build(deps): bump the web group (#135)

Bumps the web group in /src/Web/autopilot-monitor-web with 4 updates: [next](https://github.com/vercel/next.js), [@next/bundle-analyzer](https://github.com/vercel/next.js/tree/HEAD/packages/next-bundle-analyzer), [eslint-config-next](https://github.com/vercel/next.js/tree/HEAD/packages/eslint-config-next) and [vitest](https://github.com/vitest-dev/vitest/tree/HEAD/packages/vitest).


Updates `next` from 16.3.1 to 16.3.2
- [Release notes](https://github.com/vercel/next.js/releases)
- [Commits](https://github.com/vercel/next.js/compare/v16.3.1...v16.3.2)

Updates `@next/bundle-analyzer` from 16.3.1 to 16.3.2
- [Release notes](https://github.com/vercel/next.js/releases)
- [Commits](https://github.com/vercel/next.js/commits/v16.3.2/packages/next-bundle-analyzer)

Updates `eslint-config-next` from 16.3.1 to 16.3.2
- [Release notes](https://github.com/vercel/next.js/releases)
- [Commits](https://github.com/vercel/next.js/commits/v16.3.2/packages/eslint-config-next)

Updates `vitest` from 4.1.10 to 4.1.11
- [Release notes](https://github.com/vitest-dev/vitest/releases)
- [Changelog](https://github.com/vitest-dev/vitest/blob/main/docs/releases.md)
- [Commits](https://github.com/vitest-dev/vitest/commits/v4.1.11/packages/vitest)

---
updated-dependencies:
- dependency-name: next
  dependency-version: 16.3.2
  dependency-type: direct:production
  update-type: version-update:semver-patch
  dependency-group: web
- dependency-name: "@next/bundle-analyzer"
  dependency-version: 16.3.2
  dependency-type: direct:development
  update-type: version-update:semver-patch
  dependency-group: web
- dependency-name: eslint-config-next
  dependency-version: 16.3.2
  dependency-type: direct:development
  update-type: version-update:semver-patch
  dependency-group: web
- dependency-name: vitest
  dependency-version: 4.1.11
  dependency-type: direct:development
  update-type: version-update:semver-patch
  dependency-group: web
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

docs/agent/decision-engine.md

@@ -128,6 +128,15 @@ user-ESP Install-intent apps are still pending (IME logs "Completed user session
 *pass*, not once per session — sessions 14690fc2/6cb01530). The reducer stays free of
 app-pending guards; adapters sanitize evidence before it becomes a signal.
 
+**Zero user-targeted apps** (sessions 81daa77f/75d6ae8e): IME never writes its
+"Completed user session" line at all when the user has no assigned apps — its zero-app
+branch returns first — so fact 3 was unreachable by construction and every such device
+fell through to the backstop. `IME-USER-SESSION-ZERO-APPS` reads the line IME *does*
+write there and posts the same signal, guarded to the AccountSetup phase. The event
+carries an `evidence` field (`zero_user_apps_enumerated` vs `user_session_completed`)
+naming which observation carried the completion. See
+[zero-user-apps-completion-gate.md](zero-user-apps-completion-gate.md).
+
 ## Completion attempt sites and sequence
 
 Completion can be attempted from whichever fact arrives last: `HelloResolved`,
@@ -166,6 +175,11 @@ advisory backstop. All routes converge on `CompleteThroughFinalizingOrDefer`:
    scheduler contract) dead-ends as `realmjoin_timeout_stale_superseded_by_rearm`.
    The idle case (detected, no deployment activity ever) times out at 60 min
    unchanged, and phase 200/210 activity never extends.
+   A closed gate also outranks the 30-minute `AdvisoryCompletion` backstop: that
+   handler re-arms (trigger suffix `:CompletionGateHolding`) instead of failing while
+   the RealmJoin gate is closed *and* its own deadline is still armed, because the gate
+   carries a bounded resolution of its own. Before this, the shorter timer always won —
+   session 75d6ae8e was failed 2 min after its 25th successful RealmJoin package.
 2. `Finalizing` stage + ~5 s `FinalizingGrace` deadline.
 3. Grace fires → `Completed`, `EnrollmentComplete` outcome, `enrollment_complete`
    timeline event with full audit payload.

