWithLatencyButMissingRequests_DoesNotMatch()
+    {
+        var events = new List<EnrollmentEvent> { Snapshot(51_300L, totalRequests: null) };
+
+        Assert.False(EventIngestProcessor.TryComputeSessionApiLatency(events, out _, out _));
+    }
+
+    [Fact]
+    public void ZeroRequests_DoesNotMatch()
+    {
+        // No requests yet (first snapshot before the first upload completes) — no division.
+        var events = new List<EnrollmentEvent> { Snapshot(0L, 0L) };
+
+        Assert.False(EventIngestProcessor.TryComputeSessionApiLatency(events, out _, out _));
+    }
+
+    [Fact]
+    public void MatchingLastSnapshot_SkipsOlderAgentSnapshotAfterIt()
+    {
+        // A field-less (old-schema) snapshot later in the batch must not shadow an earlier
+        // one that carries the counters — the predicate selects the last MATCHING snapshot.
+        var events = new List<EnrollmentEvent>
+        {
+            Snapshot(51_300L, 100L),
+            Snapshot(totalLatencyMs: null, totalRequests: 150L),
+        };
+
+        Assert.True(EventIngestProcessor.TryComputeSessionApiLatency(events, out var avg, out var count));
+        Assert.Equal(513.0, avg);
+        Assert.Equal(100, count);
+    }
+
+    [Fact]
+    public void BatchWithoutSnapshots_DoesNotMatch()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            new EnrollmentEvent { EventType = "esp_phase_changed" },
+            new EnrollmentEvent { EventType = "app_install_started" },
+        };
+
+        Assert.False(EventIngestProcessor.TryComputeSessionApiLatency(events, out _, out _));
+    }
+
+    // -------- TryGetDouble coercion ladder (mirrors PlatformMetricsService.GetDouble) --------
+
+    [Theory]
+    [InlineData(513.5)]         // double — decimal JSON number
+    [InlineData(513)]           // int — in-memory constructed events (tests, V1 paths)
+    [InlineData(513L)]          // long — Newtonsoft wire form for integer JSON numbers
+    [InlineData(513f)]          // float
+    public void TryGetDouble_CoercesBoxedNumericPrimitives(object boxed)
+    {
+        var data = new Dictionary<string, object> { ["k"] = boxed };
+
+        Assert.True(EventIngestProcessor.TryGetDouble(data, "k", out var value));
+        Assert.Equal(Convert.ToDouble(boxed), value);
+    }
+
+    [Fact]
+    public void TryGetDouble_ParsesStrings_InvariantCulture()
+    {
+        var data = new Dictionary<string, object> { ["k"] = "513.5" };
+
+        Assert.True(EventIngestProcessor.TryGetDouble(data, "k", out var value));
+        Assert.Equal(513.5, value);
+    }
+
+    [Theory]
+    [InlineData("not-a-number")]
+    [InlineData("")]
+    public void TryGetDouble_RejectsUnparseableStrings(string raw)
+    {
+        var data = new Dictionary<string, object> { ["k"] = raw };
+
+        Assert.False(EventIngestProcessor.TryGetDouble(data, "k", out _));
+    }
+
+    [Fact]
+    public void TryGetDouble_MissingKey_ReturnsFalse()
+    {
+        Assert.False(EventIngestProcessor.TryGetDouble(new Dictionary<string, object>(), "k", out _));
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/IngestOnDemandDiagnosticsTests.cs

@@ -0,0 +1,128 @@
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Session 8e4cc4ae (2026-07-24): two on-demand ("Collect Logs") uploads landed in hosted
+/// storage, but the portal never flipped to Download — the agent's ServerActionDispatcher
+/// confirms the upload via <c>server_action_executed(actionType=request_diagnostics,
+/// blobName=…)</c> and emits NO <c>diagnostics_uploaded</c> (only the terminal path does),
+/// so the Sessions row was never stamped. These tests pin the classification predicate and
+/// the destination inference that close that gap backend-side (no agent release needed).
+/// </summary>
+public class IngestOnDemandDiagnosticsTests
+{
+    private const string TenantId = "b54dc1af-5320-4f60-b5d4-821e0cf2a359";
+    private const string HostedBlob =
+        TenantId + "/AgentDiagnostics-8e4cc8ae-1111-2222-3333-444455556666-20260724170902-server-requested.zip";
+
+    private static EnrollmentEvent Evt(
+        string eventType = "server_action_executed",
+        string? actionType = "request_diagnostics",
+        string? blobName = HostedBlob)
+    {
+        var data = new Dictionary<string, object>();
+        if (actionType != null) data["actionType"] = actionType;
+        if (blobName != null) data["blobName"] = blobName;
+        return new EnrollmentEvent { EventType = eventType, Data = data };
+    }
+
+    // -------- IsOnDemandDiagnosticsUploadConfirmation --------
+
+    [Fact]
+    public void ExecutedRequestDiagnosticsWithBlobName_IsConfirmation()
+    {
+        Assert.True(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(Evt()));
+    }
+
+    [Fact]
+    public void ActionTypeIsCaseInsensitive()
+    {
+        Assert.True(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(
+            Evt(actionType: "Request_Diagnostics")));
+    }
+
+    [Theory]
+    [InlineData("server_action_received")]
+    [InlineData("server_action_failed")]
+    [InlineData("diagnostics_uploaded")]
+    public void OtherEventTypes_AreNotConfirmations(string eventType)
+    {
+        Assert.False(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(Evt(eventType: eventType)));
+    }
+
+    [Theory]
+    [InlineData("rotate_config")]
+    [InlineData("terminate_session")]
+    [InlineData(null)]
+    public void OtherActionTypes_AreNotConfirmations(string? actionType)
+    {
+        Assert.False(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(Evt(actionType: actionType)));
+    }
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    public void MissingOrEmptyBlobName_IsNotAConfirmation(string? blobName)
+    {
+        // A failed upload never reaches server_action_executed, but an executed event without
+        // blobName (defensive) must not wipe/stamp the row either.
+        Assert.False(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(Evt(blobName: blobName)));
+    }
+
+    [Fact]
+    public void NullEventOrData_IsNotAConfirmation()
+    {
+        Assert.False(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(null));
+        // Data is non-nullable on the model but the JSON deserializer can still hand us null
+        // for a payload without "data" — null! pins the guard the predicate actually has.
+        Assert.False(EventIngestProcessor.IsOnDemandDiagnosticsUploadConfirmation(
+            new EnrollmentEvent { EventType = "server_action_executed", Data = null! }));
+    }
+
+    // -------- InferDiagnosticsDestination --------
+
+    [Fact]
+    public void ExplicitDestination_AlwaysWins()
+    {
+        Assert.Equal("CustomerSas",
+            EventIngestProcessor.InferDiagnosticsDestination("CustomerSas", HostedBlob, TenantId));
+    }
+
+    [Fact]
+    public void TenantPrefixedBlobName_InfersHosted()
+    {
+        Assert.Equal("Hosted",
+            EventIngestProcessor.InferDiagnosticsDestination(null, HostedBlob, TenantId));
+    }
+
+    [Fact]
+    public void TenantPrefixMatch_IsCaseInsensitive()
+    {
+        Assert.Equal("Hosted",
+            EventIngestProcessor.InferDiagnosticsDestination(null, HostedBlob.ToUpperInvariant(), TenantId));
+    }
+
+    [Fact]
+    public void BareFilename_InfersNothing()
+    {
+        // CustomerSas uploads persist the bare zip filename — repo must leave the
+        // destination column unchanged (read-time default CustomerSas).
+        Assert.Null(EventIngestProcessor.InferDiagnosticsDestination(
+            null, "AgentDiagnostics-x-20260724.zip", TenantId));
+    }
+
+    [Fact]
+    public void ForeignTenantPrefix_InfersNothing()
+    {
+        Assert.Null(EventIngestProcessor.InferDiagnosticsDestination(
+            null, "99999999-9999-9999-9999-999999999999/x.zip", TenantId));
+    }
+
+    [Fact]
+    public void NullBlobName_InfersNothing()
+    {
+        Assert.Null(EventIngestProcessor.InferDiagnosticsDestination(null, null, TenantId));
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/MetricsMathAppPayloadTests.cs

@@ -119,6 +119,138 @@ public class MetricsMathAppPayloadTests
         Assert.Equal(2, topCode.GetProperty("count").GetInt32());
     }
 
+    [Fact]
+    public void FailureRate_IsOverFinishedInstalls_InProgressExcluded()
+    {
+        var summaries = new List<AppInstallSummary>
+        {
+            new() { AppName = "App A", Status = "Failed" },
+            new() { AppName = "App A", Status = "Succeeded", DurationSeconds = 10 },
+            new() { AppName = "App A", Status = "InProgress" },
+            new() { AppName = "App A", Status = "InProgress" },
+        };
+
+        var app = Build(summaries).GetProperty("topFailingApps")[0];
+
+        Assert.Equal(4, app.GetProperty("totalInstalls").GetInt32());
+        // 1 / (1 + 1) = 50% (over finished), NOT 1 / 4 = 25% (over all installs).
+        Assert.Equal(50d, app.GetProperty("failureRate").GetDouble());
+    }
+
+    // ── PR0 (2026-07-26): skip/unmeasured classification ────────────────────
+
+    [Fact]
+    public void Skips_LeaveTheFailureRate_AndAreReportedSeparately()
+    {
+        // 1 Failed + 1 real install + 3 skips ("Update for X" not applicable).
+        // Rate must be 1/(1+1) = 50%, NOT 1/(1+4) = 20% — skips are not attempts.
+        var summaries = new List<AppInstallSummary>
+        {
+            new() { AppName = "App A", Status = "Failed", TerminalState = "Error" },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 30 },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Skipped" },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Skipped" },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Postponed" },
+        };
+
+        var root = Build(summaries);
+        var app = root.GetProperty("topFailingApps")[0];
+
+        Assert.Equal(50d, app.GetProperty("failureRate").GetDouble());
+        Assert.Equal(1, app.GetProperty("succeeded").GetInt32());
+        Assert.Equal(3, app.GetProperty("skipped").GetInt32());
+        Assert.Equal(0, app.GetProperty("unmeasured").GetInt32());
+        Assert.Equal(3, root.GetProperty("totalSkipped").GetInt32());
+    }
+
+    [Fact]
+    public void ZeroDurations_AreExcludedFromAverages_AndCountedAsUnmeasured()
+    {
+        // Two measured installs (100s, 200s) + one whose start was never observed (audit §0.5:
+        // agent attach window — duration UNKNOWN). avg must be 150, not (100+200+0)/3 = 100.
+        var summaries = new List<AppInstallSummary>
+        {
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 100 },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 200 },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 0 },
+        };
+
+        var root = Build(summaries);
+        Assert.Equal(1, root.GetProperty("totalUnmeasured").GetInt32());
+
+        // Rebuild with a third measured row to clear the sample floor and read the average.
+        summaries.Add(new AppInstallSummary { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 150 });
+        var slowest = Build(summaries).GetProperty("slowestApps");
+        Assert.Equal(1, slowest.GetArrayLength());
+        Assert.Equal(150d, slowest[0].GetProperty("avgDurationSeconds").GetDouble());
+        Assert.Equal(1, slowest[0].GetProperty("unmeasured").GetInt32());
+    }
+
+    [Fact]
+    public void LegacyZeroDurationRows_StayInRate_ButNeverInDurations()
+    {
+        // Rows written before TerminalState existed: cannot be classified as skip vs unmeasured.
+        // They keep the legacy rate behavior (counted as succeeded) and are excluded from
+        // durations (counted as unmeasured — duration genuinely not measured).
+        var summaries = new List<AppInstallSummary>
+        {
+            new() { AppName = "App A", Status = "Failed" },
+            new() { AppName = "App A", Status = "Succeeded", DurationSeconds = 0 },   // legacy, no TerminalState
+            new() { AppName = "App A", Status = "Succeeded", DurationSeconds = 60, TerminalState = "Installed" },
+        };
+
+        var app = Build(summaries).GetProperty("topFailingApps")[0];
+
+        Assert.Equal(2, app.GetProperty("succeeded").GetInt32());      // legacy row stays in the rate
+        Assert.Equal(33.3d, app.GetProperty("failureRate").GetDouble());
+        Assert.Equal(1, app.GetProperty("unmeasured").GetInt32());
+        Assert.Equal(0, app.GetProperty("skipped").GetInt32());
+    }
+
+    [Fact]
+    public void ImplausibleDurations_AreExcludedFromAverages_AndCountedAsUnmeasured()
+    {
+        // A duration beyond the agent's max observation lifetime is a back-stamped emission
+        // artifact (CompletedAt pinned at session end — production: 60 175 s app rows inside a
+        // 2 697 s session), not a slow install. It must fall out of the averages exactly like
+        // an unobserved start. Boundary: exactly 6 h is still plausible/measured.
+        var summaries = new List<AppInstallSummary>
+        {
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 100 },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 200 },
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed",
+                    DurationSeconds = MetricsMath.MaxPlausibleInstallDurationSeconds },       // boundary: measured
+            new() { AppName = "App A", Status = "Succeeded", TerminalState = "Installed",
+                    DurationSeconds = MetricsMath.MaxPlausibleInstallDurationSeconds + 1 },   // artifact: unmeasured
+        };
+
+        var root = Build(summaries);
+        Assert.Equal(1, root.GetProperty("totalUnmeasured").GetInt32());
+
+        var slowest = root.GetProperty("slowestApps");
+        Assert.Equal(1, slowest.GetArrayLength());
+        Assert.Equal(3, slowest[0].GetProperty("measuredInstalls").GetInt32());
+        // avg over {100, 200, 21600} — the 21601 artifact contributes nothing.
+        Assert.Equal(7300d, slowest[0].GetProperty("avgDurationSeconds").GetDouble());
+        Assert.Equal(MetricsMath.MaxPlausibleInstallDurationSeconds, slowest[0].GetProperty("maxDurationSeconds").GetInt32());
+    }
+
+    [Fact]
+    public void SlowestApps_GateOnMeasuredInstalls_NotOnSucceededCount()
+    {
+        // 3 succeeded but only 2 measured → below the floor; an app must not rank "fast"
+        // (or at all) on a handful of unobserved zeros.
+        var summaries = new List<AppInstallSummary>
+        {
+            new() { AppName = "Mostly Unmeasured", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 5 },
+            new() { AppName = "Mostly Unmeasured", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 5 },
+            new() { AppName = "Mostly Unmeasured", Status = "Succeeded", TerminalState = "Installed", DurationSeconds = 0 },
+        };
+
+        var slowest = Build(summaries).GetProperty("slowestApps");
+        Assert.Equal(0, slowest.GetArrayLength());
+    }
+
     [Fact]
     public void SlowestApps_DropsAppsBelowSampleFloor()
     {

src/Backend/AutopilotMonitor.Functions.Tests/OffboardingSessionEnumeratorTests.cs

@@ -125,5 +125,6 @@ public class OffboardingSessionEnumeratorTests
         public Task<bool> IsSessionIndexEmptyAsync() => throw new NotImplementedException();
         public Task<List<OrphanedEventSession>> GetOrphanedEventSessionsAsync(TimeSpan gracePeriod) => throw new NotImplementedException();
         public Task DeleteEventSessionIndexEntryAsync(string tenantId, string sessionId) => throw new NotImplementedException();
+        public Task<List<(string TenantId, string SessionId)>> GetAllSessionTombstoneKeysAsync() => throw new NotImplementedException();
     }
 }

src/Backend/AutopilotMonitor.Functions.Tests/PolicyEnforcementMiddlewareTests.cs

@@ -1,4 +1,5 @@
 using System.Security.Claims;
+using AutopilotMonitor.Functions.Functions.Config;
 using AutopilotMonitor.Functions.Middleware;
 using AutopilotMonitor.Functions.Security;
 using AutopilotMonitor.Functions.Services;
@@ -205,8 +206,101 @@ public class PolicyEnforcementMiddlewareTests
         Assert.Equal(403, result.StatusCode);
     }
 
-    // ── config GET (TenantAdminOrGlobalReader): own-tenant admin view vs reader view ──
-    // GetTenantConfigurationFunction redacts only when IsGlobalReader && !(IsTenantAdmin && Target==Tenant).
+    // ── TenantAdminOrOperator (POST sessions/{id}/actions — "Collect Logs") ────────
+
+    private const string SessionA = "aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa";
+
+    private static void AsTenantMemberRole(Harness h, string tenantId, string upn, string role) =>
+        h.Repo.Setup(r => r.GetTenantMemberAsync(It.IsAny<string>(), It.IsAny<string>()))
+            .ReturnsAsync(new TenantMember { Upn = upn, TenantId = tenantId, Role = role, IsEnabled = true });
+
+    [Fact]
+    public async Task SessionActions_TenantOperator_IsAllowed_WithOperatorRoleOnContext()
+    {
+        const string upn = "operator@contoso.com";
+        var h = BuildHarness();
+        AsTenantMemberRole(h, TenantA, upn, Constants.TenantRoles.Operator);
+
+        var result = await h.Middleware.DecideAsync(
+            "POST", $"/api/sessions/{SessionA}/actions", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.True(result.Allowed);
+        // The function's per-type gate depends on these exact flags: an Operator must arrive
+        // with IsTenantAdmin=false + UserRole=Operator so terminate_session/rotate_config 403.
+        Assert.False(result.Context!.IsTenantAdmin);
+        Assert.Equal(Constants.TenantRoles.Operator, result.Context!.UserRole);
+    }
+
+    [Fact]
+    public async Task SessionActions_TenantAdmin_IsAllowed_WithAdminFlag()
+    {
+        const string upn = "admin@contoso.com";
+        var h = BuildHarness();
+        h.AsTenantAdmin(TenantA, upn);
+
+        var result = await h.Middleware.DecideAsync(
+            "POST", $"/api/sessions/{SessionA}/actions", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.True(result.Allowed);
+        Assert.True(result.Context!.IsTenantAdmin);
+    }
+
+    [Fact]
+    public async Task SessionActions_TenantViewer_IsForbidden()
+    {
+        const string upn = "viewer@contoso.com";
+        var h = BuildHarness();
+        AsTenantMemberRole(h, TenantA, upn, Constants.TenantRoles.Viewer);
+
+        var result = await h.Middleware.DecideAsync(
+            "POST", $"/api/sessions/{SessionA}/actions", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.False(result.Allowed);
+        Assert.Equal(403, result.StatusCode);
+    }
+
+    [Fact]
+    public async Task SessionActions_NonMember_IsForbidden()
+    {
+        var h = BuildHarness(); // defaults: no membership, no global role
+
+        var result = await h.Middleware.DecideAsync(
+            "POST", $"/api/sessions/{SessionA}/actions", null, AuthedPrincipal(TenantA, "outsider@contoso.com"));
+
+        Assert.False(result.Allowed);
+        Assert.Equal(403, result.StatusCode);
+    }
+
+    [Fact]
+    public async Task SessionActions_GlobalAdmin_IsAllowed()
+    {
+        const string upn = "ga@platform.com";
+        var h = BuildHarness();
+        h.AsGlobalRole(Constants.GlobalRoles.GlobalAdmin);
+
+        var result = await h.Middleware.DecideAsync(
+            "POST", $"/api/sessions/{SessionA}/actions", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.True(result.Allowed);
+        Assert.True(result.Context!.IsGlobalAdmin);
+    }
+
+    [Fact]
+    public async Task SessionActions_PureGlobalReader_IsForbidden_WriteTier()
+    {
+        const string upn = "reader@platform.com";
+        var h = BuildHarness();
+        h.AsGlobalRole(Constants.GlobalRoles.GlobalReader);
+
+        var result = await h.Middleware.DecideAsync(
+            "POST", $"/api/sessions/{SessionA}/actions", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.False(result.Allowed);
+        Assert.Equal(403, result.StatusCode);
+    }
+
+    // ── config GET (MemberRead): own-tenant admin view vs redacted reader/member view ──
+    // GetTenantConfigurationFunction redacts unless GA or own-tenant Admin (CanViewSecrets).
     // These lock in the RequestContext that drives that decision.
 
     [Fact]
@@ -243,6 +337,38 @@ public class PolicyEnforcementMiddlewareTests
         Assert.Equal(TenantB, rc.TargetTenantId);
     }
 
+    [Fact]
+    public async Task ConfigGet_OwnTenantOperator_IsAllowed_RedactedMemberView()
+    {
+        // Read-only Settings view for troubleshooting staff: an own-tenant Operator is admitted
+        // (MemberRead) but arrives with IsTenantAdmin=false, so CanViewSecrets ⇒ redacted copy.
+        const string upn = "operator@contoso.com";
+        var h = BuildHarness();
+        AsTenantMemberRole(h, TenantA, upn, Constants.TenantRoles.Operator);
+
+        var result = await h.Middleware.DecideAsync("GET", $"/api/config/{TenantA}", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.True(result.Allowed);
+        var rc = result.Context!;
+        Assert.False(rc.IsTenantAdmin);
+        Assert.False(GetTenantConfigurationFunction.CanViewSecrets(rc));
+    }
+
+    [Fact]
+    public async Task ConfigGet_CrossTenantOperator_IsForbidden()
+    {
+        // Membership is resolved against the caller's JWT tenant — an Operator of A naming
+        // tenant B in the route is a cross-tenant read and must stay 403.
+        const string upn = "operator@contoso.com";
+        var h = BuildHarness();
+        AsTenantMemberRole(h, TenantA, upn, Constants.TenantRoles.Operator);
+
+        var result = await h.Middleware.DecideAsync("GET", $"/api/config/{TenantB}", null, AuthedPrincipal(TenantA, upn));
+
+        Assert.False(result.Allowed);
+        Assert.Equal(403, result.StatusCode);
+    }
+
     // ── Cross-tenant read: GlobalReader crosses tenants (HasGlobalScope gate) ───────
 
     [Fact]

src/Backend/AutopilotMonitor.Functions.Tests/RuleRegressionRadarTests.cs

@@ -0,0 +1,405 @@
+using AutopilotMonitor.Functions.DataAccess.TableStorage;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// F3 PR6 (insights spec §F3): the Wilson primitive (hand-computed vectors), the radar's
+/// detection gates and EVERY suppression branch (table-driven), the re-arm rule, the
+/// dimension-concentration gates, the tracker keyspace round-trip (tri-states included)
+/// and the alert wording contracts ("correlated, not necessarily causal").
+/// </summary>
+public class RuleRegressionRadarTests
+{
+    private static readonly DateTime Target = new(2026, 7, 26, 0, 0, 0, DateTimeKind.Utc);
+    private const string TenantA = "aaaaaaaa-0000-0000-0000-000000000001";
+    private const string RuleX = "ANALYZE-NET-001";
+
+    // ── Wilson interval (MetricsMath) — hand-computed 95 % vectors ──────────
+
+    [Theory]
+    [InlineData(10, 20, 0.2993, 0.7007)]  // symmetric mid-range
+    [InlineData(0, 20, 0.0000, 0.1611)]   // zero successes: lower pinned at 0
+    [InlineData(5, 25, 0.0886, 0.3913)]
+    public void WilsonInterval_MatchesHandComputedVectors(int successes, int trials, double lower, double upper)
+    {
+        var (l, u) = MetricsMath.WilsonInterval(successes, trials);
+        Assert.Equal(lower, l, 3);
+        Assert.Equal(upper, u, 3);
+    }
+
+    [Fact]
+    public void WilsonInterval_ZeroTrials_IsUninformative_NeverClaimsSeparation()
+    {
+        Assert.Equal((0.0, 1.0), MetricsMath.WilsonInterval(0, 0));
+        // A zero-denominator window can never be "separated" from anything.
+        Assert.False(MetricsMath.RateIncreaseSeparated(0, 0, 10, 500));
+    }
+
+    [Fact]
+    public void RateIncreaseSeparated_DisjointIntervals_True_OverlappingFalse()
+    {
+        // 15 % on n=100 (lower ≈ 0.093) vs 2 % baseline on n=500 (upper ≈ 0.036) → separated.
+        Assert.True(MetricsMath.RateIncreaseSeparated(15, 100, 10, 500));
+        // 5 % on n=100 vs 4 % baseline on n=500 → intervals overlap → noise, not a regression.
+        Assert.False(MetricsMath.RateIncreaseSeparated(5, 100, 20, 500));
+    }
+
+    // ── radar evaluation — gates + suppression branches ─────────────────────
+
+    private static RuleStatsEntry Entry(
+        string date, int fire, int sessions, string ruleId = RuleX, string ruleType = "analyze")
+        => new()
+        {
+            TenantId = TenantA,
+            Date = date,
+            RuleId = ruleId,
+            RuleType = ruleType,
+            RuleTitle = "Network check",
+            Category = "network",
+            Severity = "warning",
+            FireCount = fire,
+            EvaluationCount = sessions,
+            SessionsEvaluated = sessions,
+        };
+
+    private static string Day(int daysBeforeTarget) => Target.AddDays(-daysBeforeTarget).ToString("yyyy-MM-dd");
+
+    /// <summary>Window fire=15/100 across two days; baseline fire=10/500 across three days — fires all gates.</summary>
+    private static List<RuleStatsEntry> RegressedEntries() => new()
+    {
+        Entry(Day(0), 10, 60),
+        Entry(Day(6), 5, 40),      // windowStart boundary — still window
+        Entry(Day(7), 4, 200),     // first baseline day
+        Entry(Day(20), 4, 200),
+        Entry(Day(34), 2, 100),    // baselineStart boundary — still baseline
+    };
+
+    private static Dictionary<string, RuleRegressionRadar.RuleTimestamps> OldRule(
+        DateTime? createdAt = null, DateTime? updatedAt = null, string ruleId = RuleX)
+        => new(StringComparer.OrdinalIgnoreCase)
+        {
+            [ruleId] = new RuleRegressionRadar.RuleTimestamps(
+                createdAt ?? Target.AddDays(-100), updatedAt ?? Target.AddDays(-50)),
+        };
+
+    [Fact]
+    public void Evaluate_AllGatesPass_FiresWithFullNumbers()
+    {
+        var findings = RuleRegressionRadar.Evaluate(RegressedEntries(), Target, OldRule());
+
+        var finding = Assert.Single(findings);
+        Assert.Equal(RuleX, finding.RuleId);
+        Assert.Equal(15, finding.WindowFireCount);
+        Assert.Equal(100, finding.WindowSessionCount);
+        Assert.Equal(10, finding.BaselineFireCount);
+        Assert.Equal(500, finding.BaselineSessionCount);
+        Assert.Equal(15.0, finding.WindowRatePct);
+        Assert.Equal(2.0, finding.BaselineRatePct);
+        Assert.Equal(7.5, finding.Lift);
+        Assert.Equal(Day(6), finding.WindowStartDate);
+        Assert.Equal(Day(0), finding.WindowEndDate);
+    }
+
+    [Fact]
+    public void Evaluate_ZeroBaselineFires_FiresAsNewSignal_WithNullLift()
+    {
+        var entries = new List<RuleStatsEntry>
+        {
+            Entry(Day(0), 15, 100),
+            Entry(Day(10), 0, 500),
+        };
+
+        var finding = Assert.Single(RuleRegressionRadar.Evaluate(entries, Target, OldRule()));
+        Assert.Null(finding.Lift); // no finite lift is ever invented (truthfulness rule 1)
+        Assert.Equal(0.0, finding.BaselineRatePct);
+    }
+
+    public static TheoryData<string, List<RuleStatsEntry>> SuppressedByCounts() => new()
+    {
+        // Below the ≥5 hit-session gate.
+        { "below hit gate", new List<RuleStatsEntry> { Entry(Day(0), 4, 100), Entry(Day(10), 1, 500) } },
+        // Below the ≥20 evaluated-session gate.
+        { "below session gate", new List<RuleStatsEntry> { Entry(Day(0), 5, 19), Entry(Day(10), 1, 500) } },
+        // No baseline denominator at all (rule disabled through the baseline, re-enabled recently).
+        { "no baseline sessions", new List<RuleStatsEntry> { Entry(Day(0), 15, 100) } },
+        // Lift below 2× (10 % vs 6 %).
+        { "lift below 2x", new List<RuleStatsEntry> { Entry(Day(0), 10, 100), Entry(Day(10), 30, 500) } },
+        // Lift 2.1× but Wilson intervals overlap (5/20 vs 60/500) — small-n noise never alerts.
+        { "not separated", new List<RuleStatsEntry> { Entry(Day(0), 5, 20), Entry(Day(10), 60, 500) } },
+        // Gather rows are ignored entirely (audit Q6: per-batch dedup — bad math until fixed).
+        { "gather ignored", new List<RuleStatsEntry>
+            { Entry(Day(0), 50, 60, ruleType: "gather"), Entry(Day(10), 1, 500, ruleType: "gather") } },
+    };
+
+    [Theory]
+    [MemberData(nameof(SuppressedByCounts))]
+    public void Evaluate_SuppressionBranches_ProduceNoFinding(string branch, List<RuleStatsEntry> entries)
+    {
+        Assert.True(RuleRegressionRadar.Evaluate(entries, Target, OldRule()).Count == 0,
+            $"branch '{branch}' must not fire");
+    }
+
+    [Fact]
+    public void Evaluate_RuleEntityGates_Suppress()
+    {
+        var entries = RegressedEntries();
+
+        // Deleted rule (no entity) never alerts.
+        Assert.Empty(RuleRegressionRadar.Evaluate(
+            entries, Target, new Dictionary<string, RuleRegressionRadar.RuleTimestamps>()));
+
+        // Grace period: created inside baseline+window (< 34 days before target).
+        Assert.Empty(RuleRegressionRadar.Evaluate(
+            entries, Target, OldRule(createdAt: Target.AddDays(-20))));
+
+        // Edited inside the 7-day window — edits legitimately change hit rates.
+        Assert.Empty(RuleRegressionRadar.Evaluate(
+            entries, Target, OldRule(updatedAt: Target.AddDays(-2))));
+
+        // Edited just BEFORE the window → not suppressed.
+        Assert.Single(RuleRegressionRadar.Evaluate(
+            entries, Target, OldRule(updatedAt: Target.AddDays(-8))));
+    }
+
+    [Fact]
+    public void SumWindows_BoundaryDates_LandInTheRightBucket()
+    {
+        var entries = new List<RuleStatsEntry>
+        {
+            Entry(Target.AddDays(1).ToString("yyyy-MM-dd"), 100, 100), // future — ignored
+            Entry(Day(0), 1, 10),
+            Entry(Day(6), 2, 20),   // last window day
+            Entry(Day(7), 4, 40),   // first baseline day
+            Entry(Day(34), 8, 80),  // last baseline day
+            Entry(Day(35), 100, 100), // beyond horizon — ignored
+        };
+
+        var sums = RuleRegressionRadar.SumWindows(entries, Target);
+        Assert.Equal(3, sums.WindowFire);
+        Assert.Equal(30, sums.WindowSessions);
+        Assert.Equal(12, sums.BaselineFire);
+        Assert.Equal(120, sums.BaselineSessions);
+    }
+
+    // ── re-arm rule ─────────────────────────────────────────────────────────
+
+    [Fact]
+    public void ShouldReArm_FiresStopped_OrRateBackUnderOnePointFive()
+    {
+        // Nothing fired in the window anymore → re-arm (also covers "no stats rows at all").
+        Assert.True(RuleRegressionRadar.ShouldReArm(new List<RuleStatsEntry>(), Target));
+        Assert.True(RuleRegressionRadar.ShouldReArm(
+            new List<RuleStatsEntry> { Entry(Day(0), 0, 100), Entry(Day(10), 10, 500) }, Target));
+
+        // 2 % window vs 2 % baseline → 1.0× < 1.5 → re-arm.
+        Assert.True(RuleRegressionRadar.ShouldReArm(
+            new List<RuleStatsEntry> { Entry(Day(0), 2, 100), Entry(Day(10), 10, 500) }, Target));
+
+        // 4 % window vs 2 % baseline → 2.0× ≥ 1.5 → still elevated, keep the episode.
+        Assert.False(RuleRegressionRadar.ShouldReArm(
+            new List<RuleStatsEntry> { Entry(Day(0), 4, 100), Entry(Day(10), 10, 500) }, Target));
+
+        // Zero-baseline episode ("new signal"): no rate threshold to fall under — only the
+        // fires-stopped branch re-arms it.
+        Assert.False(RuleRegressionRadar.ShouldReArm(
+            new List<RuleStatsEntry> { Entry(Day(0), 3, 100), Entry(Day(10), 0, 500) }, Target));
+    }
+
+    // ── dimension concentration ─────────────────────────────────────────────
+
+    private static SessionSummary DimSession(string osBuild = "", string model = "", string agentVersion = "")
+        => new()
+        {
+            TenantId = TenantA,
+            SessionId = Guid.NewGuid().ToString(),
+            SerialNumber = "PF4X1ABC",
+            OsBuild = osBuild,
+            Manufacturer = "Contoso",
+            Model = model,
+            AgentVersion = agentVersion,
+            ImeAgentVersion = "",
+            StartedAt = Target,
+        };
+
+    [Fact]
+    public void DimensionConcentration_DominantOsBuild_ReportsBothShares()
+    {
+        var all = new List<SessionSummary>();
+        for (var i = 0; i < 20; i++) all.Add(DimSession(osBuild: "26100.4652"));
+        for (var i = 0; i < 80; i++) all.Add(DimSession(osBuild: "22631.3880"));
+        var hits = all.Take(16).Concat(all.Skip(20).Take(4)).ToList(); // 16 on the new build, 4 on the old
+
+        var result = RuleRegressionRadar.ComputeDimensionConcentration(hits, all)!;
+
+        Assert.Equal("osBuild", result.Dimension);
+        Assert.Equal("26100.4652", result.Value);
+        Assert.Equal(16, result.HitCount);
+        Assert.Equal(80.0, result.HitSharePct);
+        Assert.Equal(20.0, result.AllSharePct);
+        Assert.Equal(4.0, result.Lift);
+    }
+
+    [Fact]
+    public void DimensionConcentration_BelowGates_YieldsNull_NeverStretched()
+    {
+        var all = Enumerable.Range(0, 100).Select(_ => DimSession(osBuild: "26100.1")).ToList();
+
+        // Every session shares the build → lift 1.0 < 2 → no claim.
+        Assert.Null(RuleRegressionRadar.ComputeDimensionConcentration(all.Take(10).ToList(), all));
+
+        // Concentrated but only 4 hit sessions (< 5) → no claim.
+        var mixed = Enumerable.Range(0, 96).Select(_ => DimSession(osBuild: "22631.1"))
+            .Concat(Enumerable.Range(0, 4).Select(_ => DimSession(osBuild: "26100.9")))
+            .ToList();
+        var hits4 = mixed.Skip(96).ToList();
+        Assert.Null(RuleRegressionRadar.ComputeDimensionConcentration(hits4, mixed));
+
+        // No hit sessions at all → no claim.
+        Assert.Null(RuleRegressionRadar.ComputeDimensionConcentration(new List<SessionSummary>(), all));
+    }
+
+    [Fact]
+    public void DimensionConcentration_PicksTheHighestLiftAcrossDimensions()
+    {
+        var all = new List<SessionSummary>();
+        for (var i = 0; i < 25; i++) all.Add(DimSession(osBuild: "26100.1", model: "Laptop 5"));
+        for (var i = 0; i < 75; i++) all.Add(DimSession(osBuild: "22631.1", model: i < 25 ? "Laptop 5" : "Laptop 4"));
+        // Hits: 8 of 10 on the new build (lift 80/25 = 3.2); model "Contoso Laptop 5" is 50 % of
+        // fleet and 80 % of hits (lift 1.6 — below gate).
+        var hits = all.Take(8).Concat(all.Skip(25).Take(2)).ToList();
+
+        var result = RuleRegressionRadar.ComputeDimensionConcentration(hits, all)!;
+        Assert.Equal("osBuild", result.Dimension);
+        Assert.Equal(3.2, result.Lift);
+    }
+
+    // ── tracker keyspace (round-trip + key shape) ───────────────────────────
+
+    [Fact]
+    public void RuleRegressionRowKey_PrefixedAndCaseFolded()
+    {
+        Assert.Equal("ruleregression|analyze-net-001",
+            TableHardwareRejectionNotificationTracker.BuildRuleRegressionRowKey("  ANALYZE-NET-001 "));
+    }
+
+    [Fact]
+    public void RuleRegressionEntity_RoundTripsAllFields_IncludingDimension()
+    {
+        var alert = new RuleRegressionAlert
+        {
+            TenantId = TenantA,
+            RuleId = RuleX,
+            RuleTitle = "Network check",
+            WindowFireCount = 15,
+            WindowSessionCount = 100,
+            BaselineFireCount = 10,
+            BaselineSessionCount = 500,
+            WindowRatePct = 15.0,
+            BaselineRatePct = 2.0,
+            Lift = 7.5,
+            WindowStartDate = "2026-07-20",
+            WindowEndDate = "2026-07-26",
+            Dimension = new RuleRegressionDimension
+            {
+                Dimension = "osBuild", Value = "26100.4652",
+                HitCount = 12, HitSharePct = 80.0, AllSharePct = 20.0, Lift = 4.0,
+            },
+            FirstNotifiedAt = Target,
+            LastEvaluatedAt = Target.AddHours(2),
+        };
+
+        var entity = TableHardwareRejectionNotificationTracker.BuildRuleRegressionEntity(TenantA, alert);
+        Assert.Equal(TenantA.ToLowerInvariant(), entity.PartitionKey);
+        Assert.Equal("ruleregression|analyze-net-001", entity.RowKey);
+
+        var mapped = TableHardwareRejectionNotificationTracker.MapToRuleRegressionAlert(entity);
+        Assert.Equal(RuleX, mapped.RuleId);
+        Assert.Equal("Network check", mapped.RuleTitle);
+        Assert.Equal(15, mapped.WindowFireCount);
+        Assert.Equal(100, mapped.WindowSessionCount);
+        Assert.Equal(10, mapped.BaselineFireCount);
+        Assert.Equal(500, mapped.BaselineSessionCount);
+        Assert.Equal(15.0, mapped.WindowRatePct);
+        Assert.Equal(2.0, mapped.BaselineRatePct);
+        Assert.Equal(7.5, mapped.Lift);
+        Assert.Equal("2026-07-20", mapped.WindowStartDate);
+        Assert.Equal("2026-07-26", mapped.WindowEndDate);
+        Assert.Equal(Target, mapped.FirstNotifiedAt);
+        Assert.Equal(Target.AddHours(2), mapped.LastEvaluatedAt);
+        Assert.NotNull(mapped.Dimension);
+        Assert.Equal("osBuild", mapped.Dimension!.Dimension);
+        Assert.Equal(4.0, mapped.Dimension.Lift);
+    }
+
+    [Fact]
+    public void RuleRegressionEntity_TriStates_AbsentColumnsMapToNull_NeverInvented()
+    {
+        var alert = new RuleRegressionAlert
+        {
+            TenantId = TenantA, RuleId = RuleX,
+            Lift = null,       // new signal — no finite lift
+            Dimension = null,  // no clear concentration
+            FirstNotifiedAt = Target, LastEvaluatedAt = Target,
+        };
+
+        var entity = TableHardwareRejectionNotificationTracker.BuildRuleRegressionEntity(TenantA, alert);
+        Assert.False(entity.ContainsKey("Lift"));
+        Assert.False(entity.ContainsKey("DimensionJson"));
+
+        var mapped = TableHardwareRejectionNotificationTracker.MapToRuleRegressionAlert(entity);
+        Assert.Null(mapped.Lift);
+        Assert.Null(mapped.Dimension);
+    }
+
+    // ── wording contracts ───────────────────────────────────────────────────
+
+    [Fact]
+    public void DescribeDimension_UsesCorrelationWording_NeverCausal()
+    {
+        var text = MaintenanceService.DescribeDimension(new RuleRegressionDimension
+        {
+            Dimension = "osBuild", Value = "26100.4652",
+            HitCount = 12, HitSharePct = 87.0, AllSharePct = 22.0, Lift = 4.0,
+        })!;
+
+        Assert.Contains("87% of affected sessions are on osBuild 26100.4652", text);
+        Assert.Contains("vs 22% of all sessions (lift 4x)", text);
+        Assert.Contains("correlated, not necessarily causal", text);
+        Assert.Null(MaintenanceService.DescribeDimension(null));
+    }
+
+    [Fact]
+    public void BuildRegressionMessage_CarriesFullNumbers_AndHonestNoDimensionFallback()
+    {
+        var finding = new RuleRegressionFinding
+        {
+            TenantId = TenantA, RuleId = RuleX, RuleTitle = "Network check",
+            WindowFireCount = 15, WindowSessionCount = 100, WindowRatePct = 15.0,
+            BaselineFireCount = 10, BaselineSessionCount = 500, BaselineRatePct = 2.0,
+            Lift = 7.5, WindowStartDate = "2026-07-20", WindowEndDate = "2026-07-26",
+        };
+
+        var message = MaintenanceService.BuildRegressionMessage(finding, dimensionSummary: null);
+        Assert.Contains("15 of 100 evaluated sessions", message);
+        Assert.Contains("(15%)", message);
+        Assert.Contains("baseline 2% (10/500 over the prior 28 days)", message);
+        Assert.Contains("lift 7.5x", message);
+        Assert.Contains("No clear dimension concentration.", message);
+
+        var newSignal = MaintenanceService.BuildRegressionMessage(
+            new RuleRegressionFinding
+            {
+                RuleId = RuleX, RuleTitle = "Network check",
+                WindowFireCount = 15, WindowSessionCount = 100, WindowRatePct = 15.0,
+                BaselineFireCount = 0, BaselineSessionCount = 500, BaselineRatePct = 0.0,
+                Lift = null, WindowStartDate = "2026-07-20", WindowEndDate = "2026-07-26",
+            },
+            dimensionSummary: "dim summary");
+        Assert.Contains("new signal", newSignal);
+        Assert.Contains("dim summary.", newSignal);
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/ServerActionQueueTests.cs