docs/agent/index.md

@@ -7,6 +7,7 @@ How the on-device agent works:
 * [Logs, Persistence & Crash Recovery](logs-and-persistence.md) - every persisted file and why it exists; signal log vs spool; snapshot/journal; the replay-based recovery flow and quarantine.
 * [Diagnostics Package](diagnostics-package.md) - what the diagnostics ZIP contains: the built-in section catalog shared with the portal (Shared `DiagnosticsBuiltInSections`, served by GET diagnostics/paths), the RealmJoin Watcher and Device Preparation collection gates, why built-ins bypass the configured-path guards, global ∪ tenant path dedup, caps and the package manifest.
 * [Hello Wizard Un-Skip](hello-wizard-unskip.md) - why a single "WHfB disabled" policy read never decides completion: the HelloWizardStarted signal (prevention + cure) and the HelloTracker confirmation second read.
+* [Zero User-Targeted Apps](zero-user-apps-completion-gate.md) - why a user with no targeted Intune apps starved every user-phase completion path (IME never writes its "Completed user session" line in the zero-app branch), and why a completion gate that is still holding the session must outrank the 30-minute backstop.
 * [Autopilot ZTD Diagnostics](autopilot-ztd-diagnostics.md) - Windows' own diagnostic surfaces for the profile-download flow: ModernDeployment event IDs (807/815/164/...), the Diagnostics\Autopilot registry key, deployment-service endpoints, and the known-issue error-code map.
 * [MDM Reboot Coalescing](mdm-reboot-coalescing.md) - attributing the mid-ESP coalesced reboot ("second sign-in") to the device-assigned policy URIs that forced it: aggregated 2800 observations (neutral, flush-time watermark) + ANALYZE-ESP-005 gated on an actually-observed reboot.
 * [Power-State Watcher](power-state-watcher.md) - live AC↔battery transitions and 50/30/15% downward threshold crossings via a WMI push subscription (no polling): power_state_change semantics (latch/debounce/cap, never armed without a battery) and the single-condition analyze rules ANALYZE-DEV-009/-010 on top.