@@ -1,3 +1,4 @@
+using AutopilotMonitor.Functions.Functions.Sessions;
 using AutopilotMonitor.Functions.Services;
 using AutopilotMonitor.Shared.Models;
 using Newtonsoft.Json;
@@ -11,6 +12,35 @@ namespace AutopilotMonitor.Functions.Tests;
 /// </summary>
 public class ServerActionQueueTests
 {
+    // ── Per-type caller gate (route tier is TenantAdminOrOperator; the function must
+    //    re-gate admin-only types for Operators) ─────────────────────────────────────
+
+    [Theory]
+    // Operator (neither admin flag): only request_diagnostics
+    [InlineData(ServerActionTypes.RequestDiagnostics, false, false, true)]
+    [InlineData(ServerActionTypes.TerminateSession, false, false, false)]
+    [InlineData(ServerActionTypes.RotateConfig, false, false, false)]
+    // Tenant Admin: everything
+    [InlineData(ServerActionTypes.TerminateSession, true, false, true)]
+    [InlineData(ServerActionTypes.RotateConfig, true, false, true)]
+    [InlineData(ServerActionTypes.RequestDiagnostics, true, false, true)]
+    // Global Admin: everything
+    [InlineData(ServerActionTypes.TerminateSession, false, true, true)]
+    [InlineData(ServerActionTypes.RotateConfig, false, true, true)]
+    public void IsTypeAllowedForCaller_Matrix(
+        string actionType, bool isTenantAdmin, bool isGlobalAdmin, bool expected)
+    {
+        Assert.Equal(expected,
+            QueueSessionActionFunction.IsTypeAllowedForCaller(actionType, isTenantAdmin, isGlobalAdmin));
+    }
+
+    [Fact]
+    public void IsTypeAllowedForCaller_IsCaseInsensitive_ForOperator()
+    {
+        Assert.True(QueueSessionActionFunction.IsTypeAllowedForCaller(
+            "Request_Diagnostics", isTenantAdmin: false, isGlobalAdmin: false));
+    }
+
     [Fact]
     public void DeserializePendingActions_NullOrEmpty_ReturnsEmpty()
     {

src/Backend/AutopilotMonitor.Functions.Tests/SessionDeletionWorkerTests.cs

@@ -62,7 +62,7 @@ public class SessionDeletionWorkerTests
         };
         harness.EnqueueMessage(JsonConvert.SerializeObject(envelope), dequeueCount: QueuePollingWorkerBase.DefaultMaxDequeueCount + 1);
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         // Poison queue received the body and main queue deleted the original.
         harness.PoisonQueue.Verify(q => q.SendMessageAsync(
@@ -121,7 +121,7 @@ public class SessionDeletionWorkerTests
                 },
                 "etag-mock"));
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         var poisoned = Assert.Single(harness.OpsEventCalls, e => e.EventType == "SessionDeletionPoisoned");
         Assert.Contains($"\"observedResidualCount\":{capObservation}", poisoned.RawDetails);
@@ -168,7 +168,7 @@ public class SessionDeletionWorkerTests
                 },
                 "etag-mock"));
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         var poisoned = Assert.Single(harness.OpsEventCalls, e => e.EventType == "SessionDeletionPoisoned");
         Assert.Contains("\"observedResidualCount\":3", poisoned.RawDetails);
@@ -307,7 +307,7 @@ public class SessionDeletionWorkerTests
                 },
                 "etag-mock"));
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         var poisoned = Assert.Single(harness.OpsEventCalls, e => e.EventType == "SessionDeletionPoisoned");
         Assert.True(poisoned.RawDetails.Length < 4096,
@@ -357,7 +357,7 @@ public class SessionDeletionWorkerTests
                 return Task.FromResult(Response.FromValue(receipt, new Mock<Response>().Object));
             });
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         var casIdx = callOrder.IndexOf("cas-running-to-poisoned");
         var poisonIdx = callOrder.IndexOf("poison-queue-send");
@@ -402,7 +402,7 @@ public class SessionDeletionWorkerTests
                 CurrentState = SessionDeletionState.Poisoned,
             });
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         // Both CAS attempts fired, and the poison-queue send still completed.
         harness.StorageMock.Verify(s => s.CasSetSessionDeletionStateAsync(
@@ -444,7 +444,7 @@ public class SessionDeletionWorkerTests
                 CurrentState = "Unknown",
             });
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.OpsEventRecorded());
 
         harness.PoisonQueue.Verify(q => q.SendMessageAsync(It.IsAny<string>(), It.IsAny<CancellationToken>()),
             Times.AtLeastOnce);
@@ -457,7 +457,7 @@ public class SessionDeletionWorkerTests
         var harness = new Harness();
         harness.EnqueueMessage("{ \"this is not valid JSON for an envelope: garbage", dequeueCount: 1);
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.MainQueueDeleted());
 
         harness.HandlerMock.Verify(h => h.HandleAsync(
             It.IsAny<SessionDeletionEnvelope>(), It.IsAny<CancellationToken>()),
@@ -491,7 +491,7 @@ public class SessionDeletionWorkerTests
                 await Task.Delay(TimeSpan.FromMilliseconds(350), ct);
             });
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(800));
+        await harness.RunUntilAsync(() => harness.HeartbeatCount() >= 3);
 
         // Heartbeat called at least 3× while the handler was busy (350ms / 50ms ≈ 7, minus a few
         // due to scheduling jitter). One call is enough to prove the heartbeat task is wired.
@@ -512,7 +512,7 @@ public class SessionDeletionWorkerTests
         };
         harness.EnqueueMessage(JsonConvert.SerializeObject(envelope), dequeueCount: 1);
 
-        await harness.RunForAsync(TimeSpan.FromMilliseconds(500));
+        await harness.RunUntilAsync(() => harness.MainQueueDeleted());
 
         harness.HandlerMock.Verify(h => h.HandleAsync(
             It.IsAny<SessionDeletionEnvelope>(), It.IsAny<CancellationToken>()),
@@ -663,7 +663,9 @@ public class SessionDeletionWorkerTests
             opsRepo.Setup(r => r.SaveOpsEventAsync(It.IsAny<OpsEventEntry>()))
                 .Returns<OpsEventEntry>(e =>
                 {
-                    OpsEventCalls.Add(CapturedOpsEvent.From(e));
+                    // Locked so RunUntilAsync's condition poll can read Count from the test
+                    // thread while the worker thread appends.
+                    lock (OpsEventCalls) OpsEventCalls.Add(CapturedOpsEvent.From(e));
                     return Task.CompletedTask;
                 });
             var alertDispatch = new OpsAlertDispatchService(
@@ -721,6 +723,41 @@ public class SessionDeletionWorkerTests
             try { await Sut.StopAsync(CancellationToken.None); }
             catch (OperationCanceledException) { /* expected */ }
         }
+
+        /// <summary>
+        /// De-flake (CI run 30047512186): <see cref="RunForAsync"/>'s fixed wall-clock window
+        /// raced the worker's background poll loop on loaded CI runners — positive assertions
+        /// (delete issued, ops event recorded) could fire before the loop ever processed the
+        /// message. Runs the worker until <paramref name="condition"/> observes the awaited
+        /// side effect (polled every 25 ms); the ceiling only bounds a genuinely broken worker,
+        /// the asserts after this call remain the source of truth. Same pattern as the watchdog
+        /// event-gate de-flake. Only for POSITIVE expectations — Times.Never-style tests keep
+        /// the fixed window (there is no event to wait for).
+        /// </summary>
+        public async Task RunUntilAsync(Func<bool> condition, TimeSpan? timeout = null)
+        {
+            var deadline = DateTime.UtcNow + (timeout ?? TimeSpan.FromSeconds(10));
+            using var cts = new CancellationTokenSource();
+            try { await Sut.StartAsync(cts.Token); }
+            catch (OperationCanceledException) { /* expected */ }
+            while (!condition() && DateTime.UtcNow < deadline)
+                await Task.Delay(25);
+            cts.Cancel();
+            try { await Sut.StopAsync(CancellationToken.None); }
+            catch (OperationCanceledException) { /* expected */ }
+        }
+
+        // Condition probes for RunUntilAsync. Moq's Invocations list snapshots under its
+        // internal lock, so polling it from the test thread while the worker invokes the
+        // mock is safe. OpsEventCalls is guarded by its own lock (see the repo callback).
+        public bool MainQueueDeleted() =>
+            MainQueue.Invocations.Any(i => i.Method.Name == nameof(QueueClient.DeleteMessageAsync));
+        public int HeartbeatCount() =>
+            MainQueue.Invocations.Count(i => i.Method.Name == nameof(QueueClient.UpdateMessageAsync));
+        public bool OpsEventRecorded()
+        {
+            lock (OpsEventCalls) return OpsEventCalls.Count > 0;
+        }
     }
 
     private sealed record AuditEntry(

src/Backend/AutopilotMonitor.Functions.Tests/TenantNotificationAudienceCatalogTests.cs

@@ -13,6 +13,7 @@ public class TenantNotificationAudienceCatalogTests
 {
     [Theory]
     [InlineData("hardware_rejection", NotificationAudience.Admin)]
+    [InlineData("tpm_pss_unsupported", NotificationAudience.Admin)]
     [InlineData("sla_breach", NotificationAudience.Member)]
     [InlineData("sla_consecutive_failures", NotificationAudience.Member)]
     [InlineData("sla_resolved", NotificationAudience.Member)]

src/Backend/AutopilotMonitor.Functions.Tests/TimeAttributionCalculatorTests.cs

@@ -0,0 +1,562 @@
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Golden fixtures for <see cref="TimeAttributionCalculator"/> (F1 PR1, insights spec §F1) —
+/// synthetic sessions only (no customer data), each pinning one semantic decision:
+/// linear phase partition, clock skew, late start, WhiteGlove two-window split, missing /
+/// truncated blocking lists, IME retry occupancy, and reboot gaps. EVERY fixture asserts the
+/// exact-partition invariant: sum(span seconds) + unattributed == wall clock ==
+/// the session's DurationSeconds (truthfulness rule 2 — never normalized, never interpolated).
+/// </summary>
+public class TimeAttributionCalculatorTests
+{
+    private static readonly DateTime T0 = new(2026, 7, 26, 10, 0, 0, DateTimeKind.Utc);
+
+    private const string AppA = "aaaaaaaa-1111-2222-3333-444444444444"; // in blocking list
+    private const string AppB = "bbbbbbbb-1111-2222-3333-444444444444"; // in blocking list, never observed
+    private const string AppC = "cccccccc-1111-2222-3333-444444444444"; // observed, NOT listed
+
+    private static long _seq;
+
+    private static EnrollmentEvent Evt(
+        DateTime ts, string eventType,
+        EnrollmentPhase phase = EnrollmentPhase.Unknown,
+        Dictionary<string, object>? data = null)
+        => new()
+        {
+            EventType = eventType,
+            Timestamp = ts,
+            Sequence = ++_seq,
+            Phase = phase,
+            Data = data ?? new Dictionary<string, object>(),
+        };
+
+    private static Dictionary<string, object> AppData(string appId, string appName, string? state = null)
+    {
+        var d = new Dictionary<string, object> { ["appId"] = appId, ["appName"] = appName };
+        if (state != null) d["state"] = state;
+        return d;
+    }
+
+    /// <summary>Storage-shape blocking lists (EventDataNormalizer yields List&lt;object&gt;).</summary>
+    private static Dictionary<string, object> EspLists(long? win32Count = null, params string[] win32Ids)
+        => new()
+        {
+            ["espTrackedWin32AppIds"] = new List<object>(win32Ids),
+            ["espTrackedUserWin32AppIds"] = new List<object>(),
+            ["espTrackedMsiProductCodes"] = new List<object>(),
+            ["espTrackedModernAppPfns"] = new List<object>(),
+            ["espTrackedWin32Count"] = win32Count ?? win32Ids.Length,
+            ["espTrackedMsiCount"] = 0L,
+            ["espTrackedModernCount"] = 0L,
+        };
+
+    private static TimeAttributionInput Input(
+        List<EnrollmentEvent> events, DateTime completedAt, int durationSeconds,
+        string status = "Succeeded", bool wg = false, DateTime? resumedAt = null)
+        => new()
+        {
+            TenantId = "00000000-0000-0000-0000-0000000000t1",
+            SessionId = "00000000-0000-0000-0000-0000000000s1",
+            Status = status,
+            StartedAt = T0,
+            CompletedAt = completedAt,
+            DurationSeconds = durationSeconds,
+            IsPreProvisioned = wg,
+            ResumedAt = resumedAt,
+            Events = events,
+        };
+
+    /// <summary>The invariant every fixture must satisfy exactly (spec §0 rule 2).</summary>
+    private static void AssertExactPartition(SessionTimeBreakdown b)
+    {
+        Assert.Equal(b.WallClockSeconds, b.Segments.Sum(s => s.Seconds) + b.UnattributedSeconds);
+        Assert.True(b.UnattributedSeconds >= 0, "unattributed must never be negative");
+    }
+
+    // ── golden: linear UDE session ──────────────────────────────────────────
+
+    private static List<EnrollmentEvent> LinearSession()
+        => new()
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(2), "esp_config_detected", data: EspLists(win32Ids: new[] { AppA, AppB })),
+            Evt(T0.AddMinutes(5), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(6), "app_install_started", data: AppData(AppA, "App A")),
+            Evt(T0.AddMinutes(10), "app_install_completed", data: AppData(AppA, "App A", "Installed")),
+            Evt(T0.AddMinutes(11), "app_install_started", data: AppData(AppC, "Untracked App")),
+            Evt(T0.AddMinutes(13), "app_install_completed", data: AppData(AppC, "Untracked App", "Installed")),
+            Evt(T0.AddMinutes(15), "esp_phase_changed", EnrollmentPhase.AccountSetup),
+            Evt(T0.AddMinutes(20), "esp_phase_changed", EnrollmentPhase.AppsUser),
+            Evt(T0.AddMinutes(25), "esp_phase_changed", EnrollmentPhase.FinalizingSetup),
+            Evt(T0.AddMinutes(28), "desktop_arrived"),
+            Evt(T0.AddMinutes(30), "enrollment_complete"),
+        };
+
+    [Fact]
+    public void Linear_Session_PartitionsIntoAllFiveSegments_Exactly()
+    {
+        var b = TimeAttributionCalculator.Compute(
+            Input(LinearSession(), completedAt: T0.AddMinutes(30), durationSeconds: 1800))!;
+
+        AssertExactPartition(b);
+        Assert.Equal(1800, b.WallClockSeconds);
+        Assert.Equal(0, b.UnattributedSeconds);
+        Assert.Equal(TimeAttributionFlags.None, b.QualityFlags);
+        Assert.Equal(TimeAttributionCalculator.CurrentVersion, b.AttributionVersion);
+
+        var totals = b.GetSegmentTotals();
+        Assert.Equal(300, totals[TimeAttributionSegments.DevicePrep]);      // T0 → apps entry (+5m)
+        Assert.Equal(600, totals[TimeAttributionSegments.EspApps]);         // +5m → +15m
+        Assert.Equal(480, totals[TimeAttributionSegments.IdentityHello]);   // AccountSetup 300 + Finalizing 180
+        Assert.Equal(300, totals[TimeAttributionSegments.UserEsp]);         // +20m → +25m
+        Assert.Equal(120, totals[TimeAttributionSegments.DesktopHandoff]);  // desktop_arrived → terminal
+    }
+
+    [Fact]
+    public void Linear_Session_BlockingJoin_IsPositiveEvidenceOnly()
+    {
+        var b = TimeAttributionCalculator.Compute(
+            Input(LinearSession(), completedAt: T0.AddMinutes(30), durationSeconds: 1800))!;
+
+        // App A is listed → blocking, interval from EVENT timestamps (started → completed).
+        var interval = Assert.Single(b.BlockingApps);
+        Assert.Equal(AppA, interval.AppId);
+        Assert.Equal(240, interval.Seconds); // +6m → +10m
+        // App C installed fine but is NOT in the lists → unknown, NEVER "not blocking":
+        // it appears nowhere (no negative claim), and only App A counts as matched.
+        Assert.Equal(1, b.BlockingAppCount);
+        // Occupancy = App A's interval clipped to the esp_apps span.
+        Assert.Equal(240, b.EspAppsOccupancySeconds);
+    }
+
+    // ── skew: backward anchor is dropped, dust stays honest ─────────────────
+
+    [Fact]
+    public void BackwardPhaseAnchor_IsDropped_WithClockSkewFlag()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(10), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            // Replayed declaration whose timestamp runs BACKWARD in canonical order.
+            Evt(T0.AddMinutes(3), "esp_phase_changed", EnrollmentPhase.AccountSetup),
+            Evt(T0.AddMinutes(30), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(30), durationSeconds: 1800))!;
+
+        AssertExactPartition(b);
+        Assert.True(b.QualityFlags.HasFlag(TimeAttributionFlags.ClockSkewDropped));
+        var totals = b.GetSegmentTotals();
+        Assert.Equal(600, totals[TimeAttributionSegments.DevicePrep]);
+        Assert.Equal(1200, totals[TimeAttributionSegments.EspApps]); // runs to window end — no phantom AccountSetup
+        Assert.False(totals.ContainsKey(TimeAttributionSegments.IdentityHello));
+    }
+
+    [Fact]
+    public void SubSecondBoundaries_FloorIntoUnattributedDust_NeverInvented()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddSeconds(90.5), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddSeconds(300), durationSeconds: 300))!;
+
+        AssertExactPartition(b);
+        var totals = b.GetSegmentTotals();
+        Assert.Equal(90, totals[TimeAttributionSegments.DevicePrep]);   // floor(90.5)
+        Assert.Equal(209, totals[TimeAttributionSegments.EspApps]);     // floor(209.5)
+        Assert.Equal(1, b.UnattributedSeconds);                          // the dust, disclosed
+    }
+
+    // ── late start ──────────────────────────────────────────────────────────
+
+    [Fact]
+    public void AgentLateStart_SetsPartialObservation()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddSeconds(1), "agent_late_start"),
+            Evt(T0.AddMinutes(10), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(10), durationSeconds: 600))!;
+
+        AssertExactPartition(b);
+        Assert.True(b.QualityFlags.HasFlag(TimeAttributionFlags.PartialObservation));
+    }
+
+    // ── WhiteGlove: two windows in ONE session, pause never attributed ──────
+
+    [Fact]
+    public void WhiteGlove_SplitsIntoTwoWindows_PauseExcluded_WallClockIsDurationSeconds()
+    {
+        var part1End = T0.AddMinutes(20);
+        var resumedAt = T0.AddDays(2);                 // 2-day pause
+        var completedAt = resumedAt.AddMinutes(10);    // part 2 = 600 s
+        // Stored combined duration: part1 (1200 s) + part2 (600 s) — pause excluded by design.
+        const int duration = 1800;
+
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(1), "esp_config_detected", data: EspLists(win32Ids: new[] { AppA })),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(part1End, "whiteglove_part1_complete"),
+            // — pause —
+            Evt(resumedAt.AddSeconds(30), "agent_started", EnrollmentPhase.AccountSetup),
+            Evt(resumedAt.AddMinutes(5), "esp_phase_changed", EnrollmentPhase.FinalizingSetup),
+            Evt(resumedAt.AddMinutes(8), "desktop_arrived"),
+            Evt(completedAt, "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt, duration, wg: true, resumedAt: resumedAt))!;
+
+        AssertExactPartition(b);
+        // Wall clock is the stored DurationSeconds — NOT CompletedAt − StartedAt (≈ 2 days).
+        Assert.Equal(1800, b.WallClockSeconds);
+        Assert.Equal(TimeAttributionFlags.None, b.QualityFlags); // part1 anchored on the event
+
+        var totals = b.GetSegmentTotals();
+        Assert.Equal(120, totals[TimeAttributionSegments.DevicePrep]);     // W1: T0 → +2m
+        Assert.Equal(1080, totals[TimeAttributionSegments.EspApps]);       // W1: +2m → part1End
+        Assert.Equal(450, totals[TimeAttributionSegments.IdentityHello]);  // W2: +30s → +8m
+        Assert.Equal(120, totals[TimeAttributionSegments.DesktopHandoff]); // W2: +8m → +10m
+        // W2's 30 s resume gap has no phase context — unattributed, not guessed.
+        Assert.Equal(30, b.UnattributedSeconds);
+        // No span may live inside the pause.
+        Assert.DoesNotContain(b.Segments, s => s.StartUtc > part1End && s.EndUtc < resumedAt);
+    }
+
+    [Fact]
+    public void WhiteGlove_WithoutPart1CompleteEvent_FallsBackAndFlags()
+    {
+        var resumedAt = T0.AddDays(1);
+        var completedAt = resumedAt.AddMinutes(10);
+        var lastPart1Event = T0.AddMinutes(20);
+
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(lastPart1Event, "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(resumedAt.AddSeconds(10), "agent_started", EnrollmentPhase.AccountSetup),
+            Evt(completedAt, "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt, durationSeconds: 1800, wg: true, resumedAt: resumedAt))!;
+
+        AssertExactPartition(b);
+        Assert.True(b.QualityFlags.HasFlag(TimeAttributionFlags.WhiteGloveAnchorsIncomplete));
+        Assert.Equal(1800, b.WallClockSeconds);
+    }
+
+    [Fact]
+    public void WhiteGlove_Part2LongerThanStoredDuration_DegradesToSingleWindow_Flagged()
+    {
+        // Stored duration (600 s) contradicts the resume stamp (part 2 alone spans 900 s):
+        // no honest two-window split exists — degrade to the single derived window instead
+        // of inventing a negative part 1.
+        var resumedAt = T0.AddDays(1);
+        var completedAt = resumedAt.AddMinutes(15);
+
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(resumedAt.AddSeconds(5), "agent_started", EnrollmentPhase.AccountSetup),
+            Evt(completedAt, "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt, durationSeconds: 600, wg: true, resumedAt: resumedAt))!;
+
+        AssertExactPartition(b);
+        Assert.Equal(600, b.WallClockSeconds);
+        Assert.True(b.QualityFlags.HasFlag(TimeAttributionFlags.WhiteGloveAnchorsIncomplete));
+    }
+
+    // ── blocking list missing / truncated ───────────────────────────────────
+
+    [Fact]
+    public void NoBlockingLists_MeansUnknown_NotZero()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(3), "app_install_started", data: AppData(AppA, "App A")),
+            Evt(T0.AddMinutes(8), "app_install_completed", data: AppData(AppA, "App A", "Installed")),
+            Evt(T0.AddMinutes(10), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(10), durationSeconds: 600))!;
+
+        AssertExactPartition(b);
+        Assert.True(b.QualityFlags.HasFlag(TimeAttributionFlags.BlockingSetUnknown));
+        Assert.Empty(b.BlockingApps);
+        Assert.Equal(0, b.BlockingAppCount);
+        Assert.Null(b.EspAppsOccupancySeconds); // unknown — NOT 0 (rule 1)
+    }
+
+    [Fact]
+    public void EspConfigWithoutListKeys_IsNotEvidence_LaterEmissionNeverErasesEarlierLists()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(1), "esp_config_detected", data: EspLists(win32Ids: new[] { AppA })),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(3), "app_install_started", data: AppData(AppA, "App A")),
+            Evt(T0.AddMinutes(8), "app_install_completed", data: AppData(AppA, "App A", "Installed")),
+            // Later emission whose probe found no Diagnostics key — no espTracked* keys at all.
+            Evt(T0.AddMinutes(9), "esp_config_detected",
+                data: new Dictionary<string, object> { ["source"] = "registry_firstsync" }),
+            Evt(T0.AddMinutes(10), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(10), durationSeconds: 600))!;
+
+        Assert.False(b.QualityFlags.HasFlag(TimeAttributionFlags.BlockingSetUnknown));
+        Assert.Equal(AppA, Assert.Single(b.BlockingApps).AppId);
+    }
+
+    [Fact]
+    public void TruncatedBlockingList_SetsFlag_AbsentAppsStayUnknown()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            // Uncapped total (60) exceeds the emitted list (1 id) → cap dropped entries.
+            Evt(T0.AddMinutes(1), "esp_config_detected", data: EspLists(win32Count: 60, AppA)),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(3), "app_install_started", data: AppData(AppC, "Maybe Capped App")),
+            Evt(T0.AddMinutes(7), "app_install_completed", data: AppData(AppC, "Maybe Capped App", "Installed")),
+            Evt(T0.AddMinutes(10), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(10), durationSeconds: 600))!;
+
+        Assert.True(b.QualityFlags.HasFlag(TimeAttributionFlags.BlockingSetTruncated));
+        // App C is absent from the (capped) list → it stays UNKNOWN: no interval, no claim.
+        Assert.Empty(b.BlockingApps);
+        Assert.Equal(0, b.BlockingAppCount);
+    }
+
+    // ── retry: occupancy spans to the LAST terminal event (audit Q1) ────────
+
+    [Fact]
+    public void Retry_IntervalRunsFromFirstStart_ToLastTerminalEvent()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(1), "esp_config_detected", data: EspLists(win32Ids: new[] { AppA })),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(6), "app_install_started", data: AppData(AppA, "App A")),
+            Evt(T0.AddMinutes(8), "app_install_failed", data: AppData(AppA, "App A")),
+            Evt(T0.AddMinutes(9), "app_install_started", data: AppData(AppA, "App A")),   // IME retry
+            Evt(T0.AddMinutes(12), "app_install_completed", data: AppData(AppA, "App A", "Installed")),
+            Evt(T0.AddMinutes(15), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(15), durationSeconds: 900))!;
+
+        AssertExactPartition(b);
+        var interval = Assert.Single(b.BlockingApps);
+        // First started (+6m) → LAST terminal (+12m) = 360 s. The payload timing would have
+        // frozen at the first failure (+8m) — event timestamps are the only valid source.
+        Assert.Equal(360, interval.Seconds);
+        Assert.Equal(360, b.EspAppsOccupancySeconds);
+    }
+
+    // ── reboots: spans from event gaps, never the detection event's stamp ───
+
+    [Fact]
+    public void RebootSpan_ComesFromEventGapAroundLastBootUtc_NotFromDetectionTimestamp()
+    {
+        var boot = T0.AddMinutes(12);
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(5), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(10), "app_install_started", data: AppData(AppA, "App A")), // last pre-reboot
+            Evt(T0.AddMinutes(14), "system_reboot_detected",                             // next run, detection-time
+                data: new Dictionary<string, object> { ["lastBootUtc"] = boot.ToString("o") }),
+            // Duplicate detection of the SAME boot (batch replay) — must not double-count.
+            Evt(T0.AddMinutes(14.5), "system_reboot_detected",
+                data: new Dictionary<string, object> { ["lastBootUtc"] = boot.ToString("o") }),
+            Evt(T0.AddMinutes(20), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(20), durationSeconds: 1200))!;
+
+        AssertExactPartition(b);
+        var span = Assert.Single(b.RebootSpans);
+        // Gap = last event before boot (+10m) → first observation at/after boot (+14m).
+        Assert.Equal(T0.AddMinutes(10), span.StartUtc);
+        Assert.Equal(T0.AddMinutes(14), span.EndUtc);
+        Assert.Equal(240, span.Seconds);
+        Assert.Equal(240, b.RebootSeconds);
+        // The reboot began inside the apps phase — cross-cutting annotation, segment unchanged.
+        Assert.Equal(TimeAttributionSegments.EspApps, span.SegmentKey);
+        Assert.Equal(1200 - 300, b.GetSegmentTotals()[TimeAttributionSegments.EspApps]);
+    }
+
+    // ── WhiteGlove pause vs. intervals/reboots (Codex review P1): the pause lies INSIDE
+    // the windows' hull, so hull-clamping alone would count it — a reboot bracketing the
+    // pause must contribute only its in-window flanks, never the pause itself ─────────────
+
+    [Fact]
+    public void WhiteGlove_RebootGapAcrossPause_CountsOnlyInWindowFlanks()
+    {
+        var part1End = T0.AddMinutes(20);
+        var resumedAt = T0.AddDays(2);                 // 2-day pause
+        var completedAt = resumedAt.AddMinutes(10);    // part 2 = 600 s → part 1 = 1200 s
+        var bootInsidePause = T0.AddDays(1);
+
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(part1End, "whiteglove_part1_complete"),
+            // — pause: device sealed, user boots it a day later —
+            Evt(resumedAt.AddSeconds(30), "agent_started", EnrollmentPhase.AccountSetup),
+            Evt(resumedAt.AddSeconds(40), "system_reboot_detected",
+                data: new Dictionary<string, object> { ["lastBootUtc"] = bootInsidePause.ToString("o") }),
+            Evt(completedAt, "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt, durationSeconds: 1800, wg: true, resumedAt: resumedAt))!;
+
+        AssertExactPartition(b);
+        var span = Assert.Single(b.RebootSpans);
+        // Gap = part1End → first post-boot event (+30 s into part 2). In-window seconds are
+        // ONLY the part-2 flank (30 s) — the ~2-day pause never counts (pre-fix: ~172 830 s
+        // of "reboot" in a 30-minute enrollment).
+        Assert.Equal(part1End, span.StartUtc);
+        Assert.Equal(resumedAt.AddSeconds(30), span.EndUtc);
+        Assert.Equal(30, span.Seconds);
+        Assert.Equal(30, b.RebootSeconds);
+    }
+
+    [Fact]
+    public void WhiteGlove_BlockingIntervalAcrossPause_CountsOnlyInWindowSeconds()
+    {
+        var part1End = T0.AddMinutes(20);
+        var resumedAt = T0.AddDays(2);
+        var completedAt = resumedAt.AddMinutes(10);
+
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(1), "esp_config_detected", data: EspLists(win32Ids: new[] { AppA })),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(T0.AddMinutes(10), "app_install_started", data: AppData(AppA, "App A")),
+            Evt(part1End, "whiteglove_part1_complete"),
+            // — pause —
+            Evt(resumedAt.AddSeconds(30), "agent_started", EnrollmentPhase.AccountSetup),
+            // The app's final terminal lands in part 2 (user-scope finish after resume).
+            Evt(resumedAt.AddMinutes(2), "app_install_completed", data: AppData(AppA, "App A", "Installed")),
+            Evt(completedAt, "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt, durationSeconds: 1800, wg: true, resumedAt: resumedAt))!;
+
+        AssertExactPartition(b);
+        var interval = Assert.Single(b.BlockingApps);
+        // Chronological endpoints stay the real event times…
+        Assert.Equal(T0.AddMinutes(10), interval.StartUtc);
+        Assert.Equal(resumedAt.AddMinutes(2), interval.EndUtc);
+        // …but Seconds is the in-window sum: part 1 tail (10 min) + part 2 head (2 min).
+        Assert.Equal(720, interval.Seconds);
+        // Occupancy intersects with the esp_apps span (part 1 only) — pause never leaks in.
+        Assert.Equal(600, b.EspAppsOccupancySeconds);
+    }
+
+    [Fact]
+    public void WhiteGlove_IntervalWhollyInsidePause_MakesNoClaim_AndNoSkewFlag()
+    {
+        var part1End = T0.AddMinutes(20);
+        var resumedAt = T0.AddDays(2);
+        var completedAt = resumedAt.AddMinutes(10);
+
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0, "agent_started", EnrollmentPhase.DeviceSetup),
+            Evt(T0.AddMinutes(1), "esp_config_detected", data: EspLists(win32Ids: new[] { AppA })),
+            Evt(T0.AddMinutes(2), "esp_phase_changed", EnrollmentPhase.AppsDevice),
+            Evt(part1End, "whiteglove_part1_complete"),
+            // Both endpoints inside the pause (post-seal tail activity) — observed, but in no window.
+            Evt(part1End.AddHours(1), "app_install_started", data: AppData(AppA, "App A")),
+            Evt(part1End.AddHours(2), "app_install_completed", data: AppData(AppA, "App A", "Installed")),
+            Evt(resumedAt.AddSeconds(30), "agent_started", EnrollmentPhase.AccountSetup),
+            Evt(completedAt, "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt, durationSeconds: 1800, wg: true, resumedAt: resumedAt))!;
+
+        AssertExactPartition(b);
+        // No in-window observation → no interval claim; and nothing is wrong with the clocks,
+        // so the skew flag must NOT fire (it would exclude the session from fleet medians).
+        Assert.Empty(b.BlockingApps);
+        Assert.False(b.QualityFlags.HasFlag(TimeAttributionFlags.ClockSkewDropped));
+    }
+
+    // ── non-computable sessions ─────────────────────────────────────────────
+
+    [Theory]
+    [InlineData("Incomplete")]   // terminal but deliberately NO DurationSeconds
+    [InlineData("InProgress")]
+    [InlineData("Pending")]
+    public void NonComputableStatuses_ReturnNull(string status)
+    {
+        var input = Input(LinearSession(), completedAt: T0.AddMinutes(30), durationSeconds: 1800, status: status);
+        Assert.Null(TimeAttributionCalculator.Compute(input));
+    }
+
+    [Fact]
+    public void MissingDuration_ReturnsNull_NeverGuessesFromCompletedMinusStarted()
+    {
+        var input = Input(LinearSession(), completedAt: T0.AddMinutes(30), durationSeconds: 0);
+        Assert.Null(TimeAttributionCalculator.Compute(input));
+    }
+
+    // ── no phase evidence at all ────────────────────────────────────────────
+
+    [Fact]
+    public void NoPhaseDeclarations_LeaveEverythingUnattributed()
+    {
+        var events = new List<EnrollmentEvent>
+        {
+            Evt(T0.AddMinutes(1), "network_state_change"),
+            Evt(T0.AddMinutes(9), "enrollment_complete"),
+        };
+
+        var b = TimeAttributionCalculator.Compute(
+            Input(events, completedAt: T0.AddMinutes(10), durationSeconds: 600))!;
+
+        AssertExactPartition(b);
+        Assert.Empty(b.Segments);
+        Assert.Equal(600, b.UnattributedSeconds); // no evidence → no claim (rule 1)
+    }
+}

src/Backend/AutopilotMonitor.Functions.Tests/TimeAttributionPersistenceAndAggregationTests.cs