docs/agent/zero-user-apps-completion-gate.md

@@ -0,0 +1,136 @@
+---
+type: Concept
+title: Zero User-Targeted Apps — the starved user-phase completion gate
+description: Why a device with no user-targeted Intune apps could never satisfy the IME user-session gate, how that turned into a 30-minute esp_exit_without_completion_evidence false positive on a third of one tenant's enrollments, and the two fixes — reading the line IME actually writes in its zero-app branch, and refusing to fail a session while a completion gate is deliberately holding it.
+resource: /src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/Ime
+tags:
+  - agent
+  - ime
+  - esp
+  - completion
+  - realmjoin
+  - false-positive
+timestamp: 2026-08-25
+---
+
+# Summary
+
+A Classic user-driven enrollment whose user has **zero targeted Intune Win32 apps** could not
+produce the evidence every user-phase completion path required. The session parked, and the
+30-minute `advisory_completion` backstop resolved it as
+`enrollment_failed: esp_exit_without_completion_evidence` — while the user was at a working
+desktop, Windows Hello was provisioned, and RealmJoin was still installing packages.
+
+One tenant hit this on **10 of 30 enrollments in a week**. Every one of them was a false
+positive; the devices were healthy in Intune and RealmJoin.
+
+# The starvation chain
+
+The IME writes `Completed user session N, userId: …` at the end of a user-session processing
+pass. `IME-USER-SESSION-COMPLETED` matches it and the adapter posts
+`ImeUserSessionCompleted`, which sets `DecisionState.ImeUserSessionCompletedUtc`.
+
+But the IME never reaches that statement when the user has no assigned apps. Decompiled IME
+`1.104.102.0`, `Win32AppPlugIn/ApplicationPoller.cs`:
+
+```csharp
+if (!appPolicies.Any())
+{
+    AppWorkloadLog.TraceInformation($"[Win32App] Get 0 apps for user session {user.RDSSessionId}, user id = {user.IntuneId}");
+    RegisterApplicationsInIfInESP(sender, user, delegate {
+        espLockInProcessor.AttemptToLockInApplications(user, new List<ProcessorSubgraph>());
+    });
+    continue;                       // skips the "Completed user session" statement below
+}
+```
+
+Identical shape in `1.97.107.0`, `1.83.107.0`, `1.80.132.0`. With that fact missing, every door
+out of the user phase is shut at once:
+
+| Path | Requirement | Why it is dead |
+|---|---|---|
+| `ShouldTransitionToAwaitingHello` arm A | `AccountSetupProvisioningSucceededUtc` | ESP registry froze at `AccountSetup: 1 of 5`; neither `categorySucceeded` nor the all-subcategories fallback can ever hold |
+| arm B | `SkipUserEsp` | Full ESP, not a skip-user flow |
+| arm C | `IsImeUserSessionGenuine` | needs the missing IME fact |
+| arm D | `EnrollmentMode.DevicePreparation` | Autopilot v1, not WDP |
+| `MaybeSynthesizeAccountSetupCompleteFromSettledUserApps` | `AreUserEspAppsSettled` | returned `false` on an empty app list |
+| `HandleAdvisoryCompletionDeadlineFired` conjunction | `imeUserSessionGenuine \|\| isObservationAdvisory` | both false |
+
+The last row is the one that produced the verdict: the backstop that exists to resolve a
+dead-end resolved it as a failure.
+
+`AreUserEspAppsSettled` deserves its own note. Its "at least one required app" rule is
+deliberately conservative — an empty list normally means *the phase just cleared* or *apps have
+not surfaced yet*. But it made **zero apps**, the most settled state possible, indistinguishable
+from *unknown*, and there was no evidence available to tell them apart.
+
+# Fix 1 — read the line IME does write
+
+`IME-USER-SESSION-ZERO-APPS` matches the zero-app line and routes it through the **same**
+`OnUserSessionCompleted` callback. Downstream this genuinely *is* a completed user session, so
+the adapter's pending-apps deferral and its fire-once flag apply unchanged.
+
+This is not a weakened gate — it is IME's own verdict. The zero-app branch locks the ESP in with