@@ -0,0 +1,458 @@
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// F1 PR2 (insights spec §F1 "Data &amp; compute changes"): persistence round-trips for the
+/// SessionTimeBreakdowns / TimeAttributionAggregates rows (table-storage-serialization rule —
+/// every property must survive Store→Map, tri-states included), the pure daily-aggregation
+/// core (<see cref="MaintenanceService.BuildTimeAttributionAggregates"/>: clean/flagged/missing
+/// disclosure, per-class bucketing, fixed segment stack, per-app gates) and the what-if bound.
+/// </summary>
+public class TimeAttributionPersistenceAndAggregationTests
+{
+    private static readonly DateTime T0 = new(2026, 7, 26, 10, 0, 0, DateTimeKind.Utc);
+    private const string TenantA = "aaaaaaaa-0000-0000-0000-000000000001";
+    private const string TenantB = "bbbbbbbb-0000-0000-0000-000000000002";
+    private const string AppX = "aaaaaaaa-1111-2222-3333-444444444444";
+    private const string AppY = "bbbbbbbb-1111-2222-3333-444444444444";
+
+    // ── breakdown row round-trip ────────────────────────────────────────────
+
+    [Fact]
+    public void BreakdownEntity_RoundTripsAllFields_IncludingJsonLists()
+    {
+        var breakdown = new SessionTimeBreakdown
+        {
+            TenantId = TenantA,
+            SessionId = "s-1",
+            AttributionVersion = 1,
+            EventCountAtCompute = 261,
+            WallClockSeconds = 1800,
+            UnattributedSeconds = 30,
+            RebootSeconds = 240,
+            BlockingAppCount = 3,
+            EspAppsOccupancySeconds = 600,
+            QualityFlags = TimeAttributionFlags.PartialObservation | TimeAttributionFlags.BlockingSetTruncated,
+            Segments = new List<TimeAttributionSpan>
+            {
+                new() { SegmentKey = TimeAttributionSegments.DevicePrep, StartUtc = T0, EndUtc = T0.AddMinutes(5), Seconds = 300 },
+                new() { SegmentKey = TimeAttributionSegments.EspApps, StartUtc = T0.AddMinutes(5), EndUtc = T0.AddMinutes(30), Seconds = 1500 },
+            },
+            RebootSpans = new List<RebootSpan>
+            {
+                new() { StartUtc = T0.AddMinutes(10), EndUtc = T0.AddMinutes(14), Seconds = 240, SegmentKey = TimeAttributionSegments.EspApps },
+            },
+            BlockingApps = new List<BlockingAppInterval>
+            {
+                new() { AppId = AppX, AppName = "App X", StartUtc = T0.AddMinutes(6), EndUtc = T0.AddMinutes(16), Seconds = 600 },
+            },
+        };
+
+        var mapped = TableStorageService.MapToSessionTimeBreakdown(
+            TableStorageService.BuildSessionTimeBreakdownEntity(breakdown));
+
+        Assert.Equal(TenantA, mapped.TenantId);
+        Assert.Equal("s-1", mapped.SessionId);
+        Assert.Equal(1, mapped.AttributionVersion);
+        // Sweep change signal (Codex review: late batches must trigger a recompute) — a lost
+        // column would read as 0 and recompute forever OR mask real drift.
+        Assert.Equal(261, mapped.EventCountAtCompute);
+        Assert.Equal(1800, mapped.WallClockSeconds);
+        Assert.Equal(30, mapped.UnattributedSeconds);
+        Assert.Equal(240, mapped.RebootSeconds);
+        Assert.Equal(3, mapped.BlockingAppCount);
+        Assert.Equal(600, mapped.EspAppsOccupancySeconds);
+        Assert.Equal(TimeAttributionFlags.PartialObservation | TimeAttributionFlags.BlockingSetTruncated, mapped.QualityFlags);
+
+        Assert.Equal(2, mapped.Segments.Count);
+        Assert.Equal(TimeAttributionSegments.DevicePrep, mapped.Segments[0].SegmentKey);
+        Assert.Equal(T0, mapped.Segments[0].StartUtc);
+        Assert.Equal(300, mapped.Segments[0].Seconds);
+
+        var reboot = Assert.Single(mapped.RebootSpans);
+        Assert.Equal(TimeAttributionSegments.EspApps, reboot.SegmentKey);
+        Assert.Equal(240, reboot.Seconds);
+
+        var app = Assert.Single(mapped.BlockingApps);
+        Assert.Equal(AppX, app.AppId);
+        Assert.Equal("App X", app.AppName);
+        Assert.Equal(600, app.Seconds);
+    }
+
+    [Fact]
+    public void BreakdownEntity_OccupancyTriState_AbsentColumnMapsToNull_NeverZero()
+    {
+        var unknown = new SessionTimeBreakdown
+        {
+            TenantId = TenantA, SessionId = "s-1", WallClockSeconds = 600,
+            EspAppsOccupancySeconds = null, QualityFlags = TimeAttributionFlags.BlockingSetUnknown,
+        };
+
+        var entity = TableStorageService.BuildSessionTimeBreakdownEntity(unknown);
+        Assert.False(entity.ContainsKey("EspAppsOccupancySeconds"));
+        Assert.Null(TableStorageService.MapToSessionTimeBreakdown(entity).EspAppsOccupancySeconds);
+
+        // Zero is a real measured value (blocking set known, nothing occupied) — must survive.
+        var zero = new SessionTimeBreakdown
+        {
+            TenantId = TenantA, SessionId = "s-2", WallClockSeconds = 600, EspAppsOccupancySeconds = 0,
+        };
+        Assert.Equal(0, TableStorageService.MapToSessionTimeBreakdown(
+            TableStorageService.BuildSessionTimeBreakdownEntity(zero)).EspAppsOccupancySeconds);
+    }
+
+    // ── aggregate row round-trip ────────────────────────────────────────────
+
+    [Fact]
+    public void AggregateEntity_RoundTripsAllFields()
+    {
+        var aggregate = new TimeAttributionDailyAggregate
+        {
+            TenantId = TenantA,
+            Date = "2026-07-26",
+            EnrollmentClass = "user_driven",
+            AttributionVersion = 1,
+            CleanSessionCount = 23,
+            FlaggedExcludedCount = 4,
+            MissingBreakdownCount = 1,
+            ComputedAt = T0,
+            SegmentStats = new List<TimeAttributionSegmentStat>
+            {
+                new() { SegmentKey = TimeAttributionSegments.EspApps, MedianSeconds = 600, P75Seconds = 700, P90Seconds = 900 },
+            },
+            TopBlockingApps = new List<TimeAttributionBlockingAppStat>
+            {
+                new()
+                {
+                    AppId = AppX, AppName = "App X", SessionCount = 12,
+                    MedianSeconds = 300, P75Seconds = 400, MedianSavingSeconds = 120, P75SavingSeconds = 200,
+                },
+            },
+        };
+
+        var entity = TableStorageService.BuildTimeAttributionAggregateEntity(aggregate);
+        Assert.Equal("2026-07-26|user_driven", entity.RowKey);
+
+        var mapped = TableStorageService.MapToTimeAttributionAggregate(entity);
+        Assert.Equal(TenantA, mapped.TenantId);
+        Assert.Equal("2026-07-26", mapped.Date);
+        Assert.Equal("user_driven", mapped.EnrollmentClass);
+        Assert.Equal(1, mapped.AttributionVersion);
+        Assert.Equal(23, mapped.CleanSessionCount);
+        Assert.Equal(4, mapped.FlaggedExcludedCount);
+        Assert.Equal(1, mapped.MissingBreakdownCount);
+
+        var segment = Assert.Single(mapped.SegmentStats);
+        Assert.Equal(TimeAttributionSegments.EspApps, segment.SegmentKey);
+        Assert.Equal(600, segment.MedianSeconds);
+        Assert.Equal(900, segment.P90Seconds);
+
+        var app = Assert.Single(mapped.TopBlockingApps);
+        Assert.Equal(AppX, app.AppId);
+        Assert.Equal(12, app.SessionCount);
+        Assert.Equal(120, app.MedianSavingSeconds);
+    }
+
+    // ── enrollment class ────────────────────────────────────────────────────
+
+    [Fact]
+    public void EnrollmentClass_Precedence_WhiteGlove_Wdp_SelfDeploying_UserDriven()
+    {
+        Assert.Equal("whiteglove", TimeAttributionCalculator.GetEnrollmentClass(
+            new SessionSummary { IsPreProvisioned = true, EnrollmentType = "v2", IsSelfDeployingProfile = true }));
+        Assert.Equal("device_preparation", TimeAttributionCalculator.GetEnrollmentClass(
+            new SessionSummary { EnrollmentType = "v2", IsSelfDeployingProfile = true }));
+        Assert.Equal("self_deploying", TimeAttributionCalculator.GetEnrollmentClass(
+            new SessionSummary { IsSelfDeployingProfile = true }));
+        Assert.Equal("user_driven", TimeAttributionCalculator.GetEnrollmentClass(new SessionSummary()));
+    }
+
+    // ── what-if bound ───────────────────────────────────────────────────────
+
+    private static BlockingAppInterval Interval(string appId, DateTime start, DateTime end)
+        => new() { AppId = appId, AppName = appId, StartUtc = start, EndUtc = end, Seconds = (int)(end - start).TotalSeconds };
+
+    [Fact]
+    public void WhatIf_LastEndingApp_SavesTailToPreviousEnd_OthersSaveNothing()
+    {
+        var intervals = new List<BlockingAppInterval>
+        {
+            Interval(AppX, T0, T0.AddMinutes(10)),                 // ends 10:10
+            Interval(AppY, T0.AddMinutes(5), T0.AddMinutes(20)),   // ends 10:20 — critical-path end
+        };
+
+        // Removing Y moves the path end 10:20 → 10:10.
+        Assert.Equal(600, TimeAttributionCalculator.WhatIfSavingSeconds(intervals, AppY));
+        // Removing X changes nothing — Y still ends at 10:20.
+        Assert.Equal(0, TimeAttributionCalculator.WhatIfSavingSeconds(intervals, AppX));
+        // An app without a measured interval yields no claim.
+        Assert.Equal(0, TimeAttributionCalculator.WhatIfSavingSeconds(intervals, "cccccccc-0000-0000-0000-000000000000"));
+    }
+
+    [Fact]
+    public void WhatIf_OnlyBlockingApp_FallsBackToOwnStart()
+    {
+        var intervals = new List<BlockingAppInterval> { Interval(AppX, T0.AddMinutes(5), T0.AddMinutes(20)) };
+        Assert.Equal(900, TimeAttributionCalculator.WhatIfSavingSeconds(intervals, AppX));
+    }
+
+    [Fact]
+    public void WhatIf_IdleGapBeforeLastApp_CountsIntoTheBound()
+    {
+        // A ends 10:00; X runs 10:05 → 10:20. Removing X exposes 10:00 as the path end —
+        // the ESP waited for X to START too, so the bound is 20 min, not 15 (spec formula).
+        var intervals = new List<BlockingAppInterval>
+        {
+            Interval(AppY, T0.AddMinutes(-30), T0),
+            Interval(AppX, T0.AddMinutes(5), T0.AddMinutes(20)),
+        };
+        Assert.Equal(1200, TimeAttributionCalculator.WhatIfSavingSeconds(intervals, AppX));
+    }
+
+    // ── daily aggregation core ──────────────────────────────────────────────
+
+    private static SessionSummary Session(string tenantId, DateTime startedAt, bool wg = false)
+        => new()
+        {
+            TenantId = tenantId,
+            SessionId = Guid.NewGuid().ToString(),
+            StartedAt = startedAt,
+            Status = SessionStatus.Succeeded,
+            IsPreProvisioned = wg,
+        };
+
+    private static SessionTimeBreakdown Breakdown(
+        TimeAttributionFlags flags = TimeAttributionFlags.None,
+        int espAppsSeconds = 600,
+        int unattributed = 0,
+        List<BlockingAppInterval>? apps = null)
+        => new()
+        {
+            AttributionVersion = TimeAttributionCalculator.CurrentVersion,
+            WallClockSeconds = espAppsSeconds + 300 + unattributed,
+            UnattributedSeconds = unattributed,
+            QualityFlags = flags,
+            Segments = new List<TimeAttributionSpan>
+            {
+                new() { SegmentKey = TimeAttributionSegments.DevicePrep, StartUtc = T0, EndUtc = T0.AddSeconds(300), Seconds = 300 },
+                new() { SegmentKey = TimeAttributionSegments.EspApps, StartUtc = T0.AddSeconds(300), EndUtc = T0.AddSeconds(300 + espAppsSeconds), Seconds = espAppsSeconds },
+            },
+            BlockingApps = apps ?? new List<BlockingAppInterval>(),
+        };
+
+    [Fact]
+    public void Aggregates_BucketPerTenantClassDate_AndMirrorGlobal()
+    {
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>
+        {
+            (Session(TenantA, T0), Breakdown(espAppsSeconds: 100)),
+            (Session(TenantA, T0), Breakdown(espAppsSeconds: 300)),
+            (Session(TenantA, T0, wg: true), Breakdown(espAppsSeconds: 900)),   // separate class
+            (Session(TenantB, T0), Breakdown(espAppsSeconds: 500)),             // separate tenant
+        };
+
+        var aggregates = MaintenanceService.BuildTimeAttributionAggregates(
+            pairs, Array.Empty<SessionSummary>(), T0);
+
+        // Tenant A: user_driven + whiteglove; Tenant B: user_driven; global: user_driven + whiteglove.
+        Assert.Equal(5, aggregates.Count);
+
+        var tenantAUde = aggregates.Single(a => a.TenantId == TenantA && a.EnrollmentClass == "user_driven");
+        Assert.Equal("2026-07-26", tenantAUde.Date);
+        Assert.Equal(2, tenantAUde.CleanSessionCount);
+        Assert.Equal(TimeAttributionCalculator.CurrentVersion, tenantAUde.AttributionVersion);
+        // Median of {100, 300} = 100 (nearest-rank: ceil(0.5·2)−1 = index 0 — repo convention).
+        Assert.Equal(100, tenantAUde.SegmentStats.Single(s => s.SegmentKey == TimeAttributionSegments.EspApps).MedianSeconds);
+
+        var wg = aggregates.Single(a => a.TenantId == TenantA && a.EnrollmentClass == "whiteglove");
+        Assert.Equal(1, wg.CleanSessionCount); // classes never mixed
+
+        var globalUde = aggregates.Single(a => a.TenantId == "global" && a.EnrollmentClass == "user_driven");
+        Assert.Equal(3, globalUde.CleanSessionCount); // A(2) + B(1)
+    }
+
+    [Fact]
+    public void Aggregates_FlaggedSessions_AreExcludedFromStats_ButDisclosed()
+    {
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>
+        {
+            (Session(TenantA, T0), Breakdown(espAppsSeconds: 100)),
+            (Session(TenantA, T0), Breakdown(flags: TimeAttributionFlags.PartialObservation, espAppsSeconds: 99999)),
+        };
+
+        var aggregate = MaintenanceService.BuildTimeAttributionAggregates(
+                pairs, Array.Empty<SessionSummary>(), T0)
+            .Single(a => a.TenantId == TenantA);
+
+        Assert.Equal(1, aggregate.CleanSessionCount);
+        Assert.Equal(1, aggregate.FlaggedExcludedCount);
+        // The flagged 99999s outlier must not shape the median.
+        Assert.Equal(100, aggregate.SegmentStats.Single(s => s.SegmentKey == TimeAttributionSegments.EspApps).MedianSeconds);
+    }
+
+    [Fact]
+    public void Aggregates_AllFlaggedDay_YieldsDisclosureRow_WithoutStats()
+    {
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>
+        {
+            (Session(TenantA, T0), Breakdown(flags: TimeAttributionFlags.ClockSkewDropped)),
+        };
+
+        var aggregate = MaintenanceService.BuildTimeAttributionAggregates(
+                pairs, Array.Empty<SessionSummary>(), T0)
+            .Single(a => a.TenantId == TenantA);
+
+        Assert.Equal(0, aggregate.CleanSessionCount);
+        Assert.Equal(1, aggregate.FlaggedExcludedCount);
+        Assert.Empty(aggregate.SegmentStats);   // no clean sessions → no statistical claim
+        Assert.Empty(aggregate.TopBlockingApps);
+    }
+
+    [Fact]
+    public void Aggregates_BlockingOnlyFlags_StayInSegmentStats()
+    {
+        // BlockingSetUnknown/Truncated limit per-app blocking EVIDENCE, not the measured spans.
+        // Gating on any flag starved the fleet aggregates in production (903/1000 breakdowns
+        // carried BlockingSetUnknown, 2026-07-27) — blocking-only flagged sessions must count
+        // as clean; only duration-critical flags exclude.
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>
+        {
+            (Session(TenantA, T0), Breakdown(flags: TimeAttributionFlags.BlockingSetUnknown, espAppsSeconds: 100)),
+            (Session(TenantA, T0), Breakdown(flags: TimeAttributionFlags.BlockingSetTruncated, espAppsSeconds: 300)),
+            (Session(TenantA, T0), Breakdown(
+                flags: TimeAttributionFlags.BlockingSetUnknown | TimeAttributionFlags.PartialObservation,
+                espAppsSeconds: 99999)), // duration-critical bit set → still excluded
+        };
+
+        var aggregate = MaintenanceService.BuildTimeAttributionAggregates(
+                pairs, Array.Empty<SessionSummary>(), T0)
+            .Single(a => a.TenantId == TenantA);
+
+        Assert.Equal(2, aggregate.CleanSessionCount);
+        Assert.Equal(1, aggregate.FlaggedExcludedCount);
+        // Median of {100, 300} = 100 (nearest-rank, lower value — repo convention).
+        Assert.Equal(100, aggregate.SegmentStats.Single(s => s.SegmentKey == TimeAttributionSegments.EspApps).MedianSeconds);
+    }
+
+    [Fact]
+    public void Aggregates_MissingBreakdowns_AreCounted_NeverGuessed()
+    {
+        var missing = new List<SessionSummary> { Session(TenantA, T0) };
+
+        var aggregate = MaintenanceService.BuildTimeAttributionAggregates(
+                Array.Empty<(SessionSummary, SessionTimeBreakdown)>(), missing, T0)
+            .Single(a => a.TenantId == TenantA);
+
+        Assert.Equal(0, aggregate.CleanSessionCount);
+        Assert.Equal(1, aggregate.MissingBreakdownCount);
+        Assert.Empty(aggregate.SegmentStats);
+    }
+
+    [Fact]
+    public void Aggregates_SegmentStack_AlwaysCarriesAllSegmentsPlusUnattributed_AbsentAsZero()
+    {
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>
+        {
+            (Session(TenantA, T0), Breakdown(unattributed: 42)),
+        };
+
+        var aggregate = MaintenanceService.BuildTimeAttributionAggregates(
+                pairs, Array.Empty<SessionSummary>(), T0)
+            .Single(a => a.TenantId == TenantA);
+
+        Assert.Equal(6, aggregate.SegmentStats.Count); // 5 canonical + unattributed — the full stack
+        // The fixture has no user_esp span → honest 0, not absent.
+        Assert.Equal(0, aggregate.SegmentStats.Single(s => s.SegmentKey == TimeAttributionSegments.UserEsp).MedianSeconds);
+        Assert.Equal(42, aggregate.SegmentStats.Single(s => s.SegmentKey == TimeAttributionSegments.Unattributed).MedianSeconds);
+    }
+
+    [Fact]
+    public void Aggregates_RollingDateKey_MergesAllDatesIntoRangeStatistics()
+    {
+        // Two sessions on DIFFERENT days — the rolling row must merge them (a median of
+        // per-day medians would not be the range median; the rolling rows are the honest
+        // range statistics the fleet panel reads).
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>
+        {
+            (Session(TenantA, T0), Breakdown(espAppsSeconds: 100)),
+            (Session(TenantA, T0.AddDays(-3)), Breakdown(espAppsSeconds: 300)),
+        };
+
+        var rolling = MaintenanceService.BuildTimeAttributionAggregates(
+            pairs, Array.Empty<SessionSummary>(), T0, MaintenanceService.RollingAggregateDateKey);
+
+        var row = rolling.Single(a => a.TenantId == TenantA);
+        Assert.Equal(MaintenanceService.RollingAggregateDateKey, row.Date);
+        Assert.Equal(2, row.CleanSessionCount);
+        // Rolling rows sort AFTER every "yyyy-MM-dd|…" key so date-range reads and the
+        // age-based retention filter never touch them.
+        Assert.True(string.CompareOrdinal($"{row.Date}|user_driven", "2026-12-31|zzz") > 0);
+    }
+
+    [Fact]
+    public void Aggregates_PerAppRows_GateAtFiveSessions_AndCarryWhatIfBounds()
+    {
+        var pairs = new List<(SessionSummary, SessionTimeBreakdown)>();
+        // App X in 5 sessions (ends last each time → saving = its full tail over device-prep-only baseline).
+        for (var i = 0; i < 5; i++)
+        {
+            pairs.Add((Session(TenantA, T0), Breakdown(apps: new List<BlockingAppInterval>
+            {
+                Interval(AppX, T0.AddMinutes(5), T0.AddMinutes(15)), // only measured blocking app → saving 600
+            })));
+        }
+        // App Y in only 4 sessions → below the per-app row gate.
+        for (var i = 0; i < 4; i++)
+        {
+            pairs.Add((Session(TenantA, T0), Breakdown(apps: new List<BlockingAppInterval>
+            {
+                Interval(AppY, T0.AddMinutes(5), T0.AddMinutes(6)),
+            })));
+        }
+
+        var aggregate = MaintenanceService.BuildTimeAttributionAggregates(
+                pairs, Array.Empty<SessionSummary>(), T0)
+            .Single(a => a.TenantId == TenantA);
+
+        var row = Assert.Single(aggregate.TopBlockingApps);
+        Assert.Equal(AppX, row.AppId);
+        Assert.Equal(5, row.SessionCount);
+        Assert.Equal(600, row.MedianSeconds);
+        Assert.Equal(600, row.MedianSavingSeconds); // sole blocking app → bound = full interval
+    }
+
+    // ── stale-bucket reconcile key consistency (Codex review) ───────────────
+
+    [Fact]
+    public void RollingAggregateEntity_RowKey_MatchesTargetedDeleteKey()
+    {
+        // The sweep's reconcile deletes by (Date, EnrollmentClass) — the entity builder and the
+        // delete path must agree on the RK shape for BOTH daily and rolling rows, or stale
+        // rolling rows would never be removed.
+        var rolling = new TimeAttributionDailyAggregate
+        {
+            TenantId = TenantA, Date = "rolling30", EnrollmentClass = "whiteglove",
+        };
+        Assert.Equal("rolling30|whiteglove", TableStorageService.BuildTimeAttributionAggregateEntity(rolling).RowKey);
+
+        var daily = new TimeAttributionDailyAggregate
+        {
+            TenantId = TenantA, Date = "2026-07-27", EnrollmentClass = "user_driven",
+        };
+        Assert.Equal("2026-07-27|user_driven", TableStorageService.BuildTimeAttributionAggregateEntity(daily).RowKey);
+    }
+
+    [Fact]
+    public void TimeAttribution_InclusiveWindowStart_YieldsExactly30CalendarDays()
+    {
+        // Codex review: both range ends are inclusive — subtracting the full WindowDays
+        // returned 31 day keys under a "windowDays: 30" label.
+        var today = new System.DateTime(2026, 7, 27);
+        Assert.Equal(today.AddDays(-29),
+            Functions.Metrics.TimeAttributionResponse.InclusiveWindowStart(today, Functions.Metrics.TimeAttributionResponse.WindowDays));
+    }
+}

src/Backend/AutopilotMonitor.Functions/DataAccess/TableStorage/TableHardwareRejectionNotificationTracker.cs

@@ -1,6 +1,8 @@
+using System.Text.Json;
 using AutopilotMonitor.Functions.Services;
 using AutopilotMonitor.Shared;
 using AutopilotMonitor.Shared.DataAccess;
+using AutopilotMonitor.Shared.Models;
 using Azure;
 using Azure.Data.Tables;
 using Microsoft.Extensions.Logging;
@@ -9,7 +11,12 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
 {
     /// <summary>
     /// Table Storage implementation of <see cref="IHardwareRejectionNotificationTracker"/>.
-    /// PartitionKey = tenantId (lowercased), RowKey = "{manufacturer-lower}|{model-lower}" (trimmed).
+    /// PartitionKey = tenantId (lowercased). RowKey encodes the dedup subject:
+    ///   - hardware rejection: "{manufacturer-lower}|{model-lower}" (trimmed)
+    ///   - TPM PSS unsupported: "tpmpss|{serial-lower}" (trimmed; the literal "tpmpss" prefix
+    ///     cannot collide with a hardware key because no real manufacturer is named "tpmpss")
+    ///   - F3 rule regression: "ruleregression|{ruleId-lower}" (trimmed; payload-carrying rows
+    ///     that are refreshed while the episode is active and deleted on re-arm)
     /// Race-safe via AddEntityAsync: Azure Table Storage returns 409 Conflict if the entity already exists.
     /// </summary>
     public class TableHardwareRejectionNotificationTracker : IHardwareRejectionNotificationTracker
@@ -68,14 +75,140 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             }
         }
 
+        public async Task<bool> TryRegisterFirstTpmPssNotificationAsync(string tenantId, string serialNumber)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(serialNumber))
+                return false;
+
+            var partitionKey = tenantId.ToLowerInvariant();
+            var rowKey = BuildTpmPssRowKey(serialNumber);
+
+            var entity = new TableEntity(partitionKey, rowKey)
+            {
+                ["TenantId"] = tenantId,
+                ["SerialNumber"] = serialNumber,
+                ["FirstNotifiedAt"] = DateTime.UtcNow
+            };
+
+            try
+            {
+                await _table.AddEntityAsync(entity);
+                _logger.LogInformation(
+                    "TpmPssUnsupported tracker registered: tenant={TenantId} sn={SerialNumber}",
+                    tenantId, serialNumber);
+                return true;
+            }
+            catch (RequestFailedException ex) when (ex.Status == 409)
+            {
+                // Already notified for this (tenant, serial) — no second bell.
+                return false;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex,
+                    "TpmPssUnsupported tracker failed: tenant={TenantId} sn={SerialNumber}",
+                    tenantId, serialNumber);
+                // On unexpected failure, return false so we do not double-fire if the row was actually written.
+                return false;
+            }
+        }
+
+        public async Task<bool> TryRegisterRuleRegressionAsync(string tenantId, RuleRegressionAlert alert)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(alert?.RuleId))
+                return false;
+
+            try
+            {
+                await _table.AddEntityAsync(BuildRuleRegressionEntity(tenantId, alert!));
+                _logger.LogInformation(
+                    "RuleRegression tracker registered: tenant={TenantId} rule={RuleId}", tenantId, alert!.RuleId);
+                return true;
+            }
+            catch (RequestFailedException ex) when (ex.Status == 409)
+            {
+                // Episode already active — one bell per episode.
+                return false;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex,
+                    "RuleRegression tracker failed: tenant={TenantId} rule={RuleId}", tenantId, alert!.RuleId);
+                // Fail closed so we do not double-fire if the row was actually written.
+                return false;
+            }
+        }
+
+        public async Task RefreshRuleRegressionAsync(string tenantId, RuleRegressionAlert alert)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(alert?.RuleId))
+                return;
+            try
+            {
+                // Replace-mode upsert with the FULL payload incl. the caller-carried
+                // FirstNotifiedAt — the retention sweep re-arms on the ORIGINAL age, so a
+                // refresh must never rejuvenate the row.
+                await _table.UpsertEntityAsync(BuildRuleRegressionEntity(tenantId, alert!), TableUpdateMode.Replace);
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex,
+                    "RuleRegression tracker refresh failed: tenant={TenantId} rule={RuleId} (stale numbers remain)",
+                    tenantId, alert!.RuleId);
+            }
+        }
+
+        public async Task DeleteRuleRegressionAsync(string tenantId, string ruleId)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(ruleId))
+                return;
+            try
+            {
+                await _table.DeleteEntityAsync(tenantId.ToLowerInvariant(), BuildRuleRegressionRowKey(ruleId));
+            }
+            catch (RequestFailedException ex) when (ex.Status == 404)
+            {
+                // Already gone (retention sweep or concurrent pass) — idempotent.
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex,
+                    "RuleRegression tracker delete failed: tenant={TenantId} rule={RuleId} (episode stays until retention)",
+                    tenantId, ruleId);
+            }
+        }
+
+        public async Task<List<RuleRegressionAlert>> GetRuleRegressionsAsync(string tenantId)
+        {
+            var results = new List<RuleRegressionAlert>();
+            if (string.IsNullOrWhiteSpace(tenantId)) return results;
+            try
+            {
+                var partitionKey = tenantId.ToLowerInvariant();
+                // '}' (0x7D) sorts directly after '|' (0x7C) — prefix range over "ruleregression|…".
+                var filter = $"PartitionKey eq '{partitionKey}' and RowKey ge '{RuleRegressionRowKeyPrefix}' and RowKey lt 'ruleregression}}'";
+                await foreach (var entity in _table.QueryAsync<TableEntity>(filter: filter))
+                {
+                    results.Add(MapToRuleRegressionAlert(entity));
+                }
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "RuleRegression tracker list failed for tenant {TenantId}", tenantId);
+            }
+            return results;
+        }
+
         public async Task<int> DeleteOlderThanAsync(DateTime cutoffUtc)
         {
             try
             {
-                // Rows are insert-once (AddEntityAsync, never updated), so FirstNotifiedAt == creation time.
-                // Pruning here resets the lifetime dedup for a given model: if the same model is rejected
-                // again after the cutoff, the bell fires once more — acceptable, since the portal only
-                // surfaces recent rejections anyway.
+                // FirstNotifiedAt == the row's first registration (rule-regression refreshes carry
+                // it forward unchanged). Pruning here resets the dedup for EVERY key space in this
+                // table: a hardware model rejected again after the cutoff rings once more, so does
+                // a TPM-PSS device that reports again ("tpmpss|{serial}" rows), and a rule
+                // regression still active after the window fires a fresh bell — a month-old
+                // still-burning regression is worth a reminder (spec §F3 retention re-arm).
                 var filter = $"FirstNotifiedAt lt datetime'{cutoffUtc:yyyy-MM-ddTHH:mm:ss}Z'";
                 var query = _table.QueryAsync<TableEntity>(filter: filter, select: new[] { "PartitionKey", "RowKey" });
 
@@ -111,5 +244,78 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             var mdl = (model ?? string.Empty).Trim().ToLowerInvariant();
             return $"{mfr}|{mdl}";
         }
+
+        internal static string BuildTpmPssRowKey(string serialNumber)
+        {
+            var sn = (serialNumber ?? string.Empty).Trim().ToLowerInvariant();
+            return $"tpmpss|{sn}";
+        }
+
+        internal const string RuleRegressionRowKeyPrefix = "ruleregression|";
+
+        internal static string BuildRuleRegressionRowKey(string ruleId)
+        {
+            var id = (ruleId ?? string.Empty).Trim().ToLowerInvariant();
+            return $"{RuleRegressionRowKeyPrefix}{id}";
+        }
+
+        // internal static: entity builder + mapper are pinned by round-trip unit tests
+        // (table-storage-serialization rule — every property must survive Store→Map).
+        internal static TableEntity BuildRuleRegressionEntity(string tenantId, RuleRegressionAlert alert)
+        {
+            var entity = new TableEntity(tenantId.ToLowerInvariant(), BuildRuleRegressionRowKey(alert.RuleId))
+            {
+                ["TenantId"] = tenantId,
+                ["RuleId"] = alert.RuleId,
+                ["RuleTitle"] = alert.RuleTitle ?? string.Empty,
+                ["WindowFireCount"] = alert.WindowFireCount,
+                ["WindowSessionCount"] = alert.WindowSessionCount,
+                ["BaselineFireCount"] = alert.BaselineFireCount,
+                ["BaselineSessionCount"] = alert.BaselineSessionCount,
+                ["WindowRatePct"] = alert.WindowRatePct,
+                ["BaselineRatePct"] = alert.BaselineRatePct,
+                ["WindowStartDate"] = alert.WindowStartDate ?? string.Empty,
+                ["WindowEndDate"] = alert.WindowEndDate ?? string.Empty,
+                ["FirstNotifiedAt"] = alert.FirstNotifiedAt,
+                ["LastEvaluatedAt"] = alert.LastEvaluatedAt,
+            };
+            // Tri-states: absent column = "no claim" — a zero-baseline alert has no finite lift,
+            // and a missing dimension means "no clear concentration", never a guessed one.
+            if (alert.Lift.HasValue)
+                entity["Lift"] = alert.Lift.Value;
+            if (alert.Dimension != null)
+                entity["DimensionJson"] = JsonSerializer.Serialize(alert.Dimension);
+            return entity;
+        }
+
+        internal static RuleRegressionAlert MapToRuleRegressionAlert(TableEntity entity)
+        {
+            RuleRegressionDimension? dimension = null;
+            var dimensionJson = entity.GetString("DimensionJson");
+            if (!string.IsNullOrEmpty(dimensionJson))
+            {
+                try { dimension = JsonSerializer.Deserialize<RuleRegressionDimension>(dimensionJson!); }
+                catch (JsonException) { dimension = null; } // corrupt column — degrade to "no claim"
+            }
+
+            return new RuleRegressionAlert
+            {
+                TenantId = entity.GetString("TenantId") ?? entity.PartitionKey,
+                RuleId = entity.GetString("RuleId") ?? entity.RowKey.Substring(RuleRegressionRowKeyPrefix.Length),
+                RuleTitle = entity.GetString("RuleTitle") ?? string.Empty,
+                WindowFireCount = entity.GetInt32("WindowFireCount") ?? 0,
+                WindowSessionCount = entity.GetInt32("WindowSessionCount") ?? 0,
+                BaselineFireCount = entity.GetInt32("BaselineFireCount") ?? 0,
+                BaselineSessionCount = entity.GetInt32("BaselineSessionCount") ?? 0,
+                WindowRatePct = entity.GetDouble("WindowRatePct") ?? 0,
+                BaselineRatePct = entity.GetDouble("BaselineRatePct") ?? 0,
+                Lift = entity.GetDouble("Lift"),
+                WindowStartDate = entity.GetString("WindowStartDate") ?? string.Empty,
+                WindowEndDate = entity.GetString("WindowEndDate") ?? string.Empty,
+                Dimension = dimension,
+                FirstNotifiedAt = entity.GetDateTimeOffset("FirstNotifiedAt")?.UtcDateTime ?? DateTime.MinValue,
+                LastEvaluatedAt = entity.GetDateTimeOffset("LastEvaluatedAt")?.UtcDateTime ?? DateTime.MinValue,
+            };
+        }
     }
 }

src/Backend/AutopilotMonitor.Functions/DataAccess/TableStorage/TableMaintenanceRepository.cs

@@ -99,6 +99,9 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
         public Task DeleteEventSessionIndexEntryAsync(string tenantId, string sessionId)
             => _storage.DeleteEventSessionIndexEntryAsync(tenantId, sessionId);
 
+        public Task<List<(string TenantId, string SessionId)>> GetAllSessionTombstoneKeysAsync()
+            => _storage.GetAllSessionTombstoneKeysAsync();
+
         /// <summary>
         /// Tenant offboarding cascade-worker session enumerator. Yields one row per session
         /// belonging to <paramref name="tenantId"/>. Storage exceptions are intentionally

src/Backend/AutopilotMonitor.Functions/DataAccess/TableStorage/TableMetricsRepository.cs

@@ -111,5 +111,53 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
 
         public Task<int> DeleteRuleStatsOlderThanAsync(DateTime cutoffDate)
             => _storage.DeleteRuleStatsOlderThanAsync(cutoffDate);
+
+        public Task<SessionTimeBreakdown?> ComputeAndStoreSessionTimeBreakdownAsync(string tenantId, string sessionId)
+            => _storage.ComputeAndStoreSessionTimeBreakdownAsync(tenantId, sessionId);
+
+        public Task<SessionTimeBreakdown?> GetSessionTimeBreakdownAsync(string tenantId, string sessionId)
+            => _storage.GetSessionTimeBreakdownAsync(tenantId, sessionId);
+
+        public Task ResolveEspBlockingForSessionAsync(string tenantId, string sessionId)
+            => _storage.ResolveEspBlockingForSessionAsync(tenantId, sessionId);
+
+        public Task<bool> SaveTimeAttributionAggregateAsync(TimeAttributionDailyAggregate aggregate)
+            => _storage.SaveTimeAttributionAggregateAsync(aggregate);
+
+        public Task<List<TimeAttributionDailyAggregate>> GetTimeAttributionAggregatesAsync(string tenantId, DateTime startDate, DateTime endDate)
+            => _storage.GetTimeAttributionAggregatesAsync(tenantId, startDate, endDate);
+
+        public Task<List<TimeAttributionDailyAggregate>> GetRollingTimeAttributionAggregatesAsync(string tenantId)
+            => _storage.GetRollingTimeAttributionAggregatesAsync(tenantId);
+
+        public Task DeleteTimeAttributionAggregateAsync(string tenantId, string dateKey, string enrollmentClass)
+            => _storage.DeleteTimeAttributionAggregateAsync(tenantId, dateKey, enrollmentClass);
+
+        public Task<int> DeleteTimeAttributionAggregatesOlderThanAsync(DateTime cutoffDate)
+            => _storage.DeleteTimeAttributionAggregatesOlderThanAsync(cutoffDate);
+
+        public Task<DeviceHistory?> GetDeviceHistoryAsync(string tenantId, string serialKey)
+            => _storage.GetDeviceHistoryAsync(tenantId, serialKey);
+
+        public Task<List<DeviceHistory>> GetDeviceHistoriesByTenantAsync(string tenantId)
+            => _storage.GetDeviceHistoriesByTenantAsync(tenantId);
+
+        public Task<bool> UpsertDeviceHistoryAsync(DeviceHistory history)
+            => _storage.UpsertDeviceHistoryAsync(history);
+
+        public Task DeleteDeviceHistoryAsync(string tenantId, string serialKey)
+            => _storage.DeleteDeviceHistoryAsync(tenantId, serialKey);
+
+        public Task<bool> SaveDeviceJourneyAggregateAsync(DeviceJourneyDailyAggregate aggregate)
+            => _storage.SaveDeviceJourneyAggregateAsync(aggregate);
+
+        public Task<List<DeviceJourneyDailyAggregate>> GetDeviceJourneyAggregatesAsync(string tenantId, DateTime startDate, DateTime endDate)
+            => _storage.GetDeviceJourneyAggregatesAsync(tenantId, startDate, endDate);
+
+        public Task DeleteDeviceJourneyAggregateAsync(string tenantId, string dateKey)
+            => _storage.DeleteDeviceJourneyAggregateAsync(tenantId, dateKey);
+
+        public Task<int> DeleteDeviceJourneyAggregatesOlderThanAsync(DateTime cutoffDate)
+            => _storage.DeleteDeviceJourneyAggregatesOlderThanAsync(cutoffDate);
     }
 }

src/Backend/AutopilotMonitor.Functions/DataAccess/TableStorage/TableRuleRepository.cs

@@ -30,6 +30,9 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
         public Task<List<RuleResult>> GetRuleResultsAsync(string tenantId, string sessionId)
             => _storage.GetRuleResultsAsync(tenantId, sessionId);
 
+        public Task<List<string>> GetRuleHitSessionIdsAsync(string tenantId, string ruleId, DateTime sinceUtc, int maxResults = 2000)
+            => _storage.GetRuleHitSessionIdsAsync(tenantId, ruleId, sinceUtc, maxResults);
+
         public Task<bool> StoreGatherRuleAsync(GatherRule rule, string tenantId = "global")
             => _storage.StoreGatherRuleAsync(rule, tenantId);
 

src/Backend/AutopilotMonitor.Functions/DataAccess/TableStorage/TableSessionRepository.cs

@@ -84,7 +84,32 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             // otherwise a swept-Incomplete session keeps whatever stale live count it had.
             // Idempotent + fail-soft, so the redundancy is cheap.
             if (transitioned && (status == SessionStatus.Succeeded || status == SessionStatus.Failed || status == SessionStatus.Incomplete))
+            {
                 await _storage.ReconcileSessionCountersAsync(tenantId, sessionId);
+                // F1 PR1: same single-writer seam — join the session's app rows against its
+                // latest esp_config_detected lists exactly once, when the event stream is
+                // complete. Positive evidence only (listed ⇒ EspBlocking=true, absent ⇒ stays
+                // unknown). Idempotent + fail-soft like the counter reconcile above.
+                await _storage.ResolveEspBlockingForSessionAsync(tenantId, sessionId);
+
+                // F1 PR2: compute + persist the time-attribution breakdown once, now that the
+                // terminal write stamped CompletedAt/DurationSeconds and the event stream is
+                // complete. Succeeded/Failed only — Incomplete deliberately stores no duration,
+                // so there is no wall clock to partition (the compute would no-op anyway; the
+                // gate just saves the point-reads). Fail-soft; the 30d maintenance sweep
+                // self-heals any miss.
+                if (status == SessionStatus.Succeeded || status == SessionStatus.Failed)
+                    await _storage.ComputeAndStoreSessionTimeBreakdownAsync(tenantId, sessionId);
+
+                // F2 PR4: upsert this terminal session into its device's history chain at the
+                // same seam, so the session-detail "Attempt N" banner is fresh rather than a
+                // sweep-cycle old. ALL three terminal statuses — Incomplete is a non-successful
+                // journey attempt, unlike the duration-gated breakdown above. Re-terminal
+                // reclassifications (admin mark, late completion) flow through here too and
+                // update the chain entry's status. Fail-soft + idempotent; the maintenance
+                // sweep self-heals any miss.
+                await _storage.UpdateDeviceHistoryForSessionAsync(tenantId, sessionId);
+            }
 
             return transitioned;
         }
@@ -123,6 +148,10 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
         public Task UpdateSessionImeAgentVersionAsync(string tenantId, string sessionId, string version)
             => _storage.UpdateSessionImeAgentVersionAsync(tenantId, sessionId, version);
 
+        public Task UpdateSessionNetworkLatencyAsync(string tenantId, string sessionId,
+            double avgLatencyMs, int requestCount)
+            => _storage.UpdateSessionNetworkLatencyAsync(tenantId, sessionId, avgLatencyMs, requestCount);
+
         public Task<List<SessionSummary>> GetSessionsWithEventCountAboveAsync(string tenantId, int threshold)
             => _storage.GetSessionsWithEventCountAboveAsync(tenantId, threshold);
 

src/Backend/AutopilotMonitor.Functions/Functions/Apps/AppsAnalyticsHelper.cs

@@ -1,5 +1,6 @@
 using System.Collections.Specialized;
 using System.Threading;
+using AutopilotMonitor.Functions.Helpers;
 using AutopilotMonitor.Shared.DataAccess;
 using AutopilotMonitor.Shared.Models;
 
@@ -140,42 +141,39 @@ namespace AutopilotMonitor.Functions.Functions.Apps
 
             var summaries = allSummaries.Where(s => s.StartedAt >= cutoff).ToList();
 
-            var apps = summaries.GroupBy(s => s.AppName).Select(g =>
-            {
-                var total = g.Count();
-                var succeeded = g.Count(s => s.Status == "Succeeded");
-                var failed = g.Count(s => s.Status == "Failed");
-                var completed = g.Where(s => s.Status == "Succeeded").ToList();
-                var failureRate = total > 0 ? Math.Round((double)failed / total * 100, 1) : 0;
+            // F1 PR1 (audit Q3): appId-collision rows merge two distinct apps under one name —
+            // they are excluded from every per-app group and disclosed via collisionExcluded.
+            var collisionExcluded = summaries.Count(s => s.AppIdCollision);
+
+            var apps = summaries.Where(s => !s.AppIdCollision).GroupBy(s => s.AppName).Select(g =>
+            {
+                // PR0 classification (see MetricsMath.IsSkipTerminalState / HasMeasuredDuration):
+                // skips leave the rate + durations; duration stats read measured rows only.
+                var total = g.Count();
+                var succeededAll = g.Where(s => s.Status == "Succeeded").ToList();
+                var skipped = succeededAll.Count(MetricsMath.IsSkipTerminalState);
+                var installed = succeededAll.Where(s => !MetricsMath.IsSkipTerminalState(s)).ToList();
+                var measured = installed.Where(MetricsMath.HasMeasuredDuration).ToList();
+                var failed = g.Count(s => s.Status == "Failed");
+                var failureRate = MetricsMath.TerminalFailureRatePct(failed, installed.Count);
 
-                // Trend: compare failure rate in first half vs second half of window.
-                // Only emit non-stable trend if BOTH halves have >= 5 installs (too noisy otherwise).
                 var firstHalf = g.Where(s => s.StartedAt < midpoint).ToList();
                 var secondHalf = g.Where(s => s.StartedAt >= midpoint).ToList();
-
-                double? trendDelta = null;
-                string trend = "stable";
-                if (firstHalf.Count >= 5 && secondHalf.Count >= 5)
-                {
-                    var fhRate = (double)firstHalf.Count(s => s.Status == "Failed") / firstHalf.Count * 100;
-                    var shRate = (double)secondHalf.Count(s => s.Status == "Failed") / secondHalf.Count * 100;
-                    var delta = Math.Round(shRate - fhRate, 1);
-                    trendDelta = delta;
-                    if (delta < -1) trend = "improving";
-                    else if (delta > 1) trend = "worsening";
-                }
+                var (trend, trendDelta) = ComputeFailureTrend(firstHalf, secondHalf);
 
                 return new
                 {
                     appName = g.Key,
                     appType = g.Select(s => s.AppType).FirstOrDefault(t => !string.IsNullOrEmpty(t)) ?? string.Empty,
                     totalInstalls = total,
-                    succeeded,
+                    succeeded = installed.Count,
+                    skipped,
+                    unmeasured = installed.Count - measured.Count,
                     failed,
                     failureRate,
-                    avgDurationSeconds = completed.Count > 0 ? Math.Round(completed.Average(s => s.DurationSeconds), 0) : 0,
-                    maxDurationSeconds = completed.Count > 0 ? completed.Max(s => s.DurationSeconds) : 0,
-                    avgDownloadBytes = completed.Count > 0 ? (long)completed.Average(s => s.DownloadBytes) : 0,
+                    avgDurationSeconds = measured.Count > 0 ? Math.Round(measured.Average(s => s.DurationSeconds), 0) : 0,
+                    maxDurationSeconds = measured.Count > 0 ? measured.Max(s => s.DurationSeconds) : 0,
+                    avgDownloadBytes = measured.Count > 0 ? (long)measured.Average(s => s.DownloadBytes) : 0,
                     trend,
                     trendDelta,
                     lastSeenAt = g.Max(s => s.CompletedAt ?? s.StartedAt)
@@ -194,6 +192,7 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                     success = true,
                     totalApps = apps.Count,
                     totalInstalls = summaries.Count,
+                    collisionExcluded,
                     windowDays = days,
                     apps
                 };
@@ -210,6 +209,7 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                 success = true,
                 totalApps = apps.Count,
                 totalInstalls = summaries.Count,
+                collisionExcluded,
                 windowDays = days,
                 count = page.Count,
                 offset,
@@ -235,11 +235,16 @@ namespace AutopilotMonitor.Functions.Functions.Apps
             var cutoff = now.AddDays(-days);
             var midpoint = now.AddDays(-days / 2.0);
 
-            var summaries = allSummaries
+            var inWindow = allSummaries
                 .Where(s => string.Equals(s.AppName, appName, StringComparison.OrdinalIgnoreCase)
                             && s.StartedAt >= cutoff)
                 .ToList();
 
+            // F1 PR1 (audit Q3): collision rows under this name mix in a second app's outcomes —
+            // excluded from every stat below, disclosed via collisionExcluded.
+            var collisionExcluded = inWindow.Count(s => s.AppIdCollision);
+            var summaries = inWindow.Where(s => !s.AppIdCollision).ToList();
+
             if (summaries.Count == 0)
             {
                 return new
@@ -248,11 +253,14 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                     appName,
                     appType = string.Empty,
                     windowDays = days,
+                    collisionExcluded,
                     bucket = "day",
                     summary = new
                     {
                         totalInstalls = 0,
                         succeeded = 0,
+                        skipped = 0,
+                        unmeasured = 0,
                         failed = 0,
                         failureRate = 0,
                         avgDurationSeconds = 0,
@@ -272,27 +280,22 @@ namespace AutopilotMonitor.Functions.Functions.Apps
             }
 
             var total = summaries.Count;
-            var succeeded = summaries.Count(s => s.Status == "Succeeded");
+            // PR0 classification — skips leave rate + durations; durations read measured rows only.
+            var succeededAll = summaries.Where(s => s.Status == "Succeeded").ToList();
+            var skipped = succeededAll.Count(MetricsMath.IsSkipTerminalState);
+            var installed = succeededAll.Where(s => !MetricsMath.IsSkipTerminalState(s)).ToList();
+            var measured = installed.Where(MetricsMath.HasMeasuredDuration).ToList();
+            var succeeded = installed.Count;
             var failed = summaries.Count(s => s.Status == "Failed");
-            var completed = summaries.Where(s => s.Status == "Succeeded").ToList();
-            var failureRate = Math.Round((double)failed / total * 100, 1);
-            var avgDurationSeconds = completed.Count > 0 ? Math.Round(completed.Average(s => s.DurationSeconds), 0) : 0;
-            var p95DurationSeconds = Percentile(completed.Select(s => s.DurationSeconds).ToList(), 0.95);
-            var avgDownloadBytes = completed.Count > 0 ? (long)completed.Average(s => s.DownloadBytes) : 0;
+            var failureRate = MetricsMath.TerminalFailureRatePct(failed, succeeded);
+            var avgDurationSeconds = measured.Count > 0 ? Math.Round(measured.Average(s => s.DurationSeconds), 0) : 0;
+            var p95DurationSeconds = Percentile(measured.Select(s => s.DurationSeconds).ToList(), 0.95);
+            var avgDownloadBytes = measured.Count > 0 ? (long)measured.Average(s => s.DownloadBytes) : 0;
 
             // Trend (same rule as list endpoint).
             var firstHalf = summaries.Where(s => s.StartedAt < midpoint).ToList();
             var secondHalf = summaries.Where(s => s.StartedAt >= midpoint).ToList();
-            double? trendDelta = null;
-            string trend = "stable";
-            if (firstHalf.Count >= 5 && secondHalf.Count >= 5)
-            {
-                var fhRate = (double)firstHalf.Count(s => s.Status == "Failed") / firstHalf.Count * 100;
-                var shRate = (double)secondHalf.Count(s => s.Status == "Failed") / secondHalf.Count * 100;
-                trendDelta = Math.Round(shRate - fhRate, 1);
-                if (trendDelta < -1) trend = "improving";
-                else if (trendDelta > 1) trend = "worsening";
-            }
+            var (trend, trendDelta) = ComputeFailureTrend(firstHalf, secondHalf);
 
             var flakinessScore = total > 0
                 ? Math.Round((double)summaries.Count(s => s.AttemptNumber > 1) / total, 3)
@@ -308,12 +311,14 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                 {
                     var vTotal = g.Count();
                     var vFailed = g.Count(s => s.Status == "Failed");
+                    // Same PR0 convention as the top-level rate: skips are not attempts.
+                    var vSucceeded = g.Count(s => s.Status == "Succeeded" && !MetricsMath.IsSkipTerminalState(s));
                     return new
                     {
                         appVersion = g.Key,
                         installs = vTotal,
                         failed = vFailed,
-                        failureRate = vTotal > 0 ? Math.Round((double)vFailed / vTotal * 100, 1) : 0
+                        failureRate = MetricsMath.TerminalFailureRatePct(vFailed, vSucceeded)
                     };
                 })
                 .OrderByDescending(v => v.installs)
@@ -359,12 +364,17 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                     Model = sessionLookup[$"{s.TenantId}|{s.SessionId}"].Model ?? "Unknown"
                 })
                 .GroupBy(x => new { x.Manufacturer, x.Model })
-                .Where(g => g.Count() >= 5)
+                // Sample floor on FINISHED installs — the displayed metric is the terminal-only
+                // failure rate, so a model with many in-flight installs but few outcomes is still
+                // too noisy to rank.
+                .Where(g => g.Count(x => x.Summary.Status == "Failed" || x.Summary.Status == "Succeeded") >= 5)
                 .Select(g =>
                 {
                     var modelTotal = g.Count();
                     var modelFailed = g.Count(x => x.Summary.Status == "Failed");
-                    var modelFailureRate = Math.Round((double)modelFailed / modelTotal * 100, 1);
+                    // PR0 convention: skips are not attempts (same population as the headline rate).
+                    var modelSucceeded = g.Count(x => x.Summary.Status == "Succeeded" && !MetricsMath.IsSkipTerminalState(x.Summary));
+                    var modelFailureRate = MetricsMath.TerminalFailureRatePct(modelFailed, modelSucceeded);
                     var lift = failureRate > 0
                         ? Math.Round(modelFailureRate / failureRate, 2)
                         : 0;
@@ -389,11 +399,14 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                 appName,
                 appType,
                 windowDays = days,
+                collisionExcluded,
                 bucket,
                 summary = new
                 {
                     totalInstalls = total,
                     succeeded,
+                    skipped,
+                    unmeasured = installed.Count - measured.Count,
                     failed,
                     failureRate,
                     avgDurationSeconds,
@@ -496,6 +509,33 @@ namespace AutopilotMonitor.Functions.Functions.Apps
 
         // ── Helpers ─────────────────────────────────────────────────────────
 
+        /// <summary>
+        /// Failure-rate trend between the two halves of the window, shared by the list and
+        /// analytics endpoints. Rates follow the terminal-only convention
+        /// (<see cref="MetricsMath.TerminalFailureRatePct"/>), and a non-stable trend requires
+        /// at least 5 FINISHED installs in BOTH halves — too noisy otherwise, and a half made
+        /// up of in-flight installs has no outcome to trend on.
+        /// </summary>
+        private static (string Trend, double? TrendDelta) ComputeFailureTrend(
+            List<AppInstallSummary> firstHalf, List<AppInstallSummary> secondHalf)
+        {
+            // PR0 convention: skips are not attempts — keep the trend on the same population
+            // as the headline failure rate, or a wave of not-applicable update policies would
+            // fake an "improving" trend.
+            var fhFailed = firstHalf.Count(s => s.Status == "Failed");
+            var fhSucceeded = firstHalf.Count(s => s.Status == "Succeeded" && !MetricsMath.IsSkipTerminalState(s));
+            var shFailed = secondHalf.Count(s => s.Status == "Failed");
+            var shSucceeded = secondHalf.Count(s => s.Status == "Succeeded" && !MetricsMath.IsSkipTerminalState(s));
+            if (fhFailed + fhSucceeded < 5 || shFailed + shSucceeded < 5)
+                return ("stable", null);
+
+            var delta = Math.Round(
+                MetricsMath.TerminalFailureRatePct(shFailed, shSucceeded)
+                - MetricsMath.TerminalFailureRatePct(fhFailed, fhSucceeded), 1);
+            var trend = delta < -1 ? "improving" : delta > 1 ? "worsening" : "stable";
+            return (trend, delta);
+        }
+
         private static List<object> BuildTimeSeries(List<AppInstallSummary> summaries, DateTime cutoff, DateTime now, string bucket)
         {
             var start = bucket == "week" ? StartOfWeek(cutoff) : cutoff.Date;
@@ -523,16 +563,17 @@ namespace AutopilotMonitor.Functions.Functions.Apps
                     var items = kv.Value;
                     var bTotal = items.Count;
                     var bFailed = items.Count(s => s.Status == "Failed");
-                    var bSucceeded = items.Count(s => s.Status == "Succeeded");
-                    var bCompleted = items.Where(s => s.Status == "Succeeded").ToList();
+                    // PR0 convention: skips are not attempts; durations read measured rows only.
+                    var bInstalled = items.Where(s => s.Status == "Succeeded" && !MetricsMath.IsSkipTerminalState(s)).ToList();
+                    var bMeasured = bInstalled.Where(MetricsMath.HasMeasuredDuration).ToList();
                     return (object)new
                     {
                         bucketStart = DateTime.SpecifyKind(kv.Key, DateTimeKind.Utc),
                         installs = bTotal,
-                        succeeded = bSucceeded,
+                        succeeded = bInstalled.Count,
                         failed = bFailed,
-                        failureRate = bTotal > 0 ? Math.Round((double)bFailed / bTotal * 100, 1) : 0,
-                        avgDurationSeconds = bCompleted.Count > 0 ? Math.Round(bCompleted.Average(s => s.DurationSeconds), 0) : 0
+                        failureRate = MetricsMath.TerminalFailureRatePct(bFailed, bInstalled.Count),
+                        avgDurationSeconds = bMeasured.Count > 0 ? Math.Round(bMeasured.Average(s => s.DurationSeconds), 0) : 0
                     };
                 })
                 .ToList();

src/Backend/AutopilotMonitor.Functions/Functions/Config/GetTenantConfigurationFunction.cs

@@ -29,7 +29,8 @@ namespace AutopilotMonitor.Functions.Functions.Config
         {
             try
             {
-                // Authentication + TenantAdminOrGA authorization enforced by PolicyEnforcementMiddleware
+                // Authentication + MemberRead authorization enforced by PolicyEnforcementMiddleware —
+                // Operators/Viewers are admitted for a READ-ONLY view (secrets redacted below).
                 var requestCtx = req.GetRequestContext();
                 var userIdentifier = requestCtx.UserPrincipalName;
 
@@ -39,8 +40,9 @@ namespace AutopilotMonitor.Functions.Functions.Config
 
                 // Only a caller with WRITE authority over THIS target tenant sees the full config including
                 // DiagnosticsBlobSasUrl / webhook URLs / custom headers (they manage those in the Settings
-                // UI). Everyone else admitted to this read tier — a cross-tenant Global Reader, or a
-                // delegated reader viewing one of their managed tenants — gets the secrets redacted (returns
+                // UI). Everyone else admitted to this read tier — an own-tenant Operator/Viewer, a
+                // cross-tenant Global Reader, or a delegated reader viewing one of their managed
+                // tenants — gets the secrets redacted (returns
                 // a copy; the cached instance is never mutated). Redact-by-default: only Global Admin (any
                 // tenant) or the tenant's OWN admin clears secrets. The own-tenant check (target == JWT
                 // tenant) is what keeps a delegated reader — whose target is a DIFFERENT tenant — redacted
@@ -66,8 +68,9 @@ namespace AutopilotMonitor.Functions.Functions.Config
         /// <summary>
         /// Redact-by-default secret clearance for the config GET. Only a caller with WRITE authority over
         /// THIS target tenant sees the unredacted secrets (SAS URLs, webhook URLs, custom headers): a Global
-        /// Admin (any tenant), or the tenant's OWN admin. Everyone else admitted to this read tier — a
-        /// cross-tenant Global Reader, or a delegated reader viewing one of their managed tenants — is
+        /// Admin (any tenant), or the tenant's OWN admin. Everyone else admitted to this read tier — an
+        /// own-tenant Operator/Viewer (read-only Settings view), a cross-tenant Global Reader, or a
+        /// delegated reader viewing one of their managed tenants — is
         /// redacted. The own-tenant check (target == JWT tenant) is what keeps a delegated reader redacted
         /// even when they are an admin of their own home tenant, since their target is a DIFFERENT tenant.
         /// Extracted as a pure static so the security decision is unit-testable without an HttpRequestData mock.

src/Backend/AutopilotMonitor.Functions/Functions/Config/GetTenantFeatureFlagsFunction.cs

@@ -76,6 +76,14 @@ namespace AutopilotMonitor.Functions.Functions.Config
             return new
             {
                 bootstrapTokenEnabled = config.BootstrapTokenEnabled,
+                // Session-detail "Collect Logs" button: whether an on-demand diagnostics upload can
+                // succeed right now (mode not Off + a usable destination). Members below Admin cannot
+                // read the full config, so this boolean is their only signal. Deliberately exposes no
+                // destination detail — just "would an upload work".
+                diagnosticsUploadConfigured =
+                    !string.Equals(config.DiagnosticsUploadMode ?? "Off", "Off", StringComparison.OrdinalIgnoreCase)
+                    && GetAgentConfigFunction.ResolveDiagnosticsUploadEnabled(
+                        config.DiagnosticsBlobSasUrl, config.DiagnosticsUploadDestination),
                 // Drives the "Autopilot Device Validation disabled" dashboard banner
                 // (useTenantSecurityConfig).
                 validateAutopilotDevice = config.ValidateAutopilotDevice,

src/Backend/AutopilotMonitor.Functions/Functions/Infrastructure/AzureMonitorAlertWebhookFunction.cs

@@ -0,0 +1,216 @@
+using System;
+using System.IO;
+using System.Net;
+using System.Security.Cryptography;
+using System.Text;
+using System.Text.Json;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.Services;
+using Microsoft.Azure.Functions.Worker;
+using Microsoft.Azure.Functions.Worker.Http;
+using Microsoft.Extensions.Logging;
+
+namespace AutopilotMonitor.Functions.Functions.Infrastructure
+{
+    /// <summary>
+    /// Receives Azure Monitor alert notifications (Common Alert Schema) from an Action Group
+    /// webhook receiver and re-emits them as ops events, so platform alerts flow through the
+    /// same OpsEventService → OpsAlertDispatchService fan-out (Telegram/Teams/Slack providers +
+    /// the portal ops-events list) as every internally raised operational event.
+    ///
+    /// The delivery path is deliberately independent of the portal: Azure Monitor → this
+    /// function → providers. A portal (SWA) outage cannot suppress these notifications — the
+    /// action group's e-mail receiver additionally stays as the backend-independent fallback,
+    /// because a self-hosted channel cannot alert on its own outage.
+    ///
+    /// Security: anonymous route (Action Group webhook receivers can only call a bare URI — no
+    /// custom headers), gated by a shared secret in the query string compared in constant time
+    /// against the OpsAlertWebhookSecret app setting. Fail-closed: while the setting is unset,
+    /// every request is rejected. Dispatch to providers only happens if an enabled ops alert
+    /// rule for EventType "AzureMonitorAlert" matches the mapped severity — the event itself is
+    /// always stored.
+    /// </summary>
+    public class AzureMonitorAlertWebhookFunction
+    {
+        internal const string SecretSettingName = "OpsAlertWebhookSecret";
+
+        // Common Alert Schema payloads for log-search alerts carry the search query, link URLs
+        // and dimension sets — a few KB in practice. 64 KB is a generous shape gate, not a throttle.
+        internal const int MaxContentLength = 64 * 1024;
+
+        private readonly OpsEventService _opsEventService;
+        private readonly ILogger<AzureMonitorAlertWebhookFunction> _logger;
+
+        public AzureMonitorAlertWebhookFunction(
+            OpsEventService opsEventService,
+            ILogger<AzureMonitorAlertWebhookFunction> logger)
+        {
+            _opsEventService = opsEventService;
+            _logger = logger;
+        }
+
+        [Function("AzureMonitorAlertWebhook")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "post", Route = "ops/alert-webhook")] HttpRequestData req)
+        {
+            var configuredSecret = Environment.GetEnvironmentVariable(SecretSettingName);
+            var providedSecret = System.Web.HttpUtility.ParseQueryString(req.Url.Query ?? string.Empty)["secret"];
+
+            if (!SecretMatches(providedSecret, configuredSecret))
+            {
+                // Deliberately no detail in the response; the warning makes a misconfigured
+                // (unset) secret operator-visible in the logs.
+                _logger.LogWarning(
+                    "Azure Monitor alert webhook rejected: {Reason}",
+                    string.IsNullOrEmpty(configuredSecret) ? "secret not configured (fail-closed)" : "secret mismatch");
+                return req.CreateResponse(HttpStatusCode.Unauthorized);
+            }
+
+            // Bounded read instead of a Body.Length check: with ASP.NET Core integration the
+            // request body is a forward-only stream whose Length property throws
+            // NotSupportedException. Reading at most cap+1 chars detects oversize without ever
+            // needing to know the total size. (Char-based cap ≈ byte cap for a shape gate.)
+            string body;
+            using (var reader = new StreamReader(req.Body, Encoding.UTF8))
+            {
+                var buffer = new char[MaxContentLength + 1];
+                var read = await reader.ReadBlockAsync(buffer, 0, buffer.Length);
+                if (read > MaxContentLength)
+                {
+                    _logger.LogWarning("Azure Monitor alert webhook rejected: payload exceeds {Cap} char cap", MaxContentLength);
+                    return req.CreateResponse(HttpStatusCode.BadRequest);
+                }
+                body = new string(buffer, 0, read);
+            }
+
+            var alert = TryParse(body);
+            if (alert == null)
+            {
+                _logger.LogWarning("Azure Monitor alert webhook rejected: body is not a Common Alert Schema payload");
+                return req.CreateResponse(HttpStatusCode.BadRequest);
+            }
+
+            await _opsEventService.RecordAzureMonitorAlertAsync(
+                alert.AlertRule,
+                MapSeverity(alert.Severity),
+                alert.MonitorCondition,
+                alert.Description,
+                alert.Severity,
+                alert.MonitoringService,
+                alert.TargetResource,
+                alert.MetricValue);
+
+            _logger.LogInformation(
+                "Azure Monitor alert '{AlertRule}' {Condition} ({Severity}) recorded as ops event",
+                alert.AlertRule, alert.MonitorCondition, alert.Severity);
+
+            return req.CreateResponse(HttpStatusCode.OK);
+        }
+
+        /// <summary>
+        /// Constant-time secret comparison. Fail-closed: an unset/empty configured secret never
+        /// matches anything, so the endpoint stays dark until the app setting is provisioned.
+        /// </summary>
+        internal static bool SecretMatches(string? provided, string? configured)
+        {
+            if (string.IsNullOrEmpty(configured) || string.IsNullOrEmpty(provided))
+                return false;
+
+            return CryptographicOperations.FixedTimeEquals(
+                Encoding.UTF8.GetBytes(provided),
+                Encoding.UTF8.GetBytes(configured));
+        }
+
+        /// <summary>
+        /// Maps Azure Monitor severities (Sev0 = most severe … Sev4 = verbose) onto the ops event
+        /// severity scale. Unknown values fall back to Info so a schema drift can never escalate.
+        /// </summary>
+        internal static string MapSeverity(string? azureSeverity) => azureSeverity switch
+        {
+            "Sev0" => Shared.DataAccess.OpsEventSeverity.Critical,
+            "Sev1" => Shared.DataAccess.OpsEventSeverity.Error,
+            "Sev2" => Shared.DataAccess.OpsEventSeverity.Warning,
+            _ => Shared.DataAccess.OpsEventSeverity.Info,
+        };
+
+        /// <summary>
+        /// Parsed subset of the Common Alert Schema. Only <see cref="AlertRule"/> is mandatory;
+        /// everything else is best-effort so a partial payload still produces a usable ops event.
+        /// </summary>
+        internal sealed record ParsedAlert(
+            string AlertRule,
+            string? Severity,
+            string MonitorCondition,
+            string? Description,
+            string? MonitoringService,
+            string? TargetResource,
+            double? MetricValue);
+
+        /// <summary>
+        /// Defensively extracts data.essentials (+ the first condition's metricValue) from a
+        /// Common Alert Schema payload. Returns null when the body is not JSON or carries no
+        /// alert rule name — the two states in which no meaningful ops event can be formed.
+        /// </summary>
+        internal static ParsedAlert? TryParse(string body)
+        {
+            if (string.IsNullOrWhiteSpace(body))
+                return null;
+
+            try
+            {
+                using var doc = JsonDocument.Parse(body);
+                if (!doc.RootElement.TryGetProperty("data", out var data) ||
+                    !data.TryGetProperty("essentials", out var essentials))
+                    return null;
+
+                var alertRule = GetString(essentials, "alertRule");
+                if (string.IsNullOrWhiteSpace(alertRule))
+                    return null;
+
+                var targetResource = GetString(essentials, "targetResourceName");
+                if (string.IsNullOrEmpty(targetResource) &&
+                    essentials.TryGetProperty("alertTargetIDs", out var targets) &&
+                    targets.ValueKind == JsonValueKind.Array && targets.GetArrayLength() > 0 &&
+                    targets[0].ValueKind == JsonValueKind.String)
+                {
+                    // Fall back to the resource name segment of the first target resource ID.
+                    var id = targets[0].GetString() ?? string.Empty;
+                    targetResource = id[(id.LastIndexOf('/') + 1)..];
+                }
+
+                // Log-search alerts carry the observed value under alertContext.condition.allOf[0].
+                double? metricValue = null;
+                if (data.TryGetProperty("alertContext", out var ctx) &&
+                    ctx.ValueKind == JsonValueKind.Object &&
+                    ctx.TryGetProperty("condition", out var condition) &&
+                    condition.ValueKind == JsonValueKind.Object &&
+                    condition.TryGetProperty("allOf", out var allOf) &&
+                    allOf.ValueKind == JsonValueKind.Array && allOf.GetArrayLength() > 0 &&
+                    allOf[0].ValueKind == JsonValueKind.Object &&
+                    allOf[0].TryGetProperty("metricValue", out var mv) &&
+                    mv.ValueKind == JsonValueKind.Number)
+                {
+                    metricValue = mv.GetDouble();
+                }
+
+                return new ParsedAlert(
+                    alertRule!,
+                    GetString(essentials, "severity"),
+                    GetString(essentials, "monitorCondition") ?? "Fired",
+                    GetString(essentials, "description"),
+                    GetString(essentials, "monitoringService"),
+                    targetResource,
+                    metricValue);
+            }
+            catch (JsonException)
+            {
+                return null;
+            }
+        }
+
+        private static string? GetString(JsonElement element, string property)
+            => element.TryGetProperty(property, out var value) && value.ValueKind == JsonValueKind.String
+                ? value.GetString()
+                : null;
+    }
+}

src/Backend/AutopilotMonitor.Functions/Functions/Ingest/ReportDistressFunction.cs