+an *empty* application list, and the sibling emit site in `AppWorkloadAbstraction.ExecuteUserCheckIn`
+returns `ProviderResult.ProvisioningComplete` straight after logging it. All three emit sites are
+user-session-scoped (`user.RDSSessionId`, `user.IntuneId`); none is device-scoped.
+
+Guards that stay in force:
+
+* **Phase guard** — the observation only fires while the tracker is in `AccountSetup`. Firing
+  earlier would burn the adapter's fire-once flag on a pre-sign-in (defaultuser0) timestamp that
+  `IsImeUserSessionGenuine` (`>= AccountSetupEnteredUtc`) can never accept. An observation made
+  before the phase was known is remembered and replayed on the transition into it, because IME
+  writes its check-in lines independently of when the agent first parses the phase marker.
+* **Engine guards unchanged** — arm C still requires a post-AccountSetup final ESP exit *and* a
+  DAD-validated real-user desktop.
+* `AreUserEspAppsSettled` returns `true` for an empty list **only** with this explicit
+  observation — never for a merely empty list.
+
+# Fix 2 — do not fail while a completion gate is holding
+
+An independent defect, and the decisive one for the affected tenant. `advisory_completion` runs
+for 30 minutes. The RealmJoin completion gate blocks completion until phase 110 and carries its
+*own* bounded resolution: 60 minutes from detection, re-armed on deployment activity, hard-capped
+at 4 hours. The shorter timer always won.
+
+Session `75d6ae8e` installed **25 RealmJoin packages successfully inside the window**, the last
+one 2 minutes before the failure, while the engine's own `completion_waiting` still read
+`waiting on: realmjoin_resolution`.
+
+`HandleAdvisoryCompletionDeadlineFired` now re-arms instead of failing when the RealmJoin gate is
+closed *and* its deadline is still armed, emitting `completion_waiting` with trigger
+`DeadlineFired:advisory_completion:CompletionGateHolding`.
+
+Convergence comes from the gate itself: `Resolved`, `FirstDeploymentIncomplete` and `Timeout` all
+set `RealmJoinFacts.Outcome`, which opens `RealmJoinGateOpen` and stops the guard applying.
+Requiring the gate's deadline to still be armed keeps a gate that can no longer resolve itself
+from parking the session; `AgentMaxLifetimeMinutes` bounds it regardless.
+
+# Timeline evidence
+
+`ime_user_session_completed` carries an `evidence` field so the basis of the verdict is legible
+without the diagnostics archive:
+
+* `user_session_completed` — IME wrote its completion line and no required app was pending.
+* `zero_user_apps_enumerated` — IME reported zero targeted apps; the user phase had nothing to
+  enforce.
+
+No new event type: the fact is the same, only the observation behind it differs.
+
+# Deployment
+
+The pattern ships through the config API and reaches existing agents without a rebuild; an agent
+that does not know the `userSessionZeroApps` action logs one Debug line and ignores it. The
+*behaviour* needs an agent release. Sessions already stamped `Failed` are not reclassified — the
+backend mirrors the agent verdict (`VerdictPaths.AgentFailed`) and has no independent judgement
+here by design.
+
+# Citations
+
+* `rules/ime-log-patterns/IME-USER-SESSION-ZERO-APPS.json`
+* `src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/Ime/ImeLogTracker.Handlers.cs` — `HandleUserSessionZeroApps`, the phase replay in `HandleEspPhaseDetected`
+* `src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/Ime/ImeLogTracker.cs` — `AreUserEspAppsSettled`
+* `src/Agent/AutopilotMonitor.Agent.V2.Core/SignalAdapters/ImeLogTrackerAdapter.cs` — `EmitUserSessionCompleted`, evidence labels
+* `src/Shared/AutopilotMonitor.DecisionCore/Engine/DecisionEngine.Edge.cs` — `HandleAdvisoryCompletionDeadlineFired`
+* [decision-engine.md](decision-engine.md), [post-reboot-completion-recovery.md](post-reboot-completion-recovery.md)