@@ -269,6 +269,36 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
                     }
                 }
 
+                // In-app bell notification on first-ever TPM-PSS incompatibility per (tenant, serial).
+                // Same lifetime-dedup pattern as the hardware-rejection bell above. No webhook —
+                // volume is tiny and the remediation (TPM firmware update / device replacement)
+                // is not time-critical. Reports without a serial number cannot be deduped and
+                // fire no bell; they still appear in the tenant insights panel.
+                if (report.ErrorType == DistressErrorType.TpmPssUnsupported
+                    && !string.IsNullOrEmpty(serialNumber))
+                {
+                    try
+                    {
+                        var isFirst = await _hardwareBellTracker
+                            .TryRegisterFirstTpmPssNotificationAsync(tenantId, serialNumber);
+                        if (isFirst)
+                        {
+                            var deviceLabel = string.Join(" ",
+                                new[] { manufacturer, model }.Where(s => !string.IsNullOrEmpty(s)));
+                            await _tenantNotificationService.CreateNotificationAsync(
+                                tenantId,
+                                type: "tpm_pss_unsupported",
+                                title: "Device with incompatible TPM detected",
+                                message: $"Device {serialNumber}{(deviceLabel.Length > 0 ? $" ({deviceLabel})" : "")} reported a TPM that cannot perform RSA-PSS signing — the agent on this device cannot authenticate. A TPM firmware update or device replacement is required.",
+                                href: "/settings/tenant/hardware-whitelist");
+                        }
+                    }
+                    catch (Exception bellEx)
+                    {
+                        _logger.LogWarning(bellEx, "TpmPssUnsupported bell notification failed for tenant {TenantId}", tenantId);
+                    }
+                }
+
                 // Structured log (Warning, not Critical — data is unverified)
                 _logger.LogWarning(
                     "AgentDistress [{ErrorType}] tenant={TenantId} mfr={Manufacturer} model={Model} sn={SerialNumber} http={HttpStatusCode} ver={AgentVersion} certState={CertSourceState} thumbprint={CertThumbprint}: {Message}",

src/Backend/AutopilotMonitor.Functions/Functions/Metrics/GetDeviceJourneyFunctions.cs

@@ -0,0 +1,358 @@
+using System.Net;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.DataAccess;
+using AutopilotMonitor.Shared.Models;
+using Microsoft.Azure.Functions.Worker;
+using Microsoft.Azure.Functions.Worker.Http;
+using Microsoft.Extensions.Logging;
+
+namespace AutopilotMonitor.Functions.Functions.Metrics
+{
+    /// <summary>
+    /// F2 device-history / First-Time-Right read surfaces (insights spec §F2 "Surfaces", PR5).
+    /// All endpoints serve rows the terminal seam + maintenance sweep maintain — the journey
+    /// semantics live server-side (<see cref="DeviceJourneyCalculator"/>); no consumer re-derives
+    /// them. Durations in chain refs are the sessions' authoritative DurationSeconds verbatim.
+    /// </summary>
+    public class GetDeviceHistoryFunction
+    {
+        private readonly ILogger<GetDeviceHistoryFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+        private readonly ISessionRepository _sessionRepo;
+
+        public GetDeviceHistoryFunction(
+            ILogger<GetDeviceHistoryFunction> logger,
+            IMetricsRepository metricsRepo,
+            ISessionRepository sessionRepo)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+            _sessionRepo = sessionRepo;
+        }
+
+        /// <summary>
+        /// One device's enrollment history by serial number. history=null is a NORMAL outcome
+        /// (unknown device, junk/placeholder serial, or every ref pruned) — the banner simply
+        /// stays hidden. With ?sessionId= the response additionally carries that session's
+        /// attempt number within its journey (server-computed; works for live sessions too via
+        /// the virtual-attempt rule). Serves the session-detail banner and the MCP tool; a
+        /// Global Admin passes ?tenantId= for cross-tenant reads (TenantScoping.QueryParam),
+        /// with the sessionId-based owner fallback mirroring GetSessionTimeAttribution.
+        /// </summary>
+        [Function("GetDeviceHistory")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "metrics/device-history")] HttpRequestData req)
+        {
+            try
+            {
+                // Authentication + MemberRead authorization enforced by PolicyEnforcementMiddleware;
+                // cross-tenant access via TargetTenantId (TenantScoping.QueryParam).
+                var requestCtx = req.GetRequestContext();
+                var query = System.Web.HttpUtility.ParseQueryString(req.Url.Query);
+                var rawSerial = query["serialNumber"];
+                var sessionId = query["sessionId"];
+
+                if (string.IsNullOrWhiteSpace(rawSerial))
+                {
+                    var badRequest = req.CreateResponse(HttpStatusCode.BadRequest);
+                    await badRequest.WriteAsJsonAsync(new { success = false, message = "serialNumber is required" });
+                    return badRequest;
+                }
+
+                // Junk/placeholder serials have no device identity — by design they never have a
+                // chain (disclosed in the fleet aggregate instead), so null here is honest.
+                var serialKey = DeviceJourneyCalculator.NormalizeSerial(rawSerial);
+                var effectiveTenantId = requestCtx.TargetTenantId;
+                var history = serialKey == null
+                    ? null
+                    : await _metricsRepo.GetDeviceHistoryAsync(effectiveTenantId, serialKey);
+
+                // Global-scope fallback (mirrors GetSessionTimeAttribution): a GA/operator opening
+                // a session without an explicit tenantId query param reads their home partition —
+                // resolve the owning tenant via the session and retry once.
+                if (history == null && serialKey != null && requestCtx.HasGlobalScope && !string.IsNullOrEmpty(sessionId))
+                {
+                    var resolvedTenantId = await _sessionRepo.FindSessionTenantIdAsync(sessionId!);
+                    if (resolvedTenantId != null &&
+                        !string.Equals(resolvedTenantId, effectiveTenantId, StringComparison.OrdinalIgnoreCase))
+                    {
+                        effectiveTenantId = resolvedTenantId;
+                        history = await _metricsRepo.GetDeviceHistoryAsync(effectiveTenantId, serialKey);
+                    }
+                }
+
+                // Attempt number for the requesting session — fail-soft: a missing session or an
+                // empty basis yields null, never a guessed position.
+                int? attemptNumber = null;
+                if (history != null && !string.IsNullOrEmpty(sessionId))
+                {
+                    var session = await _sessionRepo.GetSessionAsync(effectiveTenantId, sessionId!);
+                    if (session != null)
+                        attemptNumber = DeviceJourneyCalculator.ComputeAttemptNumber(history.Chain, sessionId!, session.StartedAt);
+                }
+
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(new { success = true, history, attemptNumber });
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching device history");
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>
+    /// Tenant fleet view: daily First-Time-Right rows of the requested window plus their sums,
+    /// the merged attempt histogram, and the "repeat devices" violator list. FTR counts are
+    /// additive, so unlike the median-based time-attribution panel this endpoint honors the
+    /// page's days selector honestly (window rate = sum over daily rows).
+    /// </summary>
+    public class GetDeviceJourneyMetricsFunction
+    {
+        private readonly ILogger<GetDeviceJourneyMetricsFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+        private readonly ISessionRepository _sessionRepo;
+
+        public GetDeviceJourneyMetricsFunction(
+            ILogger<GetDeviceJourneyMetricsFunction> logger,
+            IMetricsRepository metricsRepo,
+            ISessionRepository sessionRepo)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+            _sessionRepo = sessionRepo;
+        }
+
+        [Function("GetDeviceJourneyMetrics")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "metrics/device-journeys")] HttpRequestData req)
+        {
+            try
+            {
+                // Authentication + MemberRead authorization enforced by PolicyEnforcementMiddleware
+                var tenantId = TenantHelper.GetTenantId(req);
+                var query = System.Web.HttpUtility.ParseQueryString(req.Url.Query);
+                var days = DeviceJourneyMetricsResponse.ClampDays(query["days"]);
+
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(await DeviceJourneyMetricsResponse.BuildAsync(
+                    _metricsRepo, _sessionRepo, tenantId, days, includeRepeatDevices: true));
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching device journey metrics");
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>
+    /// Global-admin variant: tenantId filters to one tenant (full response incl. repeat
+    /// devices); absent = the cross-tenant "global" aggregate rows, WITHOUT a repeat-devices
+    /// list — that would require scanning every tenant's DeviceHistories partition, and the
+    /// aggregated view has no per-device drill anyway. repeatDevices=null discloses the gap.
+    /// </summary>
+    public class GetGlobalDeviceJourneyMetricsFunction
+    {
+        private readonly ILogger<GetGlobalDeviceJourneyMetricsFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+        private readonly ISessionRepository _sessionRepo;
+
+        public GetGlobalDeviceJourneyMetricsFunction(
+            ILogger<GetGlobalDeviceJourneyMetricsFunction> logger,
+            IMetricsRepository metricsRepo,
+            ISessionRepository sessionRepo)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+            _sessionRepo = sessionRepo;
+        }
+
+        [Function("GetGlobalDeviceJourneyMetrics")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "global/metrics/device-journeys")] HttpRequestData req)
+        {
+            try
+            {
+                // Authentication + GlobalReadOrAdmin authorization enforced by PolicyEnforcementMiddleware
+                var query = System.Web.HttpUtility.ParseQueryString(req.Url.Query);
+                var tenantIdFilter = query["tenantId"];
+                var days = DeviceJourneyMetricsResponse.ClampDays(query["days"]);
+                var partition = string.IsNullOrWhiteSpace(tenantIdFilter) ? "global" : tenantIdFilter!;
+
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(await DeviceJourneyMetricsResponse.BuildAsync(
+                    _metricsRepo, _sessionRepo, partition, days, includeRepeatDevices: partition != "global"));
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching global device journey metrics");
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>Shared response shape for the tenant and global variants (mirrors TimeAttributionResponse).</summary>
+    internal static class DeviceJourneyMetricsResponse
+    {
+        internal const int DefaultWindowDays = 30;
+        internal const int MaxWindowDays = 180; // aggregate retention — older rows no longer exist
+        internal const int MaxRepeatDevices = 10;
+
+        internal static int ClampDays(string? raw)
+        {
+            if (!int.TryParse(raw, out var days)) return DefaultWindowDays;
+            return Math.Clamp(days, 1, MaxWindowDays);
+        }
+
+        /// <summary>
+        /// "Last N days" = exactly N calendar day keys including today — both range ends are
+        /// inclusive, so subtracting the full N would return N+1 days (Codex review: days=1
+        /// summed yesterday AND today).
+        /// </summary>
+        internal static DateTime InclusiveWindowStart(DateTime today, int days) => today.AddDays(-(days - 1));
+
+        internal static async Task<object> BuildAsync(
+            IMetricsRepository metricsRepo, ISessionRepository sessionRepo,
+            string partition, int days, bool includeRepeatDevices)
+        {
+            var today = DateTime.UtcNow.Date;
+            var daily = await metricsRepo.GetDeviceJourneyAggregatesAsync(partition, InclusiveWindowStart(today, days), today);
+            var totals = SumAggregates(daily);
+
+            List<object>? repeatDevices = null;
+            if (includeRepeatDevices)
+            {
+                var histories = await metricsRepo.GetDeviceHistoriesByTenantAsync(partition);
+                var candidates = SelectRepeatDevices(histories, today.AddDays(-days));
+                repeatDevices = new List<object>(candidates.Count);
+                foreach (var candidate in candidates)
+                {
+                    // Failure reason of the newest failed attempt — a bounded point-read per top
+                    // row (≤10), fail-soft: an unreadable/deleted session just leaves it empty.
+                    var lastFailureReason = string.Empty;
+                    if (candidate.NewestFailed != null)
+                    {
+                        try
+                        {
+                            var failedSession = await sessionRepo.GetSessionAsync(partition, candidate.NewestFailed.SessionId);
+                            lastFailureReason = failedSession?.FailureReason ?? string.Empty;
+                        }
+                        catch
+                        {
+                            // fail-soft — the row renders without a reason
+                        }
+                    }
+                    repeatDevices.Add(new
+                    {
+                        serialNumber = candidate.History.SerialNumber,
+                        manufacturer = candidate.History.Manufacturer,
+                        model = candidate.History.Model,
+                        attempts = candidate.History.CurrentJourneyAttempts,
+                        journeyCount = candidate.History.JourneyCount,
+                        lastStatus = candidate.Newest.Status,
+                        lastSessionId = candidate.Newest.SessionId,
+                        lastStartedAt = candidate.Newest.StartedAt,
+                        lastFailureReason,
+                    });
+                }
+            }
+
+            return new
+            {
+                success = true,
+                windowDays = days,
+                totals,
+                daily = daily.OrderBy(d => d.Date, StringComparer.Ordinal).ToList(),
+                repeatDevices,
+            };
+        }
+
+        /// <summary>Window totals (serialized camelCase by the worker's global serializer, like the model rows).</summary>
+        internal sealed class DeviceJourneyWindowTotals
+        {
+            public int CompletedJourneys { get; init; }
+            public int FirstTimeRight { get; init; }
+
+            /// <summary>Null with zero completed journeys — no rate claim, never 0 (truthfulness rule 1).</summary>
+            public double? FtrRatePct { get; init; }
+
+            public int ExcludedSessions { get; init; }
+            public List<DeviceJourneyAttemptBucket> AttemptHistogram { get; init; } = new();
+        }
+
+        /// <summary>
+        /// Window sums over the daily rows — honest because every field is an additive count
+        /// (rule: the FTR window rate IS the ratio of summed counts). internal static: pinned
+        /// by unit tests.
+        /// </summary>
+        internal static DeviceJourneyWindowTotals SumAggregates(IReadOnlyList<DeviceJourneyDailyAggregate> daily)
+        {
+            var completed = daily.Sum(d => d.CompletedJourneyCount);
+            var firstTimeRight = daily.Sum(d => d.FirstTimeRightCount);
+            return new DeviceJourneyWindowTotals
+            {
+                CompletedJourneys = completed,
+                FirstTimeRight = firstTimeRight,
+                FtrRatePct = completed > 0 ? Math.Round(firstTimeRight * 100.0 / completed, 1) : null,
+                ExcludedSessions = daily.Sum(d => d.ExcludedSessionCount),
+                AttemptHistogram = daily
+                    .SelectMany(d => d.AttemptHistogram)
+                    .GroupBy(b => b.Attempts)
+                    .Select(g => new DeviceJourneyAttemptBucket { Attempts = g.Key, JourneyCount = g.Sum(b => b.JourneyCount) })
+                    .OrderBy(b => b.Attempts)
+                    .ToList(),
+            };
+        }
+
+        internal sealed class RepeatDeviceCandidate
+        {
+            public DeviceHistory History { get; init; } = null!;
+            public DeviceSessionRef Newest { get; init; } = null!;
+            public DeviceSessionRef? NewestFailed { get; init; }
+        }
+
+        /// <summary>
+        /// Violator selection (spec: "repeat devices" table): devices whose LAST journey took
+        /// ≥2 attempts and whose newest terminal session falls inside the window — current pain,
+        /// not ancient history (a device that retried long ago but enrolls cleanly today does
+        /// not belong on a violator list). Ordered by attempts desc, then recency; capped at
+        /// <see cref="MaxRepeatDevices"/>. internal static: pinned by unit tests.
+        /// </summary>
+        internal static List<RepeatDeviceCandidate> SelectRepeatDevices(
+            IReadOnlyList<DeviceHistory> histories, DateTime windowStartUtc)
+        {
+            var candidates = new List<RepeatDeviceCandidate>();
+            foreach (var history in histories)
+            {
+                if (history.CurrentJourneyAttempts < 2 || history.Chain.Count == 0)
+                    continue;
+                var newest = history.Chain[history.Chain.Count - 1];
+                if (newest.StartedAt < windowStartUtc)
+                    continue;
+                candidates.Add(new RepeatDeviceCandidate
+                {
+                    History = history,
+                    Newest = newest,
+                    NewestFailed = history.Chain.LastOrDefault(r =>
+                        string.Equals(r.Status, nameof(SessionStatus.Failed), StringComparison.OrdinalIgnoreCase)),
+                });
+            }
+            return candidates
+                .OrderByDescending(c => c.History.CurrentJourneyAttempts)
+                .ThenByDescending(c => c.Newest.StartedAt)
+                .Take(MaxRepeatDevices)
+                .ToList();
+        }
+    }
+}

src/Backend/AutopilotMonitor.Functions/Functions/Metrics/GetGeographicMetricsFunction.cs

@@ -106,6 +106,13 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 var first = group.First();
                 var succeeded = group.Where(s => s.Status == SessionStatus.Succeeded).ToList();
                 var failed = group.Where(s => s.Status == SessionStatus.Failed).ToList();
+                // Success rate is an outcome quota over finished enrollments only
+                // (Succeeded + Failed), matching the SLA convention. In-flight sessions
+                // (InProgress/Pending/Stalled/AwaitingUser) and Incomplete (terminal,
+                // non-failure) must not dilute it — a site mid-rollout would otherwise
+                // show a poor rate despite zero actual failures. 0 finished → 0; the
+                // frontend renders "—" off Succeeded+Failed instead of a 0% badge.
+                var finished = succeeded.Count + failed.Count;
 
                 // Duration stats from succeeded sessions with valid duration
                 var durations = succeeded
@@ -146,6 +153,18 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                     .Where(s => appsBySession.ContainsKey(s.SessionId))
                     .Count(s => appsBySession[s.SessionId].Any(a => a.DoDownloadMode >= 0));
 
+                // Agent→backend API latency: weight each session's stored average by its request
+                // count so the location figure equals Σ(latency·requests)/Σ(requests) — the exact
+                // whole-population average, not an average of session averages. Sessions from
+                // agents predating the feature carry no data and drop out.
+                var latencySessions = group
+                    .Where(s => s.AvgApiLatencyMs is > 0 && s.ApiRequestCount is > 0)
+                    .ToList();
+                var latencyRequestSum = latencySessions.Sum(s => (double)s.ApiRequestCount!.Value);
+                var avgApiLatency = latencyRequestSum > 0
+                    ? latencySessions.Sum(s => s.AvgApiLatencyMs!.Value * s.ApiRequestCount!.Value) / latencyRequestSum
+                    : 0;
+
                 locations.Add(new LocationMetrics
                 {
                     LocationKey = group.Key,
@@ -156,7 +175,7 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                     SessionCount = group.Count(),
                     Succeeded = succeeded.Count,
                     Failed = failed.Count,
-                    SuccessRate = group.Count() > 0 ? Math.Round((double)succeeded.Count / group.Count() * 100, 1) : 0,
+                    SuccessRate = finished > 0 ? Math.Round((double)succeeded.Count / finished * 100, 1) : 0,
                     AvgDurationMinutes = Math.Round(avgDuration, 1),
                     MedianDurationMinutes = Math.Round(medianDuration, 1),
                     P95DurationMinutes = Math.Round(p95Duration, 1),
@@ -164,6 +183,8 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                     MinutesPerApp = Math.Round(minutesPerApp, 2),
                     AvgThroughputBytesPerSec = Math.Round(avgThroughput, 0),
                     TotalDownloadBytes = totalBytes,
+                    AvgApiLatencyMs = Math.Round(avgApiLatency, 0),
+                    ApiLatencySessionCount = latencySessions.Count,
                     // Delivery Optimization
                     DoSessionCount = doSessionCount,
                     AvgDoPercentPeerCaching = Math.Round(doAgg.PercentPeerCaching, 1),
@@ -195,6 +216,17 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 .Select(l => l.AvgThroughputBytesPerSec).ToList();
             var globalAvgThroughput = allThroughputs.Count > 0 ? allThroughputs.Average() : 0;
 
+            // Global API latency: request-weighted over ALL geo sessions (not location averages),
+            // so big sites weigh in proportionally — same Σ(latency·requests)/Σ(requests) math as
+            // the per-location figure.
+            var globalLatencySessions = geoSessions
+                .Where(s => s.AvgApiLatencyMs is > 0 && s.ApiRequestCount is > 0)
+                .ToList();
+            var globalLatencyRequestSum = globalLatencySessions.Sum(s => (double)s.ApiRequestCount!.Value);
+            var globalAvgApiLatency = globalLatencyRequestSum > 0
+                ? globalLatencySessions.Sum(s => s.AvgApiLatencyMs!.Value * s.ApiRequestCount!.Value) / globalLatencyRequestSum
+                : 0;
+
             // Global DO metrics (weighted by bytes, not simple average)
             var globalDoPeers = locations.Sum(l => l.TotalDoBytesFromPeers);
             var globalDoHttp = locations.Sum(l => l.TotalDoBytesFromHttp);
@@ -212,6 +244,10 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 loc.ThroughputVsGlobalPct = globalAvgThroughput > 0
                     ? Math.Round((loc.AvgThroughputBytesPerSec - globalAvgThroughput) / globalAvgThroughput * 100, 1) : 0;
 
+                // API latency vs global (positive = slower/farther from the backend region)
+                loc.ApiLatencyVsGlobalPct = globalAvgApiLatency > 0 && loc.AvgApiLatencyMs > 0
+                    ? Math.Round((loc.AvgApiLatencyMs - globalAvgApiLatency) / globalAvgApiLatency * 100, 1) : 0;
+
                 // AppLoadScore: normalize minutesPerApp to global median = 100
                 loc.AppLoadScore = globalMedianMinutesPerApp > 0
                     ? Math.Round(loc.MinutesPerApp / globalMedianMinutesPerApp * 100, 0) : 0;
@@ -236,6 +272,7 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                     AvgMinutesPerApp = Math.Round(globalAvgMinutesPerApp, 2),
                     AvgThroughputBytesPerSec = Math.Round(globalAvgThroughput, 0),
                     StdDevDurationMinutes = Math.Round(globalStdDev, 1),
+                    AvgApiLatencyMs = Math.Round(globalAvgApiLatency, 0),
                     AvgDoPercentPeerCaching = Math.Round(globalDoPct, 1),
                     TotalDoBytesFromPeers = globalDoPeers,
                     TotalDoBytesFromHttp = globalDoHttp

src/Backend/AutopilotMonitor.Functions/Functions/Metrics/GetImeVersionHistoryFunction.cs

@@ -1,6 +1,7 @@
 using System.Net;
 using AutopilotMonitor.Functions.Helpers;
 using AutopilotMonitor.Shared.DataAccess;
+using AutopilotMonitor.Shared.Models;
 using Microsoft.Azure.Functions.Worker;
 using Microsoft.Azure.Functions.Worker.Http;
 using Microsoft.Extensions.Logging;
@@ -34,24 +35,7 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 var versions = await _sessionRepo.GetImeVersionHistoryAsync();
 
                 var response = req.CreateResponse(HttpStatusCode.OK);
-
-                if (requestCtx.HasGlobalScope)
-                {
-                    await response.WriteAsJsonAsync(versions);
-                }
-                else
-                {
-                    // Strip sensitive cross-tenant data for non-Global Admins
-                    var redacted = versions.Select(v => new
-                    {
-                        v.Version,
-                        v.FirstSeenAt,
-                        v.LastSeenAt,
-                        v.SessionCount
-                    });
-                    await response.WriteAsJsonAsync(redacted);
-                }
-
+                await response.WriteAsJsonAsync(BuildResponsePayload(versions, requestCtx.HasGlobalScope));
                 return response;
             }
             catch (Exception ex)
@@ -62,5 +46,35 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 return errorResponse;
             }
         }
+
+        /// <summary>
+        /// Projects the archive for the caller's scope.
+        ///
+        /// Extracted from <see cref="Run"/> so the cross-tenant redaction boundary is
+        /// directly testable: this suite has no HTTP-level harness, so an inline projection
+        /// is unreachable from a test — and an unguarded redaction is exactly what a later
+        /// refactor opens silently.
+        ///
+        /// FirstSeenTenantId and FirstSeenSessionId name ANOTHER tenant's device and are
+        /// dropped for non-global callers. Version, dates and SessionCount deliberately stay:
+        /// the archive is a platform-wide view of Microsoft's IME rollout, and the route is
+        /// MemberRead precisely so any tenant member can read that view.
+        ///
+        /// Returned as <see cref="object"/> on purpose — System.Text.Json serializes an
+        /// `object` declared type using the RUNTIME type, so both branches keep the exact
+        /// wire format they had while the projection was inline.
+        /// </summary>
+        public static object BuildResponsePayload(
+            IEnumerable<ImeVersionHistoryEntry> versions, bool hasGlobalScope)
+        {
+            if (hasGlobalScope)
+            {
+                return versions.ToList();
+            }
+
+            return versions
+                .Select(v => new { v.Version, v.FirstSeenAt, v.LastSeenAt, v.SessionCount })
+                .ToList();
+        }
     }
 }

src/Backend/AutopilotMonitor.Functions/Functions/Metrics/GetTimeAttributionFunctions.cs

@@ -0,0 +1,196 @@
+using System.Net;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.DataAccess;
+using Microsoft.Azure.Functions.Worker;
+using Microsoft.Azure.Functions.Worker.Http;
+using Microsoft.Extensions.Logging;
+
+namespace AutopilotMonitor.Functions.Functions.Metrics
+{
+    /// <summary>
+    /// F1 time-attribution read surfaces (insights spec §F1 "Surfaces", PR3). All three
+    /// endpoints serve PRE-COMPUTED rows — the per-session breakdown written at the terminal
+    /// transition and the aggregate rows the maintenance sweep maintains. No request-time
+    /// derivation: what the calculator computed is what every consumer (web, MCP) sees.
+    /// </summary>
+    public class GetSessionTimeAttributionFunction
+    {
+        private readonly ILogger<GetSessionTimeAttributionFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+        private readonly ISessionRepository _sessionRepo;
+
+        public GetSessionTimeAttributionFunction(
+            ILogger<GetSessionTimeAttributionFunction> logger,
+            IMetricsRepository metricsRepo,
+            ISessionRepository sessionRepo)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+            _sessionRepo = sessionRepo;
+        }
+
+        /// <summary>
+        /// Per-session breakdown for the session-detail attribution lane. A missing row is a
+        /// NORMAL outcome (pre-feature session, non-terminal, Incomplete — no wall clock) and
+        /// returns success with breakdown=null so the UI simply omits the lane.
+        /// </summary>
+        [Function("GetSessionTimeAttribution")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "sessions/{sessionId}/time-attribution")] HttpRequestData req,
+            string sessionId)
+        {
+            try
+            {
+                // Authentication + MemberRead authorization enforced by PolicyEnforcementMiddleware;
+                // cross-tenant access via TargetTenantId (TenantScoping.QueryParam).
+                var requestCtx = req.GetRequestContext();
+
+                var breakdown = await _metricsRepo.GetSessionTimeBreakdownAsync(requestCtx.TargetTenantId, sessionId);
+
+                // Global-scope fallback (mirrors GetSessionEventsFunction): a GA/operator opening
+                // a session without an explicit tenantId query param reads their home partition —
+                // resolve the owning tenant and retry once.
+                if (breakdown == null && requestCtx.HasGlobalScope)
+                {
+                    var resolvedTenantId = await _sessionRepo.FindSessionTenantIdAsync(sessionId);
+                    if (resolvedTenantId != null &&
+                        !string.Equals(resolvedTenantId, requestCtx.TargetTenantId, StringComparison.OrdinalIgnoreCase))
+                    {
+                        breakdown = await _metricsRepo.GetSessionTimeBreakdownAsync(resolvedTenantId, sessionId);
+                    }
+                }
+
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(new { success = true, breakdown });
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching time attribution for session {SessionId}", sessionId);
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>
+    /// Tenant fleet view: the rolling 30-day range statistics per enrollment class (the panel's
+    /// stacked medians + top blocking apps) plus the daily rows for the per-day trend. The range
+    /// window is FIXED at the sweep's 30 days — daily medians cannot be merged into an arbitrary
+    /// range honestly, so no days parameter is offered.
+    /// </summary>
+    public class GetTimeAttributionMetricsFunction
+    {
+        private readonly ILogger<GetTimeAttributionMetricsFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+
+        public GetTimeAttributionMetricsFunction(
+            ILogger<GetTimeAttributionMetricsFunction> logger,
+            IMetricsRepository metricsRepo)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+        }
+
+        [Function("GetTimeAttributionMetrics")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "metrics/time-attribution")] HttpRequestData req)
+        {
+            try
+            {
+                // Authentication + MemberRead authorization enforced by PolicyEnforcementMiddleware
+                var tenantId = TenantHelper.GetTenantId(req);
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(await TimeAttributionResponse.BuildAsync(_metricsRepo, tenantId));
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching time attribution metrics");
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>Global-admin variant: tenantId filters to one tenant; absent = cross-tenant "global" rows.</summary>
+    public class GetGlobalTimeAttributionMetricsFunction
+    {
+        private readonly ILogger<GetGlobalTimeAttributionMetricsFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+
+        public GetGlobalTimeAttributionMetricsFunction(
+            ILogger<GetGlobalTimeAttributionMetricsFunction> logger,
+            IMetricsRepository metricsRepo)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+        }
+
+        [Function("GetGlobalTimeAttributionMetrics")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "global/metrics/time-attribution")] HttpRequestData req)
+        {
+            try
+            {
+                // Authentication + GlobalReadOrAdmin authorization enforced by PolicyEnforcementMiddleware
+                var query = System.Web.HttpUtility.ParseQueryString(req.Url.Query);
+                var tenantIdFilter = query["tenantId"];
+                var partition = string.IsNullOrWhiteSpace(tenantIdFilter) ? "global" : tenantIdFilter!;
+
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(await TimeAttributionResponse.BuildAsync(_metricsRepo, partition));
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching global time attribution metrics");
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>Shared response shape for the tenant and global variants (mirrors the MetricsMath single-source pattern).</summary>
+    internal static class TimeAttributionResponse
+    {
+        internal const int WindowDays = 30;
+
+        /// <summary>
+        /// Rolling rows are rewritten by every 2h sweep as long as the partition has window
+        /// sessions; a partition that went fully quiet stops being rewritten, so its rows would
+        /// otherwise serve arbitrarily old numbers as "last 30 days" (Codex review). 48h
+        /// tolerates a full day of sweep outage before the panel honestly goes empty.
+        /// </summary>
+        internal static readonly TimeSpan RollingMaxAge = TimeSpan.FromHours(48);
+
+        /// <summary>
+        /// "Last N days" = exactly N calendar day keys including today — both range ends are
+        /// inclusive, so subtracting the full N would return N+1 days (Codex review).
+        /// </summary>
+        internal static DateTime InclusiveWindowStart(DateTime today, int days) => today.AddDays(-(days - 1));
+
+        internal static async Task<object> BuildAsync(IMetricsRepository metricsRepo, string partition)
+        {
+            var freshCutoff = DateTime.UtcNow - RollingMaxAge;
+            var rolling = (await metricsRepo.GetRollingTimeAttributionAggregatesAsync(partition))
+                .Where(r => r.ComputedAt >= freshCutoff)
+                .ToList();
+            var today = DateTime.UtcNow.Date;
+            var daily = await metricsRepo.GetTimeAttributionAggregatesAsync(partition, InclusiveWindowStart(today, WindowDays), today);
+
+            return new
+            {
+                success = true,
+                windowDays = WindowDays,
+                // Range statistics per enrollment class (never mixed) — the UI gates rendering
+                // at cleanSessionCount >= 20 and shows "insufficient data (n=…)" below it.
+                classes = rolling.OrderBy(r => r.EnrollmentClass, StringComparer.Ordinal).ToList(),
+                daily = daily.OrderBy(d => d.Date, StringComparer.Ordinal).ToList(),
+            };
+        }
+    }
+}

src/Backend/AutopilotMonitor.Functions/Functions/Metrics/RuleStatsFunction.cs

@@ -16,13 +16,16 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
     {
         private readonly ILogger<RuleStatsFunction> _logger;
         private readonly IMetricsRepository _metricsRepo;
+        private readonly IHardwareRejectionNotificationTracker _notificationTracker;
 
         public RuleStatsFunction(
             ILogger<RuleStatsFunction> logger,
-            IMetricsRepository metricsRepo)
+            IMetricsRepository metricsRepo,
+            IHardwareRejectionNotificationTracker notificationTracker)
         {
             _logger = logger;
             _metricsRepo = metricsRepo;
+            _notificationTracker = notificationTracker;
         }
 
         /// <summary>
@@ -81,6 +84,9 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 var result = new
                 {
                     rules = aggregated,
+                    // F3 PR6: active regression episodes (tracker rows) — the rules-page badge
+                    // and MCP get_rule_stats read them from here. Empty list = nothing regressed.
+                    regressions = await _notificationTracker.GetRuleRegressionsAsync(tenantId),
                     summary = new
                     {
                         totalEvaluations,
@@ -165,6 +171,11 @@ namespace AutopilotMonitor.Functions.Functions.Metrics
                 var result = new
                 {
                     rules = aggregated,
+                    // Regression episodes are tenant-scoped: present when drilling into one
+                    // tenant, empty for the cross-tenant "global" aggregate (no episode there).
+                    regressions = queryTenantId != "global"
+                        ? await _notificationTracker.GetRuleRegressionsAsync(queryTenantId)
+                        : new List<AutopilotMonitor.Shared.Models.RuleRegressionAlert>(),
                     summary = new
                     {
                         totalEvaluations,

src/Backend/AutopilotMonitor.Functions/Functions/Reports/GetTpmPssUnsupportedFunction.cs