docs/backend/client-cert-tenant-binding.md

@@ -0,0 +1,216 @@
+---
+type: Concept
+title: Client-Certificate Tenant Binding — Closing the Cross-Tenant Gap
+description: Why a chain-valid Intune MDM certificate does not prove the caller belongs to the tenant it claims, which certificate extension carries the Entra tenant id (and which two look like it but do not), and the shadow-then-enforce rollout that measures field coverage before it can lock devices out.
+resource: src/Backend/AutopilotMonitor.Functions/Security/CertTenantBinding.cs
+tags:
+  - backend
+  - security
+  - authentication
+  - certificates
+  - intune
+  - multi-tenant
+timestamp: 2026-08-24T00:00:00+02:00
+---
+
+# Problem
+
+Agent requests authenticate with mTLS. `CertificateValidator` pins the chain to the embedded
+Intune roots (`X509ChainTrustMode.CustomRootTrust`), checks the validity window and requires the
+Client-Authentication EKU. All of that together proves exactly one thing: **the certificate was
+issued by the Microsoft Intune MDM Device CA to some tenant**.
+
+Those roots are shared by every Intune tenant on the planet. Nothing in the certificate stage ties
+the caller to the tenant in the request. The tenant scoping came entirely from the downstream
+device validators — the serial number must appear in that tenant's Autopilot / imported-device /
+Cloud PC inventory. Device serial numbers are not secrets; they are printed on the chassis.
+
+So the pre-existing attack was: enroll a device in your own Intune tenant, obtain a legitimate
+device certificate, learn one serial number of the victim tenant, and submit telemetry as that
+tenant. Certificate validation passes, device validation passes.
+
+# Mechanism
+
+The Intune MDM Device CA stamps the customer's identifiers into the leaf certificate. Since the
+agent presents this certificate in the TLS handshake, its contents are proof of possession and the
+backend can compare them against the requested tenant — no Graph call, no additional consent, and
+nothing the agent has to send.
+
+## Which extension carries what
+
+Verified by decoding a real field certificate (issued 2026-05-08, `Functions.Tests/device-cert-sample.pem`):
+
+| OID | Encoding | Content |
+| --- | --- | --- |
+| `1.2.840.113556.5.14` | nested OCTET STRING (`04 10` + 16 bytes) | **Entra tenant id** — the one to compare |
+| `1.2.840.113556.5.6` | nested OCTET STRING (`04 10` + 16 bytes) | Intune **Account** id — a *different* GUID, never equal to the tenant id |
+| `1.2.840.113556.5.4` | **16 raw bytes, no wrapper** | Intune device id, identical to the certificate's Subject CN |
+
+Two traps are worth stating explicitly, because both cost real debugging time:
+
+1. `…5.6` is widely described as "the tenant id (AccountID)". It is not the Entra tenant id.
+   Comparing it against the requested tenant would reject every request in production.
+2. The encodings are not uniform. `…5.4` holds the GUID bare while the other two nest it in a
+   second OCTET STRING, so a decoder written against one shape silently fails on the other.
+   `MsDeviceCertificateOids.TryParseGuid` accepts both, and all GUIDs are Microsoft
+   little-endian (what `new Guid(byte[])` expects).
+
+A 16-byte input is inherently ambiguous — it is indistinguishable from a truncated wrapper and is
+always read as a bare GUID. That is harmless here: the value is only ever compared for equality
+against a known tenant id, never trusted as a standalone claim.
+
+## Where it is NOT
+
+`EntraDeviceCertHelper` reads tenant and device ids from the **MS-Organization-Access** certificate
+(`1.2.840.113556.1.5.284.*`). That certificate is not part of the authentication path at all — the
+agent selects the Intune MDM certificate by exact issuer match (`CN=Microsoft Intune MDM Device CA`,
+see `CertificateHelper.SelectMdmCertificate`) and the backend pins only the MDM chain. The Entra
+certificate is used purely as a local fallback for `TenantIdResolver`; it never leaves the device.
+
+Sending it in a header instead would prove nothing: only the certificate used in the TLS handshake
+demonstrates possession of a private key. A certificate copied into a custom header can be replayed
+by anyone who has seen it once.
+
+# Rollout
+
+Enforcing immediately would be reckless: any device population whose certificates predate the
+extension, or whose CA behaves differently (sovereign clouds), would be locked out at once. The
+rollout therefore has staged consequences.
+
+**Stage 1 — shadow (current).** `SecurityValidator.ObserveCertTenantBinding` evaluates the binding
+after certificate validation and logs the outcome. It returns `void`, changes no state, and swallows
+its own exceptions, so it cannot fail a request that would otherwise be accepted. The pure comparison
+lives in `CertTenantBinding.Evaluate`; the outcome codes (`Match`, `Mismatch`, `ExtensionMissing`,
+`Unparseable`, `RequestTenantNotAGuid`) are stable and queried by exact match in KQL.
+`WouldRejectUnderEnforcement` records what stage 2 *would* have done, so the telemetry can be read
+without re-deriving the rule at query time.
+
+The telemetry uses two carriers, because the numerator and the denominator have very different
+costs and answer different questions:
+
+* **Every** outcome, `Match` included, is stamped onto the request row as the `CertTenantBinding`
+  dimension — `SecurityValidator` writes it into `FunctionContext.Items`
+  (`CertTenantBinding.RequestItemKey`) and `RequestTelemetryMiddleware` copies it onto the
+  `RequestTelemetry` it already emits. That row exists per request and is unsampled, so the
+  denominator costs no additional telemetry at all.
+* Outcomes that need an enforcement decision (mismatch, missing, undecodable) are **additionally**
+  logged at Warning with the full context — thumbprint, certificate tenant, session.
+
+`Match` is deliberately not a trace line. Worker-side `LogInformation` never reaches App Insights
+(the ApplicationInsightsLoggerProvider default rule is Warning+ and host-forwarding of worker logs
+is disabled), so a `Match` trace is dead code that silently costs the denominator. The first
+deployment of this feature made exactly that mistake: it emitted zero rows across 32k requests,
+which is indistinguishable from the code never running.
+
+**Stage 2 — enforce.** Once telemetry shows the `ExtensionMissing` rate is negligible, `Mismatch`
+becomes a rejection. The decision that still has to be made is the policy for certificates without
+the extension: skip with a warning, or reject. Grep marker for every stage-2 site:
+`CERT-TENANT-BINDING-SHADOW`.
+
+The bootstrap-token path (`X-Bootstrap-Token`) bypasses certificate validation by design and is
+unaffected by either stage.
+
+**Stage 3 — device binding (Global-Admin-only preview).** Tenant binding proves the certificate was
+issued to this tenant; it does not prove the certificate belongs to a device the tenant still has.
+`IntuneDeviceBindingValidator` closes that by resolving the Intune managedDevice id from the
+certificate's Subject CN against the tenant's own `deviceManagement/managedDevices` inventory
+(point-GET on `managedDevices/{id}` — no OData filter to inject into, and a 404 is unambiguous).
+
+It is modelled on `CloudPcDeviceValidator` and shares its cache/retry/transient contract, with one
+important difference in how it is wired: the lookup is fire-and-forget, like the DevPrep
+association shadow, because a cold Graph round-trip can take seconds and must never sit in the
+agent's request path. That rules out the request-row dimension stage 1 uses — the row is written
+the moment the function returns, before the detached task finishes — so the outcome is logged at
+Warning instead. Volume stays bounded because only Global Admins can see the toggle
+(`user?.isGlobalAdmin === true`, the same gate as the DevPrep preview). Widening it beyond preview
+should move to an inline call plus a request dimension.
+
+Outcomes: `Match`, `NotFound`, `NoDeviceIdInCert`, `PermissionMissing`, `Transient`. Only
+`Transient` is retried and left uncached — a missing grant or a foreign device is a state, not an
+outage.
+
+Graph cost is not a concern, and that is measured rather than assumed: the first live enrollment
+produced **136 requests and exactly one Graph call**, because the 30-minute positive cache absorbs
+the rest. For comparison, the per-request Autopilot serial validator made 25 calls in the same
+window. Lengthening the positive TTL would buy nothing and would delay noticing a device that has
+been wiped or removed; shortening the 5-minute negative TTL matters more, since that is how quickly
+a device object appearing late is picked up.
+
+What the 136 requests *did* cost was log lines saying the same thing 136 times, so the shadow log is
+gated on `IntuneDeviceBindingResult.ServedFromCache`: one line per real Graph lookup. Cached repeats
+add nothing — the reported age just counts up mechanically from the same `enrolledDateTime`.
+Outcomes that are never cached (`Transient`, `NoDeviceIdInCert`) still log every time, which is the
+right behaviour for an anomaly.
+
+The open question this preview exists to answer is a race, not an attack: a device object can in
+principle appear in Intune *after* the agent's first call, and enforcing before that is understood
+would reject legitimate enrollments. Every result therefore carries `enrolledDateTime`, and the log
+line records `enrolledAgeSeconds` — a device whose object was created seconds before the request
+looks completely different from a certificate that never belonged to the tenant. Grep marker:
+`CERT-DEVICE-BINDING-SHADOW`.
+
+Enabling it needs the optional Graph permission `DeviceManagementManagedDevices.Read.All`, granted
+per tenant with the customer script — it is not in the default-consent set:
+
+```powershell
+.\Grant-AutopilotMonitorAddOn.ps1 -ClientId "<client-id>" -Features IntuneDeviceBinding
+```
+
+`GraphFeatureCatalog` (feature `IntuneDeviceBinding`) and the script's catalog, `ValidateSet` and
+help block are pinned to each other by `GraphFeatureCatalogSyncTests`.
+
+# Examples
+
+Coverage — the question stage 2 depends on ("would enforcing lock anyone out?"):
+
+```kusto
+requests
+| where isnotempty(customDimensions.CertTenantBinding)
+| summarize Requests = sum(itemCount)
+    by Outcome = tostring(customDimensions.CertTenantBinding),
+       Tenant  = tostring(customDimensions.TenantId)
+| order by Requests desc
+```
+
+Detail on the individually actionable ones:
+
+```kusto
+traces
+| where message has "AgentCertTenantBinding"
+| extend Outcome = tostring(customDimensions.Outcome),
+         Tenant  = tostring(customDimensions.TenantId)
+| summarize count() by Outcome, Tenant
+```
+
+Anything in `Mismatch` is a cross-tenant certificate; anything in `ExtensionMissing` is a device
+that stage-2 enforcement would have to make a policy decision about. Note that the trace query
+alone can never show `Match`, so it cannot tell "everything matched" apart from "nothing ran" —
+use the request-row query for that.
+
+Stage 3, including the enrollment-race measurement:
+
+```kusto
+traces
+| where message has "AgentCertDeviceBinding"
+| extend Outcome  = tostring(customDimensions.Outcome),
+         AgeSecs  = toint(customDimensions.EnrolledAgeSeconds)
+| summarize Requests = count(),
+            MinAgeSecs = min(AgeSecs),
+            P05AgeSecs = percentile(AgeSecs, 5)
+    by Outcome
+```
+
+`NotFound` is the row that decides whether this can ever become a gate. A small `MinAgeSecs` on
+`Match` means device objects are appearing only just in time — the race is real and enforcement
+would need a grace path.
+
+# Citations
+
+* `src/Shared/AutopilotMonitor.Shared/Security/MsDeviceCertificateOids.cs` — OID catalog and the
+  shared ASN.1 GUID decoder (one implementation for agent and backend).
+* `src/Backend/AutopilotMonitor.Functions/Security/CertificateValidator.cs` — extension extraction
+  alongside chain validation, cached per thumbprint.
+* `src/Backend/AutopilotMonitor.Functions/Security/CertTenantBinding.cs` — the pure comparison.
+* `src/Backend/AutopilotMonitor.Functions/Security/SecurityValidator.cs` — shadow observation.
+* [W365 Cloud PC Device Validation](cloudpc-device-validation.md) — the other consumer of this
+  certificate's identity fields (Subject CN = Intune device id).