@@ -0,0 +1,95 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.DataAccess;
+using Microsoft.Azure.Functions.Worker;
+using Microsoft.Azure.Functions.Worker.Http;
+using Microsoft.Extensions.Logging;
+
+namespace AutopilotMonitor.Functions.Functions.Reports
+{
+    /// <summary>
+    /// Tenant-scoped endpoint: returns devices whose TPM-backed client certificate cannot
+    /// perform RSA-PSS signing (Schannel silently drops the cert from TLS client-auth on
+    /// Windows 11 25H2+, so the agent can never authenticate), aggregated by serial number.
+    /// Data comes from distress reports with ErrorType == "TpmPssUnsupported".
+    ///
+    /// Informational only — the remediation is a TPM firmware update or device replacement,
+    /// not a portal action. The 14-day horizon is enforced by the distress-report retention
+    /// cleanup, not by this function.
+    /// Authentication + MemberRead authorization enforced by PolicyEnforcementMiddleware.
+    /// </summary>
+    public class GetTpmPssUnsupportedFunction
+    {
+        private readonly ILogger<GetTpmPssUnsupportedFunction> _logger;
+        private readonly IDistressReportRepository _repository;
+
+        public GetTpmPssUnsupportedFunction(
+            ILogger<GetTpmPssUnsupportedFunction> logger,
+            IDistressReportRepository repository)
+        {
+            _logger = logger;
+            _repository = repository;
+        }
+
+        [Function("GetTpmPssUnsupported")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "audit/tpm-pss-unsupported")] HttpRequestData req)
+        {
+            try
+            {
+                string tenantId = TenantHelper.GetTenantId(req);
+
+                var reports = await _repository.GetDistressReportsAsync(tenantId, maxResults: 200);
+                var (aggregated, totalRawReports) = BuildAggregatedResult(reports);
+
+                return await req.OkAsync(new
+                {
+                    success = true,
+                    aggregated,
+                    totalRawReports,
+                    dataQualityNotice = "This data is from pre-authentication distress reports and is UNVERIFIED. Devices reported here have a TPM that cannot perform RSA-PSS signing, so their agent cannot authenticate to the backend. Serial number, manufacturer, and model values are self-reported by devices."
+                });
+            }
+            catch (Exception ex)
+            {
+                return await req.InternalServerErrorAsync(_logger, ex, "Get TPM PSS unsupported devices");
+            }
+        }
+
+        /// <summary>
+        /// Filters distress reports to TpmPssUnsupported and aggregates by serial number.
+        /// Reports with no serial number are grouped into a single "unknown serial" bucket.
+        /// Extracted as public static for testability.
+        /// </summary>
+        public static (List<object> aggregated, int totalRawReports) BuildAggregatedResult(
+            List<DistressReportEntry> reports)
+        {
+            var tpmReports = reports
+                .Where(r => string.Equals(r.ErrorType, "TpmPssUnsupported", StringComparison.OrdinalIgnoreCase))
+                .ToList();
+
+            var aggregated = tpmReports
+                .GroupBy(r => (r.SerialNumber ?? "").Trim().ToLowerInvariant())
+                .Select(g =>
+                {
+                    var mostRecent = g.OrderByDescending(r => r.IngestedAt).First();
+
+                    return new
+                    {
+                        serialNumber = mostRecent.SerialNumber ?? "",
+                        manufacturer = mostRecent.Manufacturer ?? "",
+                        model = mostRecent.Model ?? "",
+                        attemptCount = g.Count(),
+                        firstSeen = g.Min(r => r.IngestedAt),
+                        lastSeen = g.Max(r => r.IngestedAt)
+                    };
+                })
+                .OrderByDescending(a => a.lastSeen)
+                .ToList<object>();
+
+            return (aggregated, tpmReports.Count);
+        }
+    }
+}

src/Backend/AutopilotMonitor.Functions/Functions/Reports/SubmitDiagFilesReportFunction.cs

@@ -10,8 +10,8 @@ using Microsoft.Extensions.Logging;
 namespace AutopilotMonitor.Functions.Functions.Reports
 {
     /// <summary>
-    /// Tenant-admin entry point for submitting diagnostic files (logs, JSON state files,
-    /// screenshots) without a session context. Lands in the same SessionReports table +
+    /// Tenant-admin/-operator entry point for submitting diagnostic files (logs, JSON state
+    /// files, screenshots) without a session context. Lands in the same SessionReports table +
     /// session-reports blob container as session reports, distinguished by
     /// <see cref="SessionReportMetadata.ReportType"/> = "diagFiles".
     /// </summary>
@@ -45,7 +45,7 @@ namespace AutopilotMonitor.Functions.Functions.Reports
 
             try
             {
-                // Authentication + TenantAdminOrGA authorization enforced by PolicyEnforcementMiddleware
+                // Authentication + TenantAdminOrOperator authorization enforced by PolicyEnforcementMiddleware
                 var requestCtx = req.GetRequestContext();
                 var tenantId = requestCtx.TenantId;
                 var userIdentifier = requestCtx.UserPrincipalName;

src/Backend/AutopilotMonitor.Functions/Functions/Rules/GlobalRulesFunction.cs

@@ -1,14 +1,22 @@
 using System.Net;
 using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
 using Microsoft.Azure.Functions.Worker;
 using Microsoft.Azure.Functions.Worker.Http;
 using Microsoft.Extensions.Logging;
+using Newtonsoft.Json;
 
 namespace AutopilotMonitor.Functions.Functions.Rules
 {
     /// <summary>
-    /// Global Admin endpoints for viewing rules across tenants.
-    /// Reads tenantId from query string (not JWT) so global admins can inspect any tenant.
+    /// Global Admin endpoints for viewing AND mutating rules across tenants.
+    /// Reads tenantId from the ?tenantId= query string (not JWT) so a Global Admin can inspect
+    /// or edit any tenant. The write routes exist because the JWT-scoped GatherRulesFunction /
+    /// AnalyzeRulesFunction resolve the tenant from the caller's own token: a GA editing a foreign
+    /// tenant via those routes would silently upsert into its HOME tenant (200 → false "saved"),
+    /// while the reload re-reads the foreign tenant and shows the unchanged rule. These routes are
+    /// GlobalAdminOnly + TenantScoping.QueryParam so cross-tenant write stays platform-admin-only
+    /// (no tenant-admin / delegated rescue — the delegated grant applies to READ tiers only).
     /// </summary>
     public class GlobalRulesFunction
     {
@@ -77,5 +85,163 @@ namespace AutopilotMonitor.Functions.Functions.Rules
             await response.WriteAsJsonAsync(new { success = true, rules });
             return response;
         }
+
+        /// <summary>
+        /// PUT /api/global/rules/gather/{ruleId}?tenantId=X - Update a gather rule in any tenant (Global Admin only).
+        /// Cross-tenant counterpart of GatherRulesFunction.UpdateRule; the tenant comes from the query, not the JWT.
+        /// </summary>
+        [Function("UpdateGlobalGatherRule")]
+        public async Task<HttpResponseData> UpdateGatherRule(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "put", Route = "global/rules/gather/{ruleId}")] HttpRequestData req,
+            string ruleId)
+        {
+            // Authentication + GlobalAdminOnly authorization enforced by PolicyEnforcementMiddleware.
+            var tenantId = ResolveTenantId(req, out var tenantError);
+            if (tenantError != null)
+                return await BadRequestAsync(req, tenantError);
+
+            var (rule, bodyError) = await ReadRuleBodyAsync<GatherRule>(req);
+            if (bodyError != null)
+                return await BadRequestAsync(req, bodyError);
+
+            rule!.RuleId = ruleId;
+
+            var scopeError = GatherRulesFunction.ValidateScopeAndEmitMode(rule);
+            if (scopeError != null)
+                return await BadRequestAsync(req, scopeError);
+
+            _logger.LogInformation("Global admin updating gather rule {RuleId} for tenant {TenantId}", ruleId, tenantId);
+
+            var success = await _gatherRuleService.UpdateRuleAsync(tenantId, rule);
+            var response = req.CreateResponse(success ? HttpStatusCode.OK : HttpStatusCode.InternalServerError);
+            await response.WriteAsJsonAsync(new { success, message = success ? "Rule updated" : "Failed to update rule" });
+            return response;
+        }
+
+        /// <summary>
+        /// DELETE /api/global/rules/gather/{ruleId}?tenantId=X - Delete a gather rule in any tenant (Global Admin only).
+        /// </summary>
+        [Function("DeleteGlobalGatherRule")]
+        public async Task<HttpResponseData> DeleteGatherRule(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "delete", Route = "global/rules/gather/{ruleId}")] HttpRequestData req,
+            string ruleId)
+        {
+            // Authentication + GlobalAdminOnly authorization enforced by PolicyEnforcementMiddleware.
+            var tenantId = ResolveTenantId(req, out var tenantError);
+            if (tenantError != null)
+                return await BadRequestAsync(req, tenantError);
+
+            // Load the rule to determine its type (built-in/community vs. custom).
+            var rules = await _gatherRuleService.GetAllRulesForTenantAsync(tenantId);
+            var rule = rules.FirstOrDefault(r => r.RuleId == ruleId);
+            if (rule == null)
+            {
+                var notFound = req.CreateResponse(HttpStatusCode.NotFound);
+                await notFound.WriteAsJsonAsync(new { success = false, message = "Rule not found" });
+                return notFound;
+            }
+
+            _logger.LogInformation("Global admin deleting gather rule {RuleId} for tenant {TenantId}", ruleId, tenantId);
+
+            var success = await _gatherRuleService.DeleteRuleAsync(tenantId, rule);
+            var response = req.CreateResponse(success ? HttpStatusCode.OK : HttpStatusCode.InternalServerError);
+            await response.WriteAsJsonAsync(new { success, message = success ? "Rule deleted" : "Failed to delete rule" });
+            return response;
+        }
+
+        /// <summary>
+        /// PUT /api/global/rules/analyze/{ruleId}?tenantId=X - Update an analyze rule in any tenant (Global Admin only).
+        /// Cross-tenant counterpart of AnalyzeRulesFunction.UpdateRule.
+        /// </summary>
+        [Function("UpdateGlobalAnalyzeRule")]
+        public async Task<HttpResponseData> UpdateAnalyzeRule(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "put", Route = "global/rules/analyze/{ruleId}")] HttpRequestData req,
+            string ruleId)
+        {
+            // Authentication + GlobalAdminOnly authorization enforced by PolicyEnforcementMiddleware.
+            var tenantId = ResolveTenantId(req, out var tenantError);
+            if (tenantError != null)
+                return await BadRequestAsync(req, tenantError);
+
+            var (rule, bodyError) = await ReadRuleBodyAsync<AnalyzeRule>(req);
+            if (bodyError != null)
+                return await BadRequestAsync(req, bodyError);
+
+            rule!.RuleId = ruleId;
+
+            _logger.LogInformation("Global admin updating analyze rule {RuleId} for tenant {TenantId}", ruleId, tenantId);
+
+            var success = await _analyzeRuleService.UpdateRuleAsync(tenantId, rule);
+            var response = req.CreateResponse(success ? HttpStatusCode.OK : HttpStatusCode.InternalServerError);
+            await response.WriteAsJsonAsync(new { success, message = success ? "Rule updated" : "Failed to update rule" });
+            return response;
+        }
+
+        /// <summary>
+        /// DELETE /api/global/rules/analyze/{ruleId}?tenantId=X - Delete an analyze rule in any tenant (Global Admin only).
+        /// </summary>
+        [Function("DeleteGlobalAnalyzeRule")]
+        public async Task<HttpResponseData> DeleteAnalyzeRule(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "delete", Route = "global/rules/analyze/{ruleId}")] HttpRequestData req,
+            string ruleId)
+        {
+            // Authentication + GlobalAdminOnly authorization enforced by PolicyEnforcementMiddleware.
+            var tenantId = ResolveTenantId(req, out var tenantError);
+            if (tenantError != null)
+                return await BadRequestAsync(req, tenantError);
+
+            // Load the rule to determine its type (built-in/community vs. custom).
+            var rules = await _analyzeRuleService.GetAllRulesForTenantAsync(tenantId);
+            var rule = rules.FirstOrDefault(r => r.RuleId == ruleId);
+            if (rule == null)
+            {
+                var notFound = req.CreateResponse(HttpStatusCode.NotFound);
+                await notFound.WriteAsJsonAsync(new { success = false, message = "Rule not found" });
+                return notFound;
+            }
+
+            _logger.LogInformation("Global admin deleting analyze rule {RuleId} for tenant {TenantId}", ruleId, tenantId);
+
+            var success = await _analyzeRuleService.DeleteRuleAsync(tenantId, rule);
+            var response = req.CreateResponse(success ? HttpStatusCode.OK : HttpStatusCode.InternalServerError);
+            await response.WriteAsJsonAsync(new { success, message = success ? "Rule deleted" : "Failed to delete rule" });
+            return response;
+        }
+
+        /// <summary>
+        /// Resolves the target tenant from the ?tenantId= query string (the write routes are
+        /// TenantScoping.QueryParam). Sets <paramref name="error"/> when the parameter is missing.
+        /// </summary>
+        private static string ResolveTenantId(HttpRequestData req, out string? error)
+        {
+            var tenantId = System.Web.HttpUtility.ParseQueryString(req.Url.Query ?? "").Get("tenantId");
+            error = string.IsNullOrEmpty(tenantId) ? "tenantId query parameter is required" : null;
+            return tenantId ?? string.Empty;
+        }
+
+        /// <summary>
+        /// Reads and deserializes a rule body with the same 1 MB guard the JWT-scoped functions apply.
+        /// Returns (rule, null) on success or (null, errorMessage) for a 400.
+        /// </summary>
+        private static async Task<(T? rule, string? error)> ReadRuleBodyAsync<T>(HttpRequestData req) where T : class
+        {
+            if (req.Headers.TryGetValues("Content-Length", out var clValues)
+                && long.TryParse(clValues.FirstOrDefault(), out var contentLength)
+                && contentLength > 1_048_576) // 1 MB limit
+            {
+                return (null, "Request body too large");
+            }
+
+            var body = await new StreamReader(req.Body).ReadToEndAsync();
+            var rule = JsonConvert.DeserializeObject<T>(body);
+            return rule == null ? (null, "Invalid rule data") : (rule, null);
+        }
+
+        private static async Task<HttpResponseData> BadRequestAsync(HttpRequestData req, string message)
+        {
+            var badRequest = req.CreateResponse(HttpStatusCode.BadRequest);
+            await badRequest.WriteAsJsonAsync(new { success = false, message });
+            return badRequest;
+        }
     }
 }

src/Backend/AutopilotMonitor.Functions/Functions/Sessions/QueueSessionActionFunction.cs

@@ -32,6 +32,28 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
             ServerActionTypes.RequestDiagnostics
         };
 
+        /// <summary>
+        /// Action types an Operator may queue. The route is TenantAdminOrOperator so troubleshooting
+        /// staff can trigger an on-demand diagnostics collection; everything with config or lifecycle
+        /// impact (terminate_session, rotate_config) stays Admin/GA-only via <see cref="IsTypeAllowedForCaller"/>.
+        /// </summary>
+        private static readonly HashSet<string> OperatorAllowedTypes = new(StringComparer.OrdinalIgnoreCase)
+        {
+            ServerActionTypes.RequestDiagnostics
+        };
+
+        /// <summary>
+        /// Type-level authorization on top of the route policy: Admins and Global Admins may queue
+        /// every allowed type, Operators only the types in <see cref="OperatorAllowedTypes"/>.
+        /// Internal-static so tests can pin the matrix without an HTTP host.
+        /// </summary>
+        internal static bool IsTypeAllowedForCaller(string actionType, bool isTenantAdmin, bool isGlobalAdmin)
+        {
+            if (isTenantAdmin || isGlobalAdmin)
+                return true;
+            return OperatorAllowedTypes.Contains(actionType);
+        }
+
         public QueueSessionActionFunction(
             ILogger<QueueSessionActionFunction> logger,
             ISessionRepository sessionRepo,
@@ -53,9 +75,11 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
         {
             try
             {
-                // Authentication + TenantAdminOrGA authorization enforced by PolicyEnforcementMiddleware.
+                // Authentication + TenantAdminOrOperator authorization enforced by
+                // PolicyEnforcementMiddleware; per-type re-gating for Operators happens below.
                 var tenantId = TenantHelper.GetTenantId(req);
                 var userIdentifier = TenantHelper.GetUserIdentifier(req);
+                var requestCtx = req.GetRequestContext();
 
                 string body;
                 using (var reader = new StreamReader(req.Body))
@@ -78,6 +102,20 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
                 if (!AllowedTypes.Contains(action.Type))
                     return await BadRequestAsync(req, $"Unknown action type '{action.Type}'. Allowed: {string.Join(", ", AllowedTypes)}");
 
+                if (!IsTypeAllowedForCaller(action.Type, requestCtx.IsTenantAdmin, requestCtx.IsGlobalAdmin))
+                {
+                    _logger.LogWarning(
+                        "Operator {User} denied queueing admin-only action '{Type}' for session {SessionId}",
+                        userIdentifier, action.Type, sessionId);
+                    var forbidden = req.CreateResponse(HttpStatusCode.Forbidden);
+                    await forbidden.WriteAsJsonAsync(new
+                    {
+                        success = false,
+                        message = $"Action type '{action.Type}' requires the Tenant Admin role"
+                    });
+                    return forbidden;
+                }
+
                 // Stamp server-side fields — never trust client timestamps, and never let the caller
                 // forge a RuleId (that's reserved for the RuleEngine).
                 action.QueuedAt = DateTime.UtcNow;

src/Backend/AutopilotMonitor.Functions/Functions/Sessions/SearchSessionsFunction.cs

@@ -210,6 +210,8 @@ public class SearchSessionsFunction
             if (fields.Contains("imeAgentVersion")) dict["imeAgentVersion"] = s.ImeAgentVersion;
             if (fields.Contains("geoCountry")) dict["geoCountry"] = s.GeoCountry;
             if (fields.Contains("rebootCount")) dict["rebootCount"] = s.RebootCount;
+            if (fields.Contains("avgApiLatencyMs")) dict["avgApiLatencyMs"] = s.AvgApiLatencyMs;
+            if (fields.Contains("apiRequestCount")) dict["apiRequestCount"] = s.ApiRequestCount;
 
             if (dict.Count == 0)
             {

src/Backend/AutopilotMonitor.Functions/Helpers/DeviceJourneyCalculator.cs

@@ -0,0 +1,244 @@
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Helpers;
+
+/// <summary>
+/// One journey (a single deployment intent for one device): consecutive terminal attempts of a
+/// device key, ended by the first terminal success. Transient grouping result — never persisted.
+/// </summary>
+public sealed class DeviceJourney
+{
+    /// <summary>Terminal session refs of this journey, ordered by StartedAt ascending.</summary>
+    public List<DeviceSessionRef> Attempts { get; } = new();
+
+    /// <summary>True when the journey ended with a terminal success (its last attempt Succeeded).</summary>
+    public bool Completed { get; set; }
+
+    /// <summary>The success ref that completed the journey; null while open/abandoned.</summary>
+    public DeviceSessionRef? CompletingRef => Completed ? Attempts[Attempts.Count - 1] : null;
+}
+
+/// <summary>
+/// Deterministic, I/O-free core of the F2 device-history / First-Time-Right feature
+/// (insights spec §F2 — JourneyVersion 1). Owns the device-key normalization (trim +
+/// case-fold + junk-serial exclusion), the terminal-only chain maintenance (upsert by
+/// sessionId, cap 20 most recent) and the journey grouping:
+/// <list type="bullet">
+/// <item>Terminal set = Succeeded / Failed / Incomplete (production-verified status catalogue).
+/// A WhiteGlove <c>Pending</c> row is an OPEN session — WG part 1 + part 2 share ONE session
+/// row (second-sweep finding), so journeys never stitch WG parts and a WG device still waiting
+/// for its user session leaves the journey open, never failed.</item>
+/// <item>A journey ends with the first terminal SUCCESS; the next session starts a new journey
+/// (redeployment). A gap of &gt; 30 days since the previous terminal session also starts a new
+/// journey (device shelved/repurposed) — production gaps: median 4.9 h, 92 % &lt; 7 d.</item>
+/// <item>Attempt count = terminal sessions in the journey; Incomplete and Failed are
+/// non-successful attempts.</item>
+/// </list>
+/// </summary>
+public static class DeviceJourneyCalculator
+{
+    /// <summary>Bump on any grouping-semantics change so aggregates never mix definitions (truthfulness rule 8).</summary>
+    public const int CurrentVersion = 1;
+
+    /// <summary>Chain cap: the 20 most recent terminal sessions (spec §F2 — journey counts are chain-scoped).</summary>
+    internal const int MaxChainLength = 20;
+
+    /// <summary>New-journey boundary without a prior success. Constant in v1, not a setting (spec §F2).</summary>
+    internal const int JourneyGapDays = 30;
+
+    /// <summary>
+    /// Placeholder serials that are not device identities (spec §F2 + audit Q5: "Unknown" is the
+    /// agent's WMI-failure sentinel). Compared against the NORMALIZED (trimmed, lower-cased)
+    /// serial; anything shorter than 4 characters is junk regardless.
+    /// </summary>
+    private static readonly HashSet<string> JunkSerials = new(StringComparer.Ordinal)
+    {
+        "system serial number",
+        "to be filled by o.e.m.",
+        "default string",
+        "0",
+        "none",
+        "invalid",
+        "unknown",
+    };
+
+    /// <summary>
+    /// Normalizes a raw serial to the device-key form (trim + lower-case invariant) or returns
+    /// null for junk/placeholder serials — excluded devices never enter FTR and never get a
+    /// history chain (the exclusion is disclosed in the daily aggregate instead).
+    /// </summary>
+    public static string? NormalizeSerial(string? rawSerial)
+    {
+        if (string.IsNullOrWhiteSpace(rawSerial)) return null;
+        var normalized = rawSerial.Trim().ToLowerInvariant();
+        if (normalized.Length < 4) return null;
+        if (JunkSerials.Contains(normalized)) return null;
+        return normalized;
+    }
+
+    /// <summary>F2 terminal set — everything else (InProgress, Pending, Stalled, AwaitingUser, Unknown) is an open session.</summary>
+    public static bool IsTerminal(SessionStatus status)
+        => status == SessionStatus.Succeeded || status == SessionStatus.Failed || status == SessionStatus.Incomplete;
+
+    /// <summary>
+    /// Builds the chain ref for a terminal session; null for non-terminal statuses (defense in
+    /// depth — callers gate on <see cref="IsTerminal"/> already). Duration is the session's
+    /// authoritative <c>DurationSeconds</c> verbatim, never CompletedAt − StartedAt.
+    /// </summary>
+    public static DeviceSessionRef? BuildSessionRef(SessionSummary session)
+    {
+        if (!IsTerminal(session.Status)) return null;
+        return new DeviceSessionRef
+        {
+            SessionId = session.SessionId,
+            StartedAt = session.StartedAt,
+            CompletedAt = session.CompletedAt,
+            Status = session.Status.ToString(),
+            EnrollmentType = string.IsNullOrEmpty(session.EnrollmentType) ? "v1" : session.EnrollmentType,
+            IsPreProvisioned = session.IsPreProvisioned,
+            DurationSeconds = session.DurationSeconds,
+            AdminMarked = !string.IsNullOrEmpty(session.AdminMarkedAction),
+        };
+    }
+
+    /// <summary>
+    /// Merges refs into an existing chain: upsert by sessionId (a re-terminal reclassification,
+    /// e.g. Incomplete → Succeeded, replaces the entry), non-terminal refs dropped defensively,
+    /// ordered by StartedAt (tiebreak sessionId, ordinal) and capped to the
+    /// <see cref="MaxChainLength"/> MOST RECENT entries. Pure — inputs are not mutated.
+    /// </summary>
+    public static List<DeviceSessionRef> MergeChain(
+        IReadOnlyList<DeviceSessionRef>? existing, IEnumerable<DeviceSessionRef> updates)
+    {
+        var bySessionId = new Dictionary<string, DeviceSessionRef>(StringComparer.Ordinal);
+        if (existing != null)
+        {
+            foreach (var reference in existing)
+            {
+                if (!string.IsNullOrEmpty(reference.SessionId))
+                    bySessionId[reference.SessionId] = reference;
+            }
+        }
+        foreach (var update in updates)
+        {
+            if (update != null && !string.IsNullOrEmpty(update.SessionId))
+                bySessionId[update.SessionId] = update;
+        }
+
+        return SortAndCap(bySessionId.Values.Where(IsTerminalRef));
+    }
+
+    /// <summary>
+    /// Drops refs whose sessions were deleted (tombstoned or cascaded away) and re-normalizes
+    /// the chain. An empty result means the device has no observable history left — the caller
+    /// deletes the row rather than keeping an empty claim.
+    /// </summary>
+    public static List<DeviceSessionRef> RemoveSessionRefs(
+        IReadOnlyList<DeviceSessionRef> chain, ISet<string> deletedSessionIds)
+        => SortAndCap(chain.Where(r => IsTerminalRef(r) && !deletedSessionIds.Contains(r.SessionId)));
+
+    private static List<DeviceSessionRef> SortAndCap(IEnumerable<DeviceSessionRef> refs)
+    {
+        var sorted = refs
+            .OrderBy(r => r.StartedAt)
+            .ThenBy(r => r.SessionId, StringComparer.Ordinal)
+            .ToList();
+        if (sorted.Count > MaxChainLength)
+            sorted.RemoveRange(0, sorted.Count - MaxChainLength);
+        return sorted;
+    }
+
+    /// <summary>
+    /// A chain entry parses as terminal. Chains only ever receive terminal refs, but the JSON
+    /// column is data, not a proof — parse defensively (wire-type lesson: mirror the canonical
+    /// reader, never assume).
+    /// </summary>
+    private static bool IsTerminalRef(DeviceSessionRef reference)
+        => Enum.TryParse<SessionStatus>(reference.Status, ignoreCase: true, out var status) && IsTerminal(status);
+
+    /// <summary>
+    /// Groups an ordered terminal chain into journeys. The chain is re-sorted defensively;
+    /// non-terminal refs are ignored. See the class doc for the grouping rules.
+    /// </summary>
+    public static List<DeviceJourney> GroupJourneys(IReadOnlyList<DeviceSessionRef> chain)
+    {
+        var ordered = SortAndCap(chain.Where(IsTerminalRef));
+        var journeys = new List<DeviceJourney>();
+        DeviceJourney? current = null;
+        DeviceSessionRef? previous = null;
+
+        foreach (var reference in ordered)
+        {
+            var startNew =
+                current == null
+                || current.Completed
+                // Gap rule: > 30 days between the previous attempt's terminal moment and this
+                // start. CompletedAt is the honest "last seen terminal" timestamp; StartedAt is
+                // the defensive fallback (production: 0 terminal sessions without CompletedAt).
+                || (previous != null && reference.StartedAt - (previous.CompletedAt ?? previous.StartedAt) > TimeSpan.FromDays(JourneyGapDays));
+
+            if (startNew)
+            {
+                current = new DeviceJourney();
+                journeys.Add(current);
+            }
+
+            current!.Attempts.Add(reference);
+            if (string.Equals(reference.Status, nameof(SessionStatus.Succeeded), StringComparison.OrdinalIgnoreCase))
+                current.Completed = true;
+
+            previous = reference;
+        }
+
+        return journeys;
+    }
+
+    /// <summary>
+    /// Derived counts for the history row: journeys in the retained chain and the attempt count
+    /// of the LAST journey (open or completed) — the "Attempt N" the session banner renders.
+    /// </summary>
+    public static (int JourneyCount, int CurrentJourneyAttempts) Derive(IReadOnlyList<DeviceSessionRef> chain)
+    {
+        var journeys = GroupJourneys(chain);
+        if (journeys.Count == 0) return (0, 0);
+        return (journeys.Count, journeys[journeys.Count - 1].Attempts.Count);
+    }
+
+    /// <summary>
+    /// Attempt number of one session within its journey — the "Attempt N for this device" the
+    /// session-detail banner renders, computed server-side so no consumer re-derives journey
+    /// semantics. Terminal sessions sit in the chain and take their real position. A LIVE
+    /// (non-terminal) session is not a chain ref yet, so its position is computed by inserting
+    /// a virtual non-successful attempt at the session's StartedAt — the redeploy rule (a prior
+    /// completed journey starts a new one) and the 30-day gap rule then place it exactly like
+    /// the real ref would land once terminal. Null when the chain gives no basis (empty).
+    /// </summary>
+    public static int? ComputeAttemptNumber(
+        IReadOnlyList<DeviceSessionRef> chain, string sessionId, DateTime sessionStartedAt)
+    {
+        var effectiveChain = chain;
+        if (!chain.Any(r => string.Equals(r.SessionId, sessionId, StringComparison.Ordinal)))
+        {
+            if (chain.Count == 0) return null;
+            // Virtual ref: status is a NON-successful terminal placeholder purely for position
+            // math — it never persists and never marks the journey completed.
+            var virtualRef = new DeviceSessionRef
+            {
+                SessionId = sessionId,
+                StartedAt = sessionStartedAt,
+                Status = nameof(SessionStatus.Failed),
+            };
+            effectiveChain = MergeChain(chain, new[] { virtualRef });
+        }
+
+        foreach (var journey in GroupJourneys(effectiveChain))
+        {
+            for (var i = 0; i < journey.Attempts.Count; i++)
+            {
+                if (string.Equals(journey.Attempts[i].SessionId, sessionId, StringComparison.Ordinal))
+                    return i + 1;
+            }
+        }
+        return null; // session fell off the 20-cap when merged — no honest position claim
+    }
+}

src/Backend/AutopilotMonitor.Functions/Helpers/EspBlockingSets.cs

@@ -0,0 +1,115 @@
+using System.Collections;
+
+namespace AutopilotMonitor.Functions.Helpers;
+
+/// <summary>
+/// Parsed ESP blocking/tracking lists from an <c>esp_config_detected</c> event payload
+/// (agent emitter: <c>EspTrackingInfoProbe</c> → <c>DeviceInfoCollector.CollectEspConfiguration</c>).
+///
+/// Semantics (source-data audit Q2, insights spec §0.5): the lists are a POSITIVE-EVIDENCE set.
+/// An app id found in any list is proven ESP-blocking; an id that is absent is UNKNOWN — never
+/// "not blocking" — because the registry lists grow progressively (one timestamped subkey per
+/// CSP status write), the user-scope <c>S-&lt;SID&gt;</c> lists usually appear only after
+/// sign-in, each list is capped at 50 entries (uncapped totals ride in the <c>espTracked*Count</c>
+/// fields), and MSI ProductCodes / PFNs live in namespaces the IME <c>appId</c> may not match.
+/// Consumers must therefore always use the LATEST payload of a session that carries lists, and
+/// must map "absent" to null/unknown, never to false.
+/// </summary>
+public sealed class EspBlockingSets
+{
+    private readonly HashSet<string> _all;
+
+    private EspBlockingSets(
+        HashSet<string> all,
+        HashSet<string> userWin32AppIds,
+        int listedTotal,
+        bool isTruncated)
+    {
+        _all = all;
+        UserWin32AppIds = userWin32AppIds;
+        ListedCount = listedTotal;
+        IsTruncated = isTruncated;
+    }
+
+    /// <summary>Distinct ids across all four lists (Win32 device+user, MSI, PFN).</summary>
+    public int ListedCount { get; }
+
+    /// <summary>
+    /// True when any <c>espTracked*Count</c> exceeds its emitted list length — the 50-per-category
+    /// cap dropped entries, so even the positive evidence is incomplete for that category.
+    /// </summary>
+    public bool IsTruncated { get; }
+
+    /// <summary>
+    /// The user-scope Win32 subset (<c>espTrackedUserWin32AppIds</c>). Usually empty until the
+    /// agent-side re-emit fix ships (audit Q2) — exposed so callers can label user-ESP coverage
+    /// as unknown rather than inferring it from the merged list.
+    /// </summary>
+    public IReadOnlyCollection<string> UserWin32AppIds { get; }
+
+    /// <summary>Positive-evidence membership test (case-insensitive, all four lists).</summary>
+    public bool Contains(string? appId)
+        => !string.IsNullOrEmpty(appId) && _all.Contains(appId!);
+
+    /// <summary>
+    /// Parses the payload of one <c>esp_config_detected</c> event. Returns null when the payload
+    /// carries none of the <c>espTracked*</c> list keys (probe found no Diagnostics key — e.g.
+    /// non-Autopilot device or pre-list agent build): "no lists observed" must stay
+    /// distinguishable from "empty lists observed".
+    /// Handles both value shapes: live-ingest / storage-rehydrated data (<c>List&lt;object&gt;</c>
+    /// via EventDataNormalizer) and test fixtures (<c>string[]</c> / <c>List&lt;string&gt;</c>).
+    /// </summary>
+    public static EspBlockingSets? FromEventData(IReadOnlyDictionary<string, object>? data)
+    {
+        if (data == null) return null;
+
+        var win32 = ReadStringList(data, "espTrackedWin32AppIds");
+        var userWin32 = ReadStringList(data, "espTrackedUserWin32AppIds");
+        var msi = ReadStringList(data, "espTrackedMsiProductCodes");
+        var pfn = ReadStringList(data, "espTrackedModernAppPfns");
+
+        if (win32 == null && userWin32 == null && msi == null && pfn == null)
+            return null;
+
+        var all = new HashSet<string>(StringComparer.OrdinalIgnoreCase);
+        foreach (var list in new[] { win32, userWin32, msi, pfn })
+            if (list != null)
+                foreach (var id in list)
+                    all.Add(id);
+
+        var userSet = new HashSet<string>(userWin32 ?? new List<string>(), StringComparer.OrdinalIgnoreCase);
+
+        // The agent emits uncapped distinct totals alongside the capped lists. The Win32 count
+        // covers the merged device+user list; user-Win32 has no own count field.
+        var truncated =
+            CountExceeds(data, "espTrackedWin32Count", win32) ||
+            CountExceeds(data, "espTrackedMsiCount", msi) ||
+            CountExceeds(data, "espTrackedModernCount", pfn);
+
+        return new EspBlockingSets(all, userSet, all.Count, truncated);
+    }
+
+    private static bool CountExceeds(IReadOnlyDictionary<string, object> data, string countKey, List<string>? list)
+    {
+        if (list == null) return false;
+        if (!data.TryGetValue(countKey, out var raw)) return false;
+        return long.TryParse(raw?.ToString(), out var count) && count > list.Count;
+    }
+
+    private static List<string>? ReadStringList(IReadOnlyDictionary<string, object> data, string key)
+    {
+        if (!data.TryGetValue(key, out var raw) || raw == null) return null;
+        // Strings enumerate as chars — exclude explicitly before the IEnumerable fallback.
+        if (raw is string) return null;
+        if (raw is not IEnumerable enumerable) return null;
+
+        var result = new List<string>();
+        foreach (var item in enumerable)
+        {
+            var s = item?.ToString()?.Trim();
+            if (!string.IsNullOrEmpty(s))
+                result.Add(s!);
+        }
+        return result;
+    }
+}

src/Backend/AutopilotMonitor.Functions/Helpers/MetricsMath.cs

@@ -7,6 +7,55 @@ namespace AutopilotMonitor.Functions.Helpers;
 /// </summary>
 public static class MetricsMath
 {
+    /// <summary>
+    /// App-install failure rate over finished installs only: Failed / (Failed + Succeeded), one
+    /// decimal, 0 when nothing finished. The same outcome-quota convention as the enrollment
+    /// success rate — "InProgress" rows (still installing, or orphaned by a session that died
+    /// mid-install) never dilute the rate. Shared by the app-metrics payload and the Apps
+    /// dashboard aggregations so the definition can't drift between panes.
+    /// </summary>
+    public static double TerminalFailureRatePct(int failed, int succeeded)
+    {
+        var finished = failed + succeeded;
+        return finished > 0 ? Math.Round((double)failed / finished * 100, 1) : 0;
+    }
+
+    /// <summary>
+    /// True when the row's terminal state says no real install attempt happened —
+    /// Skipped (not applicable, e.g. WinGet "Update for X" with the target app absent)
+    /// or Postponed. PR0 decision (2026-07-26): such rows are excluded from duration
+    /// statistics AND from the failure/success rate; they are reported as their own
+    /// category instead. Rows written before the TerminalState column existed return
+    /// false here (empty state) — they keep the legacy rate behavior because they
+    /// cannot be reclassified honestly.
+    /// </summary>
+    public static bool IsSkipTerminalState(AppInstallSummary s)
+        => s.TerminalState == "Skipped" || s.TerminalState == "Postponed";
+
+    /// <summary>
+    /// Upper plausibility bound for an observed install duration: 6 h, the agent's default
+    /// max observation lifetime (AgentMaxLifetimeMinutes = 360) — nothing longer can have
+    /// been continuously watched. Rows above it are emission artifacts, not slow installs:
+    /// a batch emission at session end pins CompletedAt far past the real finish (verified
+    /// in production 2026-07-27: app rows carrying 60 175 s in a session whose own wall
+    /// clock was 2 697 s; 490 rows &gt; 6 h overall, worst 117 days — dominating the
+    /// slowest-apps ranking with six-figure averages).
+    /// </summary>
+    public const int MaxPlausibleInstallDurationSeconds = 6 * 3600;
+
+    /// <summary>
+    /// True for a succeeded row whose install duration was actually observed. Zero /
+    /// absent duration on a non-skip succeeded row means the START was never observed
+    /// (agent attach window — audit finding, §0.5 of the insights spec); a duration above
+    /// <see cref="MaxPlausibleInstallDurationSeconds"/> means the END was never observed
+    /// (completion back-stamped at session end). Either way the duration is UNKNOWN —
+    /// averaging zeros understates (measured −18 % in production), averaging back-stamps
+    /// inflates. Duration statistics must only read rows where this is true; everything
+    /// else is reported as "unmeasured".
+    /// </summary>
+    public static bool HasMeasuredDuration(AppInstallSummary s)
+        => s.DurationSeconds > 0 && s.DurationSeconds <= MaxPlausibleInstallDurationSeconds;
+
     /// <summary>
     /// Builds the complete app-metrics response object from a (pre-time-filtered) set of app
     /// install summaries. Single source of truth for both the tenant (<c>metrics/app</c>) and
@@ -23,9 +72,24 @@ public static class MetricsMath
     {
         var summaryList = summaries as IList<AppInstallSummary> ?? summaries.ToList();
 
-        var appGroups = summaryList.GroupBy(s => s.AppName).Select(g =>
+        // F1 PR1 (audit Q3): a name-keyed row that merged two distinct appIds carries an
+        // unattributable status/duration mix — it must not shape any per-app group. Excluded
+        // here with a disclosed count (truthfulness rule 7); the fleet-wide DO rollup below
+        // keeps every row, since transferred bytes are real regardless of identity mixing.
+        var totalCollisionExcluded = summaryList.Count(s => s.AppIdCollision);
+
+        var appGroups = summaryList.Where(s => !s.AppIdCollision).GroupBy(s => s.AppName).Select(g =>
         {
-            var completed = g.Where(s => s.Status == "Succeeded").ToList();
+            // PR0 (2026-07-26) classification — see IsSkipTerminalState / HasMeasuredDuration:
+            //   skipped    = no real install attempt (TerminalState Skipped/Postponed)
+            //   installed  = succeeded rows that are not known skips (includes legacy rows
+            //                without TerminalState — they cannot be reclassified)
+            //   measured   = installed rows with an observed duration — the ONLY duration input
+            //   unmeasured = installed rows whose start was never observed (duration unknown)
+            var succeededAll = g.Where(s => s.Status == "Succeeded").ToList();
+            var skipped = succeededAll.Where(IsSkipTerminalState).ToList();
+            var installed = succeededAll.Where(s => !IsSkipTerminalState(s)).ToList();
+            var measured = installed.Where(HasMeasuredDuration).ToList();
             var failed = g.Where(s => s.Status == "Failed").ToList();
             var total = g.Count();
 
@@ -38,12 +102,16 @@ public static class MetricsMath
             {
                 appName = g.Key,
                 totalInstalls = total,
-                succeeded = completed.Count,
+                succeeded = installed.Count,
+                skipped = skipped.Count,
+                unmeasured = installed.Count - measured.Count,
                 failed = failed.Count,
-                failureRate = total > 0 ? Math.Round((double)failed.Count / total * 100, 1) : 0,
-                avgDurationSeconds = completed.Count > 0 ? Math.Round(completed.Average(s => s.DurationSeconds), 0) : 0,
-                maxDurationSeconds = completed.Count > 0 ? completed.Max(s => s.DurationSeconds) : 0,
-                avgDownloadBytes = completed.Count > 0 ? (long)completed.Average(s => s.DownloadBytes) : 0,
+                // Skips leave the rate (they are not attempts); legacy rows stay in it.
+                failureRate = TerminalFailureRatePct(failed.Count, installed.Count),
+                avgDurationSeconds = measured.Count > 0 ? Math.Round(measured.Average(s => s.DurationSeconds), 0) : 0,
+                maxDurationSeconds = measured.Count > 0 ? measured.Max(s => s.DurationSeconds) : 0,
+                measuredInstalls = measured.Count,
+                avgDownloadBytes = measured.Count > 0 ? (long)measured.Average(s => s.DownloadBytes) : 0,
                 doTotalBytesDownloaded = doG.TotalBytesDownloaded,
                 doBytesFromPeers = doG.BytesFromPeers,
                 doBytesFromCacheServer = doG.BytesFromCacheServer,
@@ -58,8 +126,10 @@ public static class MetricsMath
             };
         }).ToList();
 
+        // Slowest ranking gates on MEASURED installs — an app whose durations were mostly
+        // unobserved must not rank as "fast" on a handful of zeros (audit §0.5).
         var slowestApps = SelectSlowestApps(
-            appGroups, a => a.succeeded, a => (double)a.avgDurationSeconds, minSamples: 3, take: 10);
+            appGroups, a => a.measuredInstalls, a => (double)a.avgDurationSeconds, minSamples: 3, take: 10);
 
         var topFailingApps = appGroups
             .Where(a => a.failed > 0)
@@ -75,6 +145,9 @@ public static class MetricsMath
             success = true,
             totalApps = appGroups.Count,
             totalInstalls = summaryList.Count,
+            totalSkipped = appGroups.Sum(a => a.skipped),
+            totalUnmeasured = appGroups.Sum(a => a.unmeasured),
+            totalCollisionExcluded,
             slowestApps,
             topFailingApps,
             deliveryOptimization = new
@@ -93,9 +166,10 @@ public static class MetricsMath
     /// Single source of truth for the tenant (<c>metrics/fleet-health</c>) and global
     /// (<c>global/metrics/fleet-health</c>) functions. Replaces the previous client-side path that
     /// drained up to 200k raw sessions into the browser and ran these aggregations on the main
-    /// thread. Semantics are preserved verbatim from that client code, notably: success rate is
-    /// over ALL sessions (not just terminal ones), and average duration counts every
-    /// non-in-progress session that carries a positive duration (failures included).
+    /// thread. Success rate follows the SLA convention: Succeeded / (Succeeded + Failed) — finished
+    /// enrollments only, so in-flight sessions and Incomplete (terminal, non-failure) never dilute
+    /// it. Average duration counts every non-in-progress session that carries a positive duration
+    /// (failures included).
     /// </summary>
     public static FleetHealthMetrics BuildFleetHealthPayload(IReadOnlyList<SessionSummary> sessions, int days)
     {
@@ -121,6 +195,7 @@ public static class MetricsMath
         }
 
         int total = sessions.Count;
+        int finished = succeeded + failed;
         var stats = new FleetHealthStats
         {
             Total = total,
@@ -128,7 +203,7 @@ public static class MetricsMath
             Failed = failed,
             InProgress = inProgress,
             Incomplete = incomplete,
-            SuccessRate = total > 0 ? Math.Round((double)succeeded / total * 100, 1) : 0,
+            SuccessRate = finished > 0 ? Math.Round((double)succeeded / finished * 100, 1) : 0,
             AvgDurationMinutes = completedWithDurationCount > 0
                 ? (int)Math.Round((double)completedDurationSeconds / completedWithDurationCount / 60.0, MidpointRounding.AwayFromZero)
                 : 0,
@@ -179,19 +254,29 @@ public static class MetricsMath
 
     private static List<FleetFailureReason> BuildFleetFailureReasons(IReadOnlyList<SessionSummary> sessions)
     {
-        var reasons = new Dictionary<string, int>();
+        // Group near-identical messages by a 50-char prefix so variants collapse
+        // into one row, but keep the longest full reason as the display value. The
+        // UI truncates and expands it on demand, so the complete text must survive.
+        var groups = new Dictionary<string, (string Display, int Count)>();
         foreach (var s in sessions)
         {
             if (s.Status != SessionStatus.Failed) continue;
             var reason = string.IsNullOrEmpty(s.FailureReason) ? "Unknown" : s.FailureReason;
-            // Collapse very long reasons to a 50-char prefix so near-identical messages group.
-            if (reason.Length > 50) reason = reason.Substring(0, 50) + "...";
-            reasons[reason] = reasons.GetValueOrDefault(reason) + 1;
+            var key = reason.Length > 50 ? reason.Substring(0, 50) : reason;
+            if (groups.TryGetValue(key, out var g))
+            {
+                var display = reason.Length > g.Display.Length ? reason : g.Display;
+                groups[key] = (display, g.Count + 1);
+            }
+            else
+            {
+                groups[key] = (reason, 1);
+            }
         }
-        return reasons
-            .OrderByDescending(kv => kv.Value)
+        return groups
+            .OrderByDescending(kv => kv.Value.Count)
             .Take(5)
-            .Select(kv => new FleetFailureReason { Reason = kv.Key, Count = kv.Value })
+            .Select(kv => new FleetFailureReason { Reason = kv.Value.Display, Count = kv.Value.Count })
             .ToList();
     }
 
@@ -208,6 +293,7 @@ public static class MetricsMath
             }
             m.Total++;
             if (s.Status == SessionStatus.Succeeded) m.Succeeded++;
+            else if (s.Status == SessionStatus.Failed) m.Failed++;
         }
         return models.Values
             .OrderByDescending(m => m.Total)
@@ -240,15 +326,18 @@ public static class MetricsMath
 
     private static List<FleetFailingModel> BuildFleetTopFailingModels(IReadOnlyList<SessionSummary> sessions)
     {
-        var acc = new Dictionary<string, (int Failed, int Total)>();
+        var acc = new Dictionary<string, (int Failed, int Succeeded, int Total)>();
         foreach (var s in sessions)
         {
             var key = FleetModelKey(s);
             acc.TryGetValue(key, out var cur);
             cur.Total++;
             if (s.Status == SessionStatus.Failed) cur.Failed++;
+            else if (s.Status == SessionStatus.Succeeded) cur.Succeeded++;
             acc[key] = cur;
         }
+        // FailureRate over finished enrollments only (mirror of the success-rate convention);
+        // the Where(Failed > 0) guard also keeps the denominator non-zero.
         return acc
             .Where(kv => kv.Value.Failed > 0)
             .Select(kv => new FleetFailingModel
@@ -256,7 +345,8 @@ public static class MetricsMath
                 Model = kv.Key,
                 Failed = kv.Value.Failed,
                 Total = kv.Value.Total,
-                FailureRate = (int)Math.Round((double)kv.Value.Failed / kv.Value.Total * 100, MidpointRounding.AwayFromZero),
+                FailureRate = (int)Math.Round(
+                    (double)kv.Value.Failed / (kv.Value.Failed + kv.Value.Succeeded) * 100, MidpointRounding.AwayFromZero),
             })
             .OrderByDescending(m => m.Failed)
             .Take(5)
@@ -308,6 +398,38 @@ public static class MetricsMath
             .Take(take)
             .ToList();
     }
+
+    /// <summary>
+    /// Wilson score interval (95 % default) for a binomial proportion — the F3 regression
+    /// radar's statistical primitive (insights spec §F3: deterministic, unit-pinned). Bounds
+    /// are clamped to [0,1]; n = 0 yields the uninformative (0,1) so a zero-denominator
+    /// side can never claim separation.
+    /// </summary>
+    public static (double Lower, double Upper) WilsonInterval(int successes, int trials, double z = 1.96)
+    {
+        if (trials <= 0) return (0.0, 1.0);
+        var n = (double)trials;
+        var p = Math.Clamp((double)successes / n, 0.0, 1.0);
+        var z2 = z * z;
+        var denominator = 1.0 + z2 / n;
+        var center = (p + z2 / (2.0 * n)) / denominator;
+        var half = (z / denominator) * Math.Sqrt(p * (1.0 - p) / n + z2 / (4.0 * n * n));
+        return (Math.Max(0.0, center - half), Math.Min(1.0, center + half));
+    }
+
+    /// <summary>
+    /// One-sided two-proportion separation for a rate INCREASE: true when the current
+    /// window's Wilson lower bound lies strictly above the baseline's Wilson upper bound —
+    /// the intervals are disjoint in the regression direction, so the lift is statistically
+    /// real rather than small-n noise (insights spec §F3 detection gate 3).
+    /// </summary>
+    public static bool RateIncreaseSeparated(
+        int windowHits, int windowTrials, int baselineHits, int baselineTrials, double z = 1.96)
+    {
+        var window = WilsonInterval(windowHits, windowTrials, z);
+        var baseline = WilsonInterval(baselineHits, baselineTrials, z);
+        return window.Lower > baseline.Upper;
+    }
 }
 
 /// <summary>

src/Backend/AutopilotMonitor.Functions/Helpers/RuleRegressionRadar.cs

@@ -0,0 +1,255 @@
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Helpers;
+
+/// <summary>One fired regression: a rule whose hit rate rose beyond every gate (see <see cref="RuleRegressionRadar"/>).</summary>
+public sealed class RuleRegressionFinding
+{
+    public string TenantId { get; init; } = string.Empty;
+    public string RuleId { get; init; } = string.Empty;
+    public string RuleTitle { get; init; } = string.Empty;
+    public string Severity { get; init; } = string.Empty;
+    public string Category { get; init; } = string.Empty;
+
+    public int WindowFireCount { get; init; }
+    public int WindowSessionCount { get; init; }
+    public int BaselineFireCount { get; init; }
+    public int BaselineSessionCount { get; init; }
+
+    public double WindowRatePct { get; init; }
+    public double BaselineRatePct { get; init; }
+
+    /// <summary>Window rate ÷ baseline rate, one decimal. Null when the baseline rate is 0 — a NEW signal has no finite lift, and we never invent one.</summary>
+    public double? Lift { get; init; }
+
+    public string WindowStartDate { get; init; } = string.Empty;
+    public string WindowEndDate { get; init; } = string.Empty;
+}
+
+/// <summary>
+/// Deterministic, I/O-free core of the F3 regression radar (insights spec §F3). Detects
+/// "this analyze rule suddenly fires much more often" from the daily <see cref="RuleStatsEntry"/>
+/// rows — RATES, not counts (absolute counts spike with every rollout wave): per (tenant, rule),
+/// hit rate = FireCount ÷ SessionsEvaluated, audit-verified to be session-deduplicated for
+/// analyze rules (gather rules are excluded until their per-batch dedup is fixed — Q6 follow-up,
+/// never silently included with bad math). A regression fires only when ALL hold:
+/// <list type="number">
+/// <item>trailing 7-day window has ≥ 5 hit sessions AND ≥ 20 evaluated sessions;</item>
+/// <item>window rate ≥ 2× the prior 28-day baseline rate;</item>
+/// <item>the Wilson 95 % intervals are disjoint in the increase direction
+/// (<see cref="MetricsMath.RateIncreaseSeparated"/>) — small-n noise never alerts.</item>
+/// </list>
+/// Suppressed (false positives are trust damage on a monitoring product): rules younger than
+/// baseline + window (grace period), rules edited inside the window (edits legitimately change
+/// hit rates), rules without an entity (deleted), and rules with an empty baseline denominator.
+/// </summary>
+public static class RuleRegressionRadar
+{
+    public const int WindowDays = 7;
+    public const int BaselineDays = 28;
+    public const int MinWindowHitSessions = 5;
+    public const int MinWindowSessions = 20;
+    public const double MinLift = 2.0;
+
+    /// <summary>An active alert re-arms once the window rate falls back under 1.5× baseline (spec §F3 suppression).</summary>
+    public const double ReArmLiftFactor = 1.5;
+
+    /// <summary>Dimension-concentration gates: a value must cover ≥5 hit sessions at ≥2× lift, or the alert says "no clear dimension concentration".</summary>
+    public const int MinDimensionSessions = 5;
+    public const double MinDimensionLift = 2.0;
+
+    /// <summary>Rule entity timestamps the suppression gates read (CreatedAt = grace period; UpdatedAt = edit-in-window).</summary>
+    public readonly record struct RuleTimestamps(DateTime CreatedAt, DateTime UpdatedAt);
+
+    /// <summary>
+    /// Evaluates one tenant's analyze-rule stats over the [target − 34d, target] horizon.
+    /// <paramref name="tenantEntries"/> must be the tenant's rows only (no "global" mirror);
+    /// gather rows are ignored defensively. <paramref name="ruleTimestamps"/> maps ruleId →
+    /// entity timestamps — a rule without an entry is suppressed (deleted rules never alert).
+    /// </summary>
+    public static List<RuleRegressionFinding> Evaluate(
+        IReadOnlyList<RuleStatsEntry> tenantEntries,
+        DateTime targetDateUtc,
+        IReadOnlyDictionary<string, RuleTimestamps> ruleTimestamps)
+    {
+        var findings = new List<RuleRegressionFinding>();
+        foreach (var group in AnalyzeRuleGroups(tenantEntries))
+        {
+            var sums = SumWindows(group.Value, targetDateUtc);
+
+            if (sums.WindowFire < MinWindowHitSessions) continue;
+            if (sums.WindowSessions < MinWindowSessions) continue;
+
+            // Entity gates: deleted rules never alert; new rules get the full baseline+window
+            // grace period; an edit inside the window legitimately changes the rate.
+            if (!ruleTimestamps.TryGetValue(group.Key, out var timestamps)) continue;
+            var windowStart = targetDateUtc.Date.AddDays(-(WindowDays - 1));
+            if (timestamps.CreatedAt > targetDateUtc.Date.AddDays(-(WindowDays - 1 + BaselineDays))) continue;
+            if (timestamps.UpdatedAt >= windowStart) continue;
+
+            // No observed baseline denominator (e.g. rule disabled throughout the baseline,
+            // re-enabled recently) — no honest comparison basis.
+            if (sums.BaselineSessions <= 0) continue;
+
+            var windowRate = (double)sums.WindowFire / sums.WindowSessions;
+            var baselineRate = (double)sums.BaselineFire / sums.BaselineSessions;
+            if (windowRate < MinLift * baselineRate) continue;
+            if (!MetricsMath.RateIncreaseSeparated(
+                    sums.WindowFire, sums.WindowSessions, sums.BaselineFire, sums.BaselineSessions))
+                continue;
+
+            var first = group.Value[0];
+            findings.Add(new RuleRegressionFinding
+            {
+                TenantId = first.TenantId,
+                RuleId = group.Key,
+                RuleTitle = first.RuleTitle,
+                Severity = first.Severity,
+                Category = first.Category,
+                WindowFireCount = sums.WindowFire,
+                WindowSessionCount = sums.WindowSessions,
+                BaselineFireCount = sums.BaselineFire,
+                BaselineSessionCount = sums.BaselineSessions,
+                WindowRatePct = Math.Round(windowRate * 100, 1),
+                BaselineRatePct = Math.Round(baselineRate * 100, 1),
+                Lift = baselineRate > 0 ? Math.Round(windowRate / baselineRate, 1) : null,
+                WindowStartDate = windowStart.ToString("yyyy-MM-dd"),
+                WindowEndDate = targetDateUtc.Date.ToString("yyyy-MM-dd"),
+            });
+        }
+        return findings
+            .OrderByDescending(f => f.WindowRatePct)
+            .ThenBy(f => f.RuleId, StringComparer.Ordinal)
+            .ToList();
+    }
+
+    /// <summary>
+    /// True when an ACTIVE alert for this rule may re-arm: nothing fired in the window anymore,
+    /// or the window rate fell back under <see cref="ReArmLiftFactor"/>× the baseline rate.
+    /// A zero-baseline alert ("new signal") re-arms only via the fires-stopped branch — there
+    /// is no rate threshold to fall under.
+    /// </summary>
+    public static bool ShouldReArm(IReadOnlyList<RuleStatsEntry> ruleEntries, DateTime targetDateUtc)
+    {
+        var sums = SumWindows(ruleEntries, targetDateUtc);
+        if (sums.WindowFire == 0) return true;
+        if (sums.BaselineFire <= 0 || sums.BaselineSessions <= 0 || sums.WindowSessions <= 0) return false;
+        var windowRate = (double)sums.WindowFire / sums.WindowSessions;
+        var baselineRate = (double)sums.BaselineFire / sums.BaselineSessions;
+        return windowRate < ReArmLiftFactor * baselineRate;
+    }
+
+    /// <summary>
+    /// The dominant dimension concentration among a regression's hit sessions, or null when no
+    /// dimension value reaches ≥<see cref="MinDimensionLift"/>× lift with
+    /// ≥<see cref="MinDimensionSessions"/> hit sessions — the alert then says
+    /// "no clear dimension concentration" instead of stretching for one (truthfulness rules 1/6).
+    /// Wording contract for consumers: "correlated — not necessarily causal".
+    /// </summary>
+    public static RuleRegressionDimension? ComputeDimensionConcentration(
+        IReadOnlyList<SessionSummary> hitSessions, IReadOnlyList<SessionSummary> allSessions)
+    {
+        if (hitSessions.Count == 0 || allSessions.Count == 0) return null;
+
+        var dimensions = new (string Name, Func<SessionSummary, string?> Selector)[]
+        {
+            ("osBuild", s => s.OsBuild),
+            ("model", s => $"{s.Manufacturer} {s.Model}".Trim()),
+            ("agentVersion", s => s.AgentVersion),
+            ("imeVersion", s => s.ImeAgentVersion),
+        };
+
+        RuleRegressionDimension? best = null;
+        foreach (var (name, selector) in dimensions)
+        {
+            var allCounts = CountValues(allSessions, selector);
+            foreach (var hit in CountValues(hitSessions, selector))
+            {
+                if (hit.Value < MinDimensionSessions) continue;
+                if (!allCounts.TryGetValue(hit.Key, out var allCount) || allCount <= 0) continue;
+
+                var hitShare = (double)hit.Value / hitSessions.Count;
+                var allShare = (double)allCount / allSessions.Count;
+                var lift = hitShare / allShare;
+                if (lift < MinDimensionLift) continue;
+
+                if (best == null || lift > best.Lift || (lift == best.Lift && hit.Value > best.HitCount))
+                {
+                    best = new RuleRegressionDimension
+                    {
+                        Dimension = name,
+                        Value = hit.Key,
+                        HitCount = hit.Value,
+                        HitSharePct = Math.Round(hitShare * 100, 1),
+                        AllSharePct = Math.Round(allShare * 100, 1),
+                        Lift = Math.Round(lift, 1),
+                    };
+                }
+            }
+        }
+        return best;
+    }
+
+    private static Dictionary<string, int> CountValues(
+        IReadOnlyList<SessionSummary> sessions, Func<SessionSummary, string?> selector)
+    {
+        var counts = new Dictionary<string, int>(StringComparer.OrdinalIgnoreCase);
+        foreach (var session in sessions)
+        {
+            var value = selector(session)?.Trim();
+            if (string.IsNullOrEmpty(value)) continue;
+            counts.TryGetValue(value!, out var current);
+            counts[value!] = current + 1;
+        }
+        return counts;
+    }
+
+    /// <summary>Groups a tenant's entries by ruleId, analyze rows only (defensive re-filter).</summary>
+    internal static Dictionary<string, List<RuleStatsEntry>> AnalyzeRuleGroups(IReadOnlyList<RuleStatsEntry> entries)
+    {
+        var groups = new Dictionary<string, List<RuleStatsEntry>>(StringComparer.OrdinalIgnoreCase);
+        foreach (var entry in entries)
+        {
+            if (!string.Equals(entry.RuleType, "analyze", StringComparison.OrdinalIgnoreCase)) continue;
+            if (string.IsNullOrEmpty(entry.RuleId)) continue;
+            if (!groups.TryGetValue(entry.RuleId, out var list))
+            {
+                list = new List<RuleStatsEntry>();
+                groups[entry.RuleId] = list;
+            }
+            list.Add(entry);
+        }
+        return groups;
+    }
+
+    internal readonly record struct WindowSums(int WindowFire, int WindowSessions, int BaselineFire, int BaselineSessions);
+
+    /// <summary>
+    /// Sums one rule's daily rows into the trailing window [target−6d, target] and the prior
+    /// baseline [target−34d, target−7d]. Dates are compared as ISO strings (the rows' own key
+    /// format); rows outside the horizon are ignored.
+    /// </summary>
+    internal static WindowSums SumWindows(IReadOnlyList<RuleStatsEntry> ruleEntries, DateTime targetDateUtc)
+    {
+        var windowStart = targetDateUtc.Date.AddDays(-(WindowDays - 1)).ToString("yyyy-MM-dd");
+        var baselineStart = targetDateUtc.Date.AddDays(-(WindowDays - 1 + BaselineDays)).ToString("yyyy-MM-dd");
+        var target = targetDateUtc.Date.ToString("yyyy-MM-dd");
+
+        int windowFire = 0, windowSessions = 0, baselineFire = 0, baselineSessions = 0;
+        foreach (var entry in ruleEntries)
+        {
+            if (string.CompareOrdinal(entry.Date, target) > 0) continue;
+            if (string.CompareOrdinal(entry.Date, windowStart) >= 0)
+            {
+                windowFire += entry.FireCount;
+                windowSessions += entry.SessionsEvaluated;
+            }
+            else if (string.CompareOrdinal(entry.Date, baselineStart) >= 0)
+            {
+                baselineFire += entry.FireCount;
+                baselineSessions += entry.SessionsEvaluated;
+            }
+        }
+        return new WindowSums(windowFire, windowSessions, baselineFire, baselineSessions);
+    }
+}

src/Backend/AutopilotMonitor.Functions/Helpers/TimeAttributionCalculator.cs

@@ -0,0 +1,666 @@
+using System.Globalization;
+using AutopilotMonitor.Shared;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Helpers;
+
+/// <summary>
+/// Inputs for <see cref="TimeAttributionCalculator.Compute"/> — the session facts the terminal
+/// writer persisted plus the session's full event stream in canonical order.
+/// </summary>
+public sealed class TimeAttributionInput
+{
+    public string TenantId { get; set; } = string.Empty;
+    public string SessionId { get; set; } = string.Empty;
+
+    /// <summary>Session status string. Only "Succeeded" and "Failed" are computable (Incomplete deliberately carries no DurationSeconds).</summary>
+    public string Status { get; set; } = string.Empty;
+
+    public DateTime StartedAt { get; set; }
+    public DateTime? CompletedAt { get; set; }
+
+    /// <summary>
+    /// The session's authoritative duration (second-sweep finding, insights spec §0.5): for
+    /// WhiteGlove it is part 1 + part 2 WITHOUT the pause, and re-terminal stamps move
+    /// CompletedAt — so this is the wall clock, never CompletedAt − StartedAt.
+    /// </summary>
+    public int? DurationSeconds { get; set; }
+
+    public bool IsPreProvisioned { get; set; }
+
+    /// <summary>WhiteGlove part-2 resume timestamp (session row), when present.</summary>
+    public DateTime? ResumedAt { get; set; }
+
+    /// <summary>Session events. Canonical order is Sequence ascending; the calculator re-sorts defensively.</summary>
+    public IReadOnlyList<EnrollmentEvent> Events { get; set; } = Array.Empty<EnrollmentEvent>();
+}
+
+/// <summary>
+/// Deterministic, I/O-free computation of a session's <see cref="SessionTimeBreakdown"/>
+/// (F1, insights spec — AttributionVersion 1). Every number is traceable to observed events;
+/// missing signals produce "unknown"/unattributed, never guesses (truthfulness rules 1/2).
+///
+/// Semantics pinned by the golden-fixture tests:
+/// <list type="bullet">
+/// <item><b>Wall clock = session DurationSeconds.</b> Observation windows are constructed to sum
+/// to it exactly: non-WhiteGlove = [CompletedAt − Duration, CompletedAt]; WhiteGlove = two windows
+/// in ONE session — [part1End − part1, part1End] and [ResumedAt, CompletedAt], where
+/// part1 = Duration − part2 and part1End is anchored on <c>whiteglove_part1_complete</c>.
+/// The pause between the windows is never attributed.</item>
+/// <item><b>Segments from phase-declaration events only</b> (phase strategy: only declaration
+/// events carry Phase != Unknown), bucketed Start/DevicePreparation/DeviceSetup → device_prep,
+/// AppsDevice → esp_apps, AccountSetup + FinalizingSetup → identity_hello, AppsUser → user_esp,
+/// Complete / <c>desktop_arrived</c> → desktop_handoff. A Failed-phase declaration ends
+/// attribution (the tail is unattributed, not relabeled). Anchors that go backward in canonical
+/// order are dropped (<see cref="TimeAttributionFlags.ClockSkewDropped"/>).</item>
+/// <item><b>Exact-partition invariant:</b> span seconds are floored, the remainder —
+/// unattributed holes plus sub-second dust — is reported as UnattributedSeconds, so
+/// sum(spans) + unattributed == wall clock holds exactly and is never normalized.</item>
+/// <item><b>Per-app intervals use event timestamps</b> (first started/download event → LAST
+/// terminal event, covering IME retries) — never the agent payload timing fields, which freeze
+/// at the first terminal transition (audit Q1).</item>
+/// <item><b>Blocking membership is positive evidence</b> from the LATEST
+/// <c>esp_config_detected</c> lists: listed ⇒ blocking; absent ⇒ unknown, never false (audit Q2).
+/// No lists ⇒ <see cref="TimeAttributionFlags.BlockingSetUnknown"/> and occupancy null.</item>
+/// <item><b>Reboot spans from event gaps</b> around the <c>lastBootUtc</c> payload — the
+/// <c>system_reboot_detected</c> event is detection-time-stamped by the next run and never used
+/// as the reboot moment (audit Q7). Reboots are a cross-cutting annotation, not a segment.</item>
+/// </list>
+/// </summary>
+public static class TimeAttributionCalculator
+{
+    /// <summary>Bump on any semantic change so aggregates never mix definitions (truthfulness rule 8).</summary>
+    public const int CurrentVersion = 1;
+
+    /// <summary>
+    /// Enrollment class for fleet aggregation — classes are NEVER mixed in one aggregate (a
+    /// WhiteGlove or WDP flow has a structurally different time profile than user-driven ESP).
+    /// Precedence: WhiteGlove (both parts live in one session row) → Windows Device Preparation
+    /// (EnrollmentType "v2" — different phase machinery) → self-deploying profile → user-driven.
+    /// </summary>
+    public static string GetEnrollmentClass(SessionSummary session)
+    {
+        if (session.IsPreProvisioned) return "whiteglove";
+        if (session.EnrollmentType == "v2") return "device_preparation";
+        if (session.IsSelfDeployingProfile) return "self_deploying";
+        return "user_driven";
+    }
+
+    /// <summary>Cap for the persisted per-app interval list (spec: top 20 by duration + uncapped count).</summary>
+    internal const int MaxBlockingAppIntervals = 20;
+
+    private readonly struct Window
+    {
+        public Window(DateTime start, DateTime end) { Start = start; End = end; }
+        public DateTime Start { get; }
+        public DateTime End { get; }
+    }
+
+    private sealed class Anchor
+    {
+        public DateTime Ts;
+        /// <summary>Segment key, or null = attribution ends here (Failed-phase declaration).</summary>
+        public string? Bucket;
+    }
+
+    private sealed class AppActivity
+    {
+        public string AppId = string.Empty;
+        public string AppName = string.Empty;
+        public DateTime? FirstStart;
+        public DateTime? LastTerminal;
+    }
+
+    /// <summary>
+    /// Returns the breakdown, or null when the session has no computable wall clock
+    /// (non-terminal, Incomplete — which deliberately stores no DurationSeconds — or a
+    /// missing/non-positive duration).
+    /// </summary>
+    public static SessionTimeBreakdown? Compute(TimeAttributionInput input)
+    {
+        if (input.Status != "Succeeded" && input.Status != "Failed") return null;
+        if (!input.CompletedAt.HasValue) return null;
+        if (!input.DurationSeconds.HasValue || input.DurationSeconds.Value <= 0) return null;
+
+        var completedAt = input.CompletedAt.Value;
+        var wallClock = input.DurationSeconds.Value;
+        var flags = TimeAttributionFlags.None;
+
+        // Canonical order (Sequence asc) — the authoritative event order per repo convention.
+        var events = input.Events.OrderBy(e => e.Sequence).ToList();
+
+        var windows = BuildWindows(input, completedAt, wallClock, events, ref flags);
+
+        var anchors = BuildAnchors(events, ref flags);
+        var spans = BuildSegmentSpans(windows, anchors);
+
+        var attributedSeconds = spans.Sum(s => s.Seconds);
+        var unattributed = wallClock - attributedSeconds;
+
+        // Blocking set: latest esp_config_detected emission that carries lists (positive
+        // evidence is never erased by a later probe failure; lists grow progressively).
+        EspBlockingSets? blockingSets = null;
+        for (var i = events.Count - 1; i >= 0 && blockingSets == null; i--)
+        {
+            if (events[i].EventType == Constants.EventTypes.EspConfigDetected)
+                blockingSets = EspBlockingSets.FromEventData(events[i].Data);
+        }
+        if (blockingSets == null || blockingSets.ListedCount == 0)
+            flags |= TimeAttributionFlags.BlockingSetUnknown;
+        else if (blockingSets.IsTruncated)
+            flags |= TimeAttributionFlags.BlockingSetTruncated;
+
+        var (blockingApps, blockingAppCount) = BuildBlockingAppIntervals(events, blockingSets, windows, ref flags);
+
+        int? occupancy = null;
+        if (blockingSets != null && blockingSets.ListedCount > 0)
+        {
+            var espAppsSpans = spans.Where(s => s.SegmentKey == TimeAttributionSegments.EspApps).ToList();
+            occupancy = ComputeOccupancySeconds(blockingApps, espAppsSpans);
+        }
+
+        var rebootSpans = BuildRebootSpans(events, windows, spans);
+
+        if (events.Any(e => e.EventType == Constants.EventTypes.AgentLateStart))
+            flags |= TimeAttributionFlags.PartialObservation;
+
+        return new SessionTimeBreakdown
+        {
+            TenantId = input.TenantId,
+            SessionId = input.SessionId,
+            AttributionVersion = CurrentVersion,
+            WallClockSeconds = wallClock,
+            Segments = spans,
+            UnattributedSeconds = unattributed,
+            RebootSeconds = rebootSpans.Sum(r => r.Seconds),
+            RebootSpans = rebootSpans,
+            BlockingApps = blockingApps
+                .OrderByDescending(a => a.Seconds)
+                .ThenBy(a => a.AppName, StringComparer.OrdinalIgnoreCase)
+                .Take(MaxBlockingAppIntervals)
+                .ToList(),
+            BlockingAppCount = blockingAppCount,
+            EspAppsOccupancySeconds = occupancy,
+            QualityFlags = flags,
+        };
+    }
+
+    // ── observation windows ─────────────────────────────────────────────────
+
+    /// <summary>
+    /// Windows are END-anchored on the authoritative timestamps (CompletedAt; part1End for WG
+    /// part 1) with the start derived so window lengths sum EXACTLY to DurationSeconds — the
+    /// invariant is against the stored duration, not against CompletedAt − StartedAt.
+    /// </summary>
+    private static List<Window> BuildWindows(
+        TimeAttributionInput input, DateTime completedAt, int wallClock,
+        List<EnrollmentEvent> events, ref TimeAttributionFlags flags)
+    {
+        if (input.IsPreProvisioned)
+        {
+            if (input.ResumedAt.HasValue && input.ResumedAt.Value < completedAt)
+            {
+                var resumedAt = input.ResumedAt.Value;
+                // Same truncation the terminal writer used for part2, so part1 comes out as the
+                // exact stored part-1 duration.
+                var part2 = (int)(completedAt - resumedAt).TotalSeconds;
+                var part1 = wallClock - part2;
+                if (part1 > 0)
+                {
+                    var part1End = ResolvePart1End(events, resumedAt, ref flags);
+                    if (part1End.HasValue)
+                    {
+                        var w1Start = part1End.Value.AddSeconds(-part1);
+                        return new List<Window>
+                        {
+                            new Window(w1Start, part1End.Value),
+                            new Window(resumedAt, completedAt),
+                        };
+                    }
+                }
+                // part1 <= 0 (stored duration contradicts the resume stamp) or no usable part-1
+                // anchor: fall through to a single derived window rather than inventing a split.
+            }
+            // WhiteGlove without a resolvable two-window split — the partition is still exact,
+            // but the WG semantics (pause exclusion boundaries) rest on no anchor.
+            flags |= TimeAttributionFlags.WhiteGloveAnchorsIncomplete;
+        }
+
+        return new List<Window> { new Window(completedAt.AddSeconds(-wallClock), completedAt) };
+    }
+
+    /// <summary>
+    /// Part-1 end anchor: the <c>whiteglove_part1_complete</c> event; fallback: the last event
+    /// at/before ResumedAt (flagged — the boundary is then observational, not declarative).
+    /// An anchor after ResumedAt is clamped to ResumedAt (windows must not overlap).
+    /// </summary>
+    private static DateTime? ResolvePart1End(
+        List<EnrollmentEvent> events, DateTime resumedAt, ref TimeAttributionFlags flags)
+    {
+        DateTime? part1End = null;
+        foreach (var evt in events)
+        {
+            if (evt.EventType == Constants.EventTypes.WhiteGlovePart1Complete)
+                part1End = evt.Timestamp;
+        }
+
+        if (!part1End.HasValue)
+        {
+            flags |= TimeAttributionFlags.WhiteGloveAnchorsIncomplete;
+            foreach (var evt in events)
+            {
+                if (evt.Timestamp <= resumedAt &&
+                    (!part1End.HasValue || evt.Timestamp > part1End.Value))
+                {
+                    part1End = evt.Timestamp;
+                }
+            }
+        }
+        else if (part1End.Value > resumedAt)
+        {
+            flags |= TimeAttributionFlags.WhiteGloveAnchorsIncomplete;
+            part1End = resumedAt;
+        }
+
+        return part1End;
+    }
+
+    // ── segment spans ───────────────────────────────────────────────────────
+
+    private static string? MapPhaseToSegment(EnrollmentPhase phase) => phase switch
+    {
+        EnrollmentPhase.Start => TimeAttributionSegments.DevicePrep,
+        EnrollmentPhase.DevicePreparation => TimeAttributionSegments.DevicePrep,
+        EnrollmentPhase.DeviceSetup => TimeAttributionSegments.DevicePrep,
+        EnrollmentPhase.AppsDevice => TimeAttributionSegments.EspApps,
+        EnrollmentPhase.AccountSetup => TimeAttributionSegments.IdentityHello,
+        EnrollmentPhase.AppsUser => TimeAttributionSegments.UserEsp,
+        EnrollmentPhase.FinalizingSetup => TimeAttributionSegments.IdentityHello,
+        EnrollmentPhase.Complete => TimeAttributionSegments.DesktopHandoff,
+        // Failed: attribution ends — the post-failure tail stays unattributed rather than
+        // being relabeled into a segment no phase ever declared.
+        EnrollmentPhase.Failed => null,
+        _ => null,
+    };
+
+    /// <summary>
+    /// Anchor list in canonical order: phase-declaration events (Phase != Unknown) plus
+    /// <c>desktop_arrived</c> (Phase=Unknown by convention, but the spec-designated handoff
+    /// anchor). Consecutive same-bucket anchors collapse; anchors whose timestamp goes
+    /// backward vs. the previously accepted anchor are dropped with ClockSkewDropped.
+    /// </summary>
+    private static List<Anchor> BuildAnchors(List<EnrollmentEvent> events, ref TimeAttributionFlags flags)
+    {
+        var anchors = new List<Anchor>();
+        foreach (var evt in events)
+        {
+            string? bucket;
+            var isDeclaration = false;
+            if (evt.EventType == Constants.EventTypes.DesktopArrived)
+            {
+                bucket = TimeAttributionSegments.DesktopHandoff;
+                isDeclaration = true;
+            }
+            else if (evt.Phase != EnrollmentPhase.Unknown)
+            {
+                bucket = MapPhaseToSegment(evt.Phase);
+                isDeclaration = true;
+            }
+            else
+            {
+                bucket = null;
+            }
+
+            if (!isDeclaration) continue;
+
+            if (anchors.Count > 0)
+            {
+                var prev = anchors[anchors.Count - 1];
+                if (evt.Timestamp < prev.Ts)
+                {
+                    flags |= TimeAttributionFlags.ClockSkewDropped;
+                    continue;
+                }
+                if (prev.Bucket == bucket)
+                    continue; // same-phase re-declaration (re-registration, replay) — no boundary
+            }
+
+            anchors.Add(new Anchor { Ts = evt.Timestamp, Bucket = bucket });
+        }
+        return anchors;
+    }
+
+    /// <summary>
+    /// Slices each window at the in-window anchors and buckets every slice by the phase context
+    /// active at its start. Context carries INTO window 0 from anchors at/before its start, but
+    /// never across the WhiteGlove pause into window 1+ (the resume gap makes no phase claim).
+    /// The leading gap of window 0 with no context is device_prep only when the first observed
+    /// anchor is still pre/at apps entry (the spec defines device_prep as session start → first
+    /// apps-phase entry); otherwise it stays unattributed.
+    /// </summary>
+    private static List<TimeAttributionSpan> BuildSegmentSpans(List<Window> windows, List<Anchor> anchors)
+    {
+        var spans = new List<TimeAttributionSpan>();
+
+        for (var w = 0; w < windows.Count; w++)
+        {
+            var window = windows[w];
+
+            string? context = null;
+            if (w == 0)
+            {
+                foreach (var a in anchors)
+                {
+                    if (a.Ts <= window.Start) context = a.Bucket;
+                    else break;
+                }
+            }
+
+            var inWindow = anchors.Where(a => a.Ts > window.Start && a.Ts < window.End).ToList();
+
+            if (w == 0 && context == null)
+            {
+                var first = inWindow.FirstOrDefault();
+                if (first != null &&
+                    (first.Bucket == TimeAttributionSegments.DevicePrep ||
+                     first.Bucket == TimeAttributionSegments.EspApps))
+                {
+                    context = TimeAttributionSegments.DevicePrep;
+                }
+            }
+
+            var cursor = window.Start;
+            foreach (var anchor in inWindow)
+            {
+                AddSpan(spans, context, cursor, anchor.Ts);
+                cursor = anchor.Ts;
+                context = anchor.Bucket;
+            }
+            AddSpan(spans, context, cursor, window.End);
+        }
+
+        return spans;
+    }
+
+    private static void AddSpan(List<TimeAttributionSpan> spans, string? bucket, DateTime start, DateTime end)
+    {
+        if (bucket == null || end <= start) return;
+        var seconds = (int)Math.Floor((end - start).TotalSeconds);
+        if (seconds <= 0) return; // sub-second slivers land in the unattributed remainder
+        spans.Add(new TimeAttributionSpan { SegmentKey = bucket, StartUtc = start, EndUtc = end, Seconds = seconds });
+    }
+
+    // ── blocking-app intervals + occupancy ──────────────────────────────────
+
+    private static (List<BlockingAppInterval> Intervals, int MatchedCount) BuildBlockingAppIntervals(
+        List<EnrollmentEvent> events, EspBlockingSets? sets, List<Window> windows, ref TimeAttributionFlags flags)
+    {
+        var intervals = new List<BlockingAppInterval>();
+        if (sets == null || sets.ListedCount == 0)
+            return (intervals, 0);
+
+        // Fold per-app activity from EVENT timestamps (never payload timing — audit Q1).
+        var apps = new Dictionary<string, AppActivity>(StringComparer.OrdinalIgnoreCase);
+        foreach (var evt in events)
+        {
+            var isStart = evt.EventType == Constants.EventTypes.AppInstallStart ||
+                          evt.EventType == Constants.EventTypes.AppDownloadStarted;
+            var isTerminal = evt.EventType == Constants.EventTypes.AppInstallComplete ||
+                             evt.EventType == Constants.EventTypes.AppInstallFailed ||
+                             evt.EventType == Constants.EventTypes.AppInstallSkipped;
+            if (!isStart && !isTerminal) continue;
+            if (evt.Data == null || !evt.Data.TryGetValue("appId", out var appIdObj)) continue;
+
+            var appId = appIdObj?.ToString()?.Trim();
+            if (string.IsNullOrEmpty(appId)) continue;
+
+            if (!apps.TryGetValue(appId!, out var activity))
+            {
+                activity = new AppActivity { AppId = appId! };
+                apps[appId!] = activity;
+            }
+            if (string.IsNullOrEmpty(activity.AppName) &&
+                evt.Data.TryGetValue("appName", out var nameObj))
+            {
+                activity.AppName = nameObj?.ToString()?.Trim() ?? string.Empty;
+            }
+
+            if (isStart && (!activity.FirstStart.HasValue || evt.Timestamp < activity.FirstStart.Value))
+                activity.FirstStart = evt.Timestamp;
+            // LAST terminal wins — an IME retry (Error → retry → Installed) extends the
+            // occupancy to the final outcome instead of freezing at the first failure.
+            if (isTerminal && (!activity.LastTerminal.HasValue || evt.Timestamp > activity.LastTerminal.Value))
+                activity.LastTerminal = evt.Timestamp;
+        }
+
+        var matched = 0;
+        foreach (var activity in apps.Values)
+        {
+            if (!sets.Contains(activity.AppId)) continue; // absent ⇒ unknown, never "not blocking"
+            matched++;
+
+            if (!activity.FirstStart.HasValue || !activity.LastTerminal.HasValue)
+                continue; // unobserved endpoint → no interval claim (attach window, audit §0.5)
+            if (activity.LastTerminal.Value < activity.FirstStart.Value)
+            {
+                flags |= TimeAttributionFlags.ClockSkewDropped;
+                continue;
+            }
+
+            var clamped = ClampToWindows(activity.FirstStart.Value, activity.LastTerminal.Value, windows);
+            if (!clamped.HasValue)
+            {
+                // Outside the hull entirely → pathological timestamps (flagged, as before).
+                // Inside the hull but touching no window (wholly inside the WhiteGlove pause):
+                // no in-window observation → no interval claim, and nothing wrong with clocks.
+                if (activity.LastTerminal.Value < windows[0].Start ||
+                    activity.FirstStart.Value > windows[windows.Count - 1].End)
+                {
+                    flags |= TimeAttributionFlags.ClockSkewDropped;
+                }
+                continue;
+            }
+
+            intervals.Add(new BlockingAppInterval
+            {
+                AppId = activity.AppId,
+                AppName = activity.AppName,
+                StartUtc = clamped.Value.Start,
+                EndUtc = clamped.Value.End,
+                // In-window seconds — a WG-pause-straddling interval never counts the pause.
+                Seconds = clamped.Value.Seconds,
+            });
+        }
+
+        return (intervals, matched);
+    }
+
+    /// <summary>
+    /// Clips [start, end] to the observation windows. Start/End are the clamped chronological
+    /// endpoints; <c>Seconds</c> counts ONLY time inside the windows — the WhiteGlove pause
+    /// between part 1 and part 2 lies inside the hull but is excluded, matching the session's
+    /// own <c>DurationSeconds</c> semantics (pause excluded by design). Null when the interval
+    /// overlaps no window at all: either it lies outside the hull entirely (pathological
+    /// timestamps) or wholly inside the pause (no in-window observation).
+    /// </summary>
+    private static (DateTime Start, DateTime End, int Seconds)? ClampToWindows(DateTime start, DateTime end, List<Window> windows)
+    {
+        DateTime? s = null, e = null;
+        double total = 0;
+        foreach (var window in windows)
+        {
+            var ws = start < window.Start ? window.Start : start;
+            var we = end > window.End ? window.End : end;
+            if (we < ws) continue;
+            if (s == null || ws < s.Value) s = ws;
+            if (e == null || we > e.Value) e = we;
+            total += (we - ws).TotalSeconds;
+        }
+        if (s == null || e == null) return null;
+        return (s.Value, e.Value, (int)Math.Floor(total));
+    }
+
+    /// <summary>
+    /// Critical-path occupancy: overlap-merged union of the blocking intervals, intersected
+    /// with the esp_apps spans. Whole-second floor on the total (consistent with span math).
+    /// </summary>
+    internal static int ComputeOccupancySeconds(
+        List<BlockingAppInterval> intervals, List<TimeAttributionSpan> espAppsSpans)
+    {
+        if (intervals.Count == 0 || espAppsSpans.Count == 0) return 0;
+
+        var merged = new List<(DateTime Start, DateTime End)>();
+        foreach (var iv in intervals.OrderBy(i => i.StartUtc))
+        {
+            if (merged.Count > 0 && iv.StartUtc <= merged[merged.Count - 1].End)
+            {
+                var last = merged[merged.Count - 1];
+                if (iv.EndUtc > last.End)
+                    merged[merged.Count - 1] = (last.Start, iv.EndUtc);
+            }
+            else
+            {
+                merged.Add((iv.StartUtc, iv.EndUtc));
+            }
+        }
+
+        double total = 0;
+        foreach (var span in espAppsSpans)
+        {
+            foreach (var (ms, me) in merged)
+            {
+                var s = ms > span.StartUtc ? ms : span.StartUtc;
+                var e = me < span.EndUtc ? me : span.EndUtc;
+                if (e > s) total += (e - s).TotalSeconds;
+            }
+        }
+        return (int)Math.Floor(total);
+    }
+
+    // ── what-if bound (fleet aggregation, F1 PR2) ───────────────────────────
+
+    /// <summary>
+    /// Upper-bound saving from removing app <paramref name="appId"/> from the ESP blocking set
+    /// of ONE session: the critical-path end (latest interval end) recomputed without the app's
+    /// interval — <c>max(0, cpEnd − cpEndWithoutX)</c> per the spec. When the app is the only
+    /// measured blocking interval, the fallback baseline is its own start (nothing else was
+    /// observed holding the path). An upper bound BY CONSTRUCTION — removing an app cannot slow
+    /// anything down, but hidden serialization may reduce real savings — so consumers must
+    /// phrase it as "up to", never as a promise (truthfulness rule 3). Returns 0 when the app
+    /// has no measured interval or does not end last.
+    /// </summary>
+    public static int WhatIfSavingSeconds(IReadOnlyList<BlockingAppInterval> intervals, string appId)
+    {
+        BlockingAppInterval? target = null;
+        DateTime? cpEnd = null, cpEndWithout = null;
+
+        foreach (var interval in intervals)
+        {
+            if (!cpEnd.HasValue || interval.EndUtc > cpEnd.Value)
+                cpEnd = interval.EndUtc;
+
+            if (string.Equals(interval.AppId, appId, StringComparison.OrdinalIgnoreCase))
+            {
+                target = interval; // one interval per app per session by construction
+            }
+            else if (!cpEndWithout.HasValue || interval.EndUtc > cpEndWithout.Value)
+            {
+                cpEndWithout = interval.EndUtc;
+            }
+        }
+
+        if (target == null || !cpEnd.HasValue) return 0;
+
+        // Spec formula verbatim: cpEnd − cpEndWithoutX. When X's removal exposes an earlier
+        // path end, the idle gap before X's start counts too — the ESP was waiting for X to
+        // START as much as to finish. Negative/zero → another app ends at or after X: no claim.
+        var baseline = cpEndWithout ?? target.StartUtc;
+        var saving = (cpEnd.Value - baseline).TotalSeconds;
+        return saving > 0 ? (int)Math.Floor(saving) : 0;
+    }
+
+    // ── reboot spans ────────────────────────────────────────────────────────
+
+    /// <summary>
+    /// One span per distinct observed boot: the event-stream gap bracketing <c>lastBootUtc</c>
+    /// (last event before the boot → first event at/after it). Fallback without a usable
+    /// payload: previous event → the detection event itself (which is genuinely post-boot —
+    /// usable as an observation bound, never as the boot moment). Spans are clipped to the
+    /// observation hull, so a WhiteGlove-pause reboot contributes nothing.
+    /// </summary>
+    private static List<RebootSpan> BuildRebootSpans(
+        List<EnrollmentEvent> events, List<Window> windows, List<TimeAttributionSpan> spans)
+    {
+        var result = new List<RebootSpan>();
+        var seenBoots = new HashSet<DateTime>();
+
+        for (var i = 0; i < events.Count; i++)
+        {
+            var evt = events[i];
+            if (evt.EventType != Constants.EventTypes.SystemRebootDetected) continue;
+
+            DateTime? lastBoot = null;
+            if (evt.Data != null && evt.Data.TryGetValue("lastBootUtc", out var lastBootObj))
+            {
+                var raw = lastBootObj?.ToString();
+                if (!string.IsNullOrEmpty(raw) &&
+                    DateTime.TryParse(raw, CultureInfo.InvariantCulture,
+                        DateTimeStyles.AssumeUniversal | DateTimeStyles.AdjustToUniversal, out var parsed))
+                {
+                    lastBoot = parsed;
+                }
+            }
+
+            DateTime? gapStart = null, gapEnd = null;
+            if (lastBoot.HasValue)
+            {
+                if (!seenBoots.Add(lastBoot.Value)) continue; // duplicate detection of the same boot
+                foreach (var e in events)
+                {
+                    if (e.Timestamp < lastBoot.Value)
+                    {
+                        if (!gapStart.HasValue || e.Timestamp > gapStart.Value) gapStart = e.Timestamp;
+                    }
+                    else if (!gapEnd.HasValue || e.Timestamp < gapEnd.Value)
+                    {
+                        gapEnd = e.Timestamp;
+                    }
+                }
+            }
+            else if (i > 0)
+            {
+                gapStart = events[i - 1].Timestamp;
+                gapEnd = evt.Timestamp;
+            }
+
+            if (!gapStart.HasValue || !gapEnd.HasValue || gapEnd.Value <= gapStart.Value) continue;
+
+            var clamped = ClampToWindows(gapStart.Value, gapEnd.Value, windows);
+            if (!clamped.HasValue) continue;
+            // In-window seconds — a reboot gap bracketing the WhiteGlove pause contributes
+            // only its in-window flanks, never the pause itself.
+            var seconds = clamped.Value.Seconds;
+            if (seconds <= 0) continue;
+
+            result.Add(new RebootSpan
+            {
+                StartUtc = clamped.Value.Start,
+                EndUtc = clamped.Value.End,
+                Seconds = seconds,
+                SegmentKey = SegmentAt(spans, clamped.Value.Start),
+            });
+        }
+
+        return result;
+    }
+
+    private static string SegmentAt(List<TimeAttributionSpan> spans, DateTime ts)
+    {
+        foreach (var span in spans)
+        {
+            if (ts >= span.StartUtc && ts < span.EndUtc)
+                return span.SegmentKey;
+        }
+        return TimeAttributionSegments.Unattributed;
+    }
+}

src/Backend/AutopilotMonitor.Functions/Middleware/PolicyEnforcementMiddleware.cs

@@ -336,6 +336,9 @@ public class PolicyEnforcementMiddleware : IFunctionsWorkerMiddleware
             case EndpointPolicy.TenantAdminOrGlobalReader:
                 return await EvaluateTenantAdminOrGlobalReaderAsync(tenantId, upn, principal, userIdentifier);
 
+            case EndpointPolicy.TenantAdminOrOperator:
+                return await EvaluateTenantAdminOrOperatorAsync(tenantId, upn, principal, userIdentifier);
+
             case EndpointPolicy.TenantAdminOrGA:
                 return await EvaluateTenantAdminOrGAAsync(tenantId, upn, principal, userIdentifier);
 
@@ -469,6 +472,28 @@ public class PolicyEnforcementMiddleware : IFunctionsWorkerMiddleware
         return CatalogDecisionResult.Deny(userIdentifier, role?.Role ?? "NonMember", "NotAdminOrGA");
     }
 
+    /// <summary>
+    /// Operational write tier: own-tenant Admin or Operator (no Viewer), or Global Admin.
+    /// The admitted member's role flows onto RequestContext.UserRole so a function that
+    /// serves multiple operation kinds can re-gate the admin-only ones on IsTenantAdmin.
+    /// The read-only Global Reader is deliberately NOT admitted — this is a write tier.
+    /// </summary>
+    private async Task<CatalogDecisionResult> EvaluateTenantAdminOrOperatorAsync(
+        string? tenantId, string? upn, ClaimsPrincipal? principal, string userIdentifier)
+    {
+        if (string.IsNullOrEmpty(tenantId) || string.IsNullOrEmpty(upn))
+            return CatalogDecisionResult.Deny(userIdentifier, "N/A", "MissingClaims");
+
+        if (await _globalAdminService.IsGlobalAdminAsync(upn))
+            return CatalogDecisionResult.Allow(userIdentifier, "GlobalAdmin", "GABypass");
+
+        var role = await ResolveEffectiveRoleAsync(tenantId, upn, principal);
+        if (role?.Role == Constants.TenantRoles.Admin || role?.Role == Constants.TenantRoles.Operator)
+            return CatalogDecisionResult.Allow(userIdentifier, role.Role, "TenantAdminOrOperator");
+
+        return CatalogDecisionResult.Deny(userIdentifier, role?.Role ?? "NonMember", "NotAdminOperatorOrGA");
+    }
+
     private async Task<CatalogDecisionResult> EvaluateBootstrapManagerOrGAAsync(
         string? tenantId, string? upn, ClaimsPrincipal? principal, string userIdentifier)
     {

src/Backend/AutopilotMonitor.Functions/Security/EndpointAccessPolicyCatalog.cs

@@ -55,11 +55,22 @@ public enum EndpointPolicy
     /// </summary>
     MemberRead,
 
+    /// <summary>
+    /// Own-tenant Admin or Operator (no Viewer), OR Global Admin. Tenant-scoped OPERATIONAL write
+    /// access: actions a troubleshooting operator may take on a session/device without holding
+    /// config-level admin power. Routes on this tier whose function accepts multiple operation
+    /// kinds MUST re-gate the admin-only kinds in the function on
+    /// <c>RequestContext.IsTenantAdmin || IsGlobalAdmin</c> (see QueueSessionActionFunction).
+    /// </summary>
+    TenantAdminOrOperator,
+
     /// <summary>
     /// Own-tenant Admin, OR a platform role (Global Admin / Global Reader). Tenant-scoped admin-tier
-    /// READ access (e.g. GET config/{tenantId}). Like <see cref="TenantAdminOrGA"/> but additionally
-    /// admits the read-only Global Reader — used only on GET routes so the reader gets visibility
-    /// without write power, while tenant Operators/Viewers still cannot read admin-tier config.
+    /// READ access (e.g. GET tenants/{tenantId}/graph-permissions/status). Like
+    /// <see cref="TenantAdminOrGA"/> but additionally admits the read-only Global Reader — used only
+    /// on GET routes so the reader gets visibility without write power, while tenant
+    /// Operators/Viewers stay excluded. (GET config/{tenantId} moved OFF this tier to MemberRead:
+    /// its handler redacts secrets for every non-admin caller, so member-tier read is safe there.)
     /// </summary>
     TenantAdminOrGlobalReader,
 
@@ -159,6 +170,11 @@ public static class EndpointAccessPolicyCatalog
         // ticket is the sole authority — authz happened at mint time (diagnostics/download-ticket,
         // MemberRead). See DiagnosticsTicketDownloadFunction.
         new("GET",    "diagnostics/download",       EndpointPolicy.PublicAnonymous),
+        // Azure Monitor action-group webhook (Common Alert Schema → ops event). Action groups
+        // cannot send custom headers, so the shared secret travels as ?secret= and is compared
+        // in constant time against the OpsAlertWebhookSecret app setting (fail-closed while
+        // unset). See AzureMonitorAlertWebhookFunction.
+        new("POST",   "ops/alert-webhook",          EndpointPolicy.PublicAnonymous),
 
         // ── DeviceOrBootstrapAuth ───────────────────────────────────────
         new("POST",   "agent/register-session",    EndpointPolicy.DeviceOrBootstrapAuth),
@@ -216,8 +232,14 @@ public static class EndpointAccessPolicyCatalog
         // (Inspector v1 — Plan §M6). Lift back to MemberRead+QueryParam at the v2 adminMode lift.
         new("GET",    "sessions/{sessionId}/analysis", EndpointPolicy.MemberRead, TenantScoping.QueryParam),
         new("GET",    "sessions/{sessionId}/vulnerability-report", EndpointPolicy.MemberRead, TenantScoping.QueryParam),
+        new("GET",    "sessions/{sessionId}/time-attribution", EndpointPolicy.MemberRead, TenantScoping.QueryParam),
         new("GET",    "metrics/app",               EndpointPolicy.MemberRead),
         new("GET",    "metrics/fleet-health",      EndpointPolicy.MemberRead),
+        new("GET",    "metrics/time-attribution",  EndpointPolicy.MemberRead),
+        // F2 device history: QueryParam scoping so the session-detail banner works cross-tenant
+        // (GA/delegated drill-in) over the SAME route MCP uses — no global duplicate needed.
+        new("GET",    "metrics/device-history",    EndpointPolicy.MemberRead, TenantScoping.QueryParam),
+        new("GET",    "metrics/device-journeys",   EndpointPolicy.MemberRead),
         new("GET",    "apps/list",                 EndpointPolicy.MemberRead),
         new("GET",    "apps/{appName}/analytics",  EndpointPolicy.MemberRead),
         new("GET",    "apps/{appName}/sessions",   EndpointPolicy.MemberRead),
@@ -233,12 +255,18 @@ public static class EndpointAccessPolicyCatalog
         new("GET",    "audit/logs",                EndpointPolicy.MemberRead),
         new("GET",    "audit/hardware-rejected", EndpointPolicy.MemberRead),
         new("GET",    "audit/device-not-registered", EndpointPolicy.MemberRead),
+        new("GET",    "audit/tpm-pss-unsupported", EndpointPolicy.MemberRead),
         new("GET",    "diagnostics/download-url",  EndpointPolicy.MemberRead, TenantScoping.QueryParam),
         new("POST",   "diagnostics/download-ticket", EndpointPolicy.MemberRead, TenantScoping.QueryParam),
         new("GET",    "rules/gather",              EndpointPolicy.MemberRead),
         new("GET",    "rules/analyze",             EndpointPolicy.MemberRead),
         new("GET",    "rules/ime-log-patterns",    EndpointPolicy.MemberRead),
-        new("GET",    "config/{tenantId}",         EndpointPolicy.TenantAdminOrGlobalReader, TenantScoping.RouteParam),
+        // Member tier so Operators/Viewers can see their tenant's configuration read-only in the
+        // Settings UI. Safe because the handler redacts by default: only a Global Admin or the
+        // tenant's OWN admin gets the unredacted secrets — every other admitted caller (Operator,
+        // Viewer, read-only Global Reader, delegated reader) receives RedactedCopyForReader()
+        // (see GetTenantConfigurationFunction.CanViewSecrets). Writes stay TenantAdminOrGA below.
+        new("GET",    "config/{tenantId}",         EndpointPolicy.MemberRead, TenantScoping.RouteParam),
         new("GET",    "config/{tenantId}/feature-flags", EndpointPolicy.MemberRead, TenantScoping.RouteParam),
 
         // ── TenantAdminOrGA ─────────────────────────────────────────────
@@ -254,9 +282,15 @@ public static class EndpointAccessPolicyCatalog
         new("PUT",    "rules/ime-log-patterns/{patternId}", EndpointPolicy.TenantAdminOrGA),
         new("POST",   "sessions/{sessionId}/mark-failed",     EndpointPolicy.TenantAdminOrGA),
         new("POST",   "sessions/{sessionId}/mark-succeeded", EndpointPolicy.TenantAdminOrGA),
-        new("POST",   "sessions/{sessionId}/actions",          EndpointPolicy.TenantAdminOrGA),
+        // Operator tier so troubleshooting staff can queue request_diagnostics ("Collect Logs" in
+        // the portal); the function re-gates the admin-only action types (terminate_session,
+        // rotate_config) on RequestContext.IsTenantAdmin || IsGlobalAdmin.
+        new("POST",   "sessions/{sessionId}/actions",          EndpointPolicy.TenantAdminOrOperator),
         new("POST",   "sessions/{sessionId}/report",        EndpointPolicy.TenantAdminOrGA),
-        new("POST",   "diagnostics/files",                  EndpointPolicy.TenantAdminOrGA),
+        // Operator tier: submitting diagnostic files to the Autopilot Monitor team is a
+        // troubleshooting action, not a config change — a support-driven Operator may use it.
+        // Viewer and the read-only Global Reader stay excluded (the tier admits no Viewer/Reader).
+        new("POST",   "diagnostics/files",                  EndpointPolicy.TenantAdminOrOperator),
         new("DELETE", "sessions/{sessionId}",      EndpointPolicy.TenantAdminOrGA, TenantScoping.QueryParam),
         new("GET",    "tenants/{tenantId}/admins",           EndpointPolicy.TenantAdminOrGA, TenantScoping.RouteParam),
         new("POST",   "tenants/{tenantId}/admins",           EndpointPolicy.TenantAdminOrGA, TenantScoping.RouteParam),
@@ -371,6 +405,8 @@ public static class EndpointAccessPolicyCatalog
         new("GET",    "global/metrics/platform",    EndpointPolicy.GlobalReadOrAdmin),
         new("GET",    "global/metrics/app",         EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
         new("GET",    "global/metrics/fleet-health", EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
+        new("GET",    "global/metrics/time-attribution", EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
+        new("GET",    "global/metrics/device-journeys", EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
         new("GET",    "global/apps/list",           EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
         new("GET",    "global/apps/{appName}/analytics", EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
         new("GET",    "global/apps/{appName}/sessions",  EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
@@ -388,6 +424,15 @@ public static class EndpointAccessPolicyCatalog
         new("PATCH",  "global/session-reports/{reportId}/note", EndpointPolicy.GlobalAdminOnly),
         new("GET",    "global/rules/gather",        EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
         new("GET",    "global/rules/analyze",       EndpointPolicy.GlobalReadOrAdmin, TenantScoping.QueryParam),
+        // Cross-tenant rule WRITES (edit/toggle/delete an existing rule in a foreign tenant). GlobalAdminOnly
+        // — a write tier, so the delegated-read rescue never fires: only a platform admin may mutate another
+        // tenant's rules. QueryParam because the tenant is named by ?tenantId= (the {ruleId} segment is not a
+        // tenant). No global CREATE route: creating rules in a foreign tenant stays disabled (the portal hides
+        // Create in the cross-tenant override), so the JWT-scoped POST routes remain the only create path.
+        new("PUT",    "global/rules/gather/{ruleId}",  EndpointPolicy.GlobalAdminOnly, TenantScoping.QueryParam),
+        new("DELETE", "global/rules/gather/{ruleId}",  EndpointPolicy.GlobalAdminOnly, TenantScoping.QueryParam),
+        new("PUT",    "global/rules/analyze/{ruleId}", EndpointPolicy.GlobalAdminOnly, TenantScoping.QueryParam),
+        new("DELETE", "global/rules/analyze/{ruleId}", EndpointPolicy.GlobalAdminOnly, TenantScoping.QueryParam),
         new("GET",    "global/devices/blocked",     EndpointPolicy.GlobalReadOrAdmin),
         new("GET",    "devices/blocked",            EndpointPolicy.GlobalReadOrAdmin),
         new("POST",   "devices/block",              EndpointPolicy.GlobalAdminOnly),

src/Backend/AutopilotMonitor.Functions/Services/Deletion/DeletionManifestBuilder.cs

@@ -114,17 +114,20 @@ namespace AutopilotMonitor.Functions.Services.Deletion
             await AddDiscriminatorPkPropStepAsync(manifest, order: 14, table: Constants.TableNames.ClassifierVerdictsByIdLevel, tenantId, sessionId, safeTenantId, safeSessionId, cancellationToken);
             await AddDiscriminatorPkPropStepAsync(manifest, order: 15, table: Constants.TableNames.SignalsByKind, tenantId, sessionId, safeTenantId, safeSessionId, cancellationToken);
 
-            // ---- Steps 16 + 17: SoftwareInventory side-row (omit both for pre-side-row sessions). ----
+            // ---- Step 16: F1 time-attribution breakdown (PK=tenant, RK=sessionId; PR2). ----
+            await AddPkRkExactStepAsync(manifest, order: 16, table: Constants.TableNames.SessionTimeBreakdowns, partitionKey: tenantId, rowKey: sessionId, cancellationToken);
+
+            // ---- Steps 17 + 18: SoftwareInventory side-row (omit both for pre-side-row sessions). ----
             var contributionsRow = await _reader.GetEntityOrNullAsync(
                 Constants.TableNames.SessionInventoryContributions, tenantId, sessionId, cancellationToken);
             if (contributionsRow != null)
             {
-                AddSoftwareInventoryDecrementStep(manifest, order: 16, contributionsRow);
-                AddContributionsRowStep(manifest, order: 17, contributionsRow);
+                AddSoftwareInventoryDecrementStep(manifest, order: 17, contributionsRow);
+                AddContributionsRowStep(manifest, order: 18, contributionsRow);
             }
 
-            // ---- Step 18: Tombstone (SessionsIndex first, then Sessions). ----
-            AddTombstoneStep(manifest, order: 18, sessionsIndexRow, sessionRow);
+            // ---- Step 19: Tombstone (SessionsIndex first, then Sessions). ----
+            AddTombstoneStep(manifest, order: 19, sessionsIndexRow, sessionRow);
 
             // PreflightCounts derive from each step's RowCount, plus the AGGREGATE decrements length.
             manifest.PreflightCounts = ComputePreflightCounts(manifest);

src/Backend/AutopilotMonitor.Functions/Services/Diagnostics/DiagnosticsBlobCascadeDeleter.cs

@@ -75,10 +75,23 @@ namespace AutopilotMonitor.Functions.Services.Diagnostics
 
             if (destination == DestinationHosted)
             {
+                // Prefix sweep first: the session may have produced several packages (on-demand
+                // server-requested uploads plus the terminal one) while the manifest carries only
+                // the last row-referenced name. Guarded by TryParse because a malformed legacy
+                // manifest must degrade to the single-blob delete, not poison the cascade.
+                var swept = 0;
+                if (Guid.TryParse(manifest.SessionId, out _))
+                {
+                    swept = await _hostedDiagnostics.DeleteBySessionPrefixAsync(
+                        manifest.TenantId, manifest.SessionId, cancellationToken);
+                }
+
+                // Belt-and-braces: the manifest-named blob covers any shape outside the canonical
+                // {tenantId}/AgentDiagnostics-{sessionId}- layout. Idempotent (404 = no-op).
                 await _hostedDiagnostics.DeleteIfExistsAsync(manifest.DiagnosticsBlobName, cancellationToken);
                 _logger.LogInformation(
-                    "Cascade deleted hosted diagnostics blob tenant={TenantId} blob={Blob}",
-                    manifest.TenantId, manifest.DiagnosticsBlobName);
+                    "Cascade deleted hosted diagnostics blobs tenant={TenantId} blob={Blob} sweptByPrefix={Swept}",
+                    manifest.TenantId, manifest.DiagnosticsBlobName, swept);
                 return DiagnosticsBlobDeleteOutcome.HostedDeleted;
             }
 

src/Backend/AutopilotMonitor.Functions/Services/Diagnostics/HostedDiagnosticsBlobService.cs

@@ -165,6 +165,40 @@ namespace AutopilotMonitor.Functions.Services.Diagnostics
             return response.Value;
         }
 
+        /// <summary>
+        /// Deletes every hosted diagnostics blob belonging to one session
+        /// (prefix <c>{tenantId}/AgentDiagnostics-{sessionId}-</c>). A session can leave more
+        /// than one blob behind — an on-demand ("Collect Logs" / server-requested) upload is
+        /// later overwritten on the Sessions row by the terminal package, so deleting only the
+        /// row-referenced name would strand the earlier blobs at offboarding.
+        /// Returns the number of blobs deleted. Idempotent; a missing container is a no-op.
+        /// </summary>
+        public virtual async Task<int> DeleteBySessionPrefixAsync(
+            string tenantId, string sessionId, CancellationToken cancellationToken = default)
+        {
+            SecurityValidator.EnsureValidGuid(tenantId, nameof(tenantId));
+            SecurityValidator.EnsureValidGuid(sessionId, nameof(sessionId));
+
+            var containerClient = GetContainerClient();
+            if (!await containerClient.ExistsAsync(cancellationToken))
+                return 0;
+
+            // Trailing dash bounds the match to this exact session id (GUIDs are fixed-length,
+            // but the explicit separator keeps the invariant obvious and layout-proof).
+            var prefix = $"{tenantId}/AgentDiagnostics-{sessionId}-";
+            var deleted = 0;
+            await foreach (var item in containerClient.GetBlobsAsync(
+                BlobTraits.None, BlobStates.None, prefix: prefix, cancellationToken: cancellationToken))
+            {
+                var response = await containerClient.GetBlobClient(item.Name)
+                    .DeleteIfExistsAsync(cancellationToken: cancellationToken);
+                if (response.Value)
+                    deleted++;
+            }
+
+            return deleted;
+        }
+
         /// <summary>
         /// Enumerates blobs under a single tenant's prefix (<c>{tenantId}/</c>).
         /// Used by ad-hoc operator scripts and (later) any orphan-scan job.