docs/backend/device-journeys.md

@@ -78,7 +78,8 @@ Durations shown for attempts are the sessions' authoritative `DurationSeconds` v
   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
+  the sweep, which slices the tick's shared projected window scan — see
+  `time-attribution.md`). 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=…)").

docs/backend/silent-session-classification.md

@@ -0,0 +1,110 @@
+---
+type: Concept
+title: Silent-Session Classification (Sweep, Max-Lifetime, Retro-Reconcile)
+description: How a session whose agent went silent gets its honest status — the two-stage maintenance sweep (2h Stalled marker, 5h timeout classification), the EnrollmentTimeoutClassifier rule order incl. the WhiteGlove Part-2 and self-deploying gates, which ingest paths share the classifier, and the admin retro-reconcile modes.
+resource: src/Backend/AutopilotMonitor.Functions/Services/EnrollmentTimeoutClassifier.cs
+tags:
+  - backend
+  - maintenance
+  - session-status
+  - classification
+  - whiteglove
+  - self-deploying
+timestamp: 2026-08-23T18:00:00+02:00
+---
+
+# Silent-Session Classification
+
+A session only ever becomes terminal through evidence. When the agent stops reporting
+(power-off, reboot without relaunch, Wi-Fi never re-associating, max-lifetime watchdog),
+the backend must decide the status from what it last saw — it never hard-fails a session
+for being silent. One pure function holds that decision,
+`EnrollmentTimeoutClassifier.ClassifyTimedOutSession`, and every silent-session path calls it
+so a session gets the same verdict regardless of which path noticed the silence first.
+
+# Schema
+
+## Inputs (`EspProvisioningRollup` + session flags)
+
+`ExtractRollup(events)` distills the event stream into facts: Device/Account Setup
+subcategory counts (incl. the 30s fallback-confirmed messages), `HasExplicitFailure`,
+`HasEnrollmentComplete` vs. `HasTerminalComplete` (the Part-1 `whiteglove_complete`),
+`HasAgentEmergencyBreak`, `DesktopArrived`, `HelloResolved`, `HelloPolicyDisabled`,
+`SkipUserEsp`, RealmJoin detection/resolution. Session-row flags are passed alongside:
+`IsPreProvisioned` + `ResumedAt` (WhiteGlove Part 2) and `IsSelfDeployingProfile`
+(registry-confirmed `CloudAssignedOobeConfig 0x20|0x40`, set at registration).
+
+## Rule order (first match wins)
+
+| # | Condition | Verdict |
+| --- | --- | --- |
+| 1 | `HasExplicitFailure` | Failed |
+| 1b | `IsWhiteGloveAwaitingUser`: pre-provisioned + resumed + Device Setup done + **no** user evidence since resume + no explicit failure/enrollment_complete | AwaitingUser within grace; Succeeded (honest WG reason) past grace |
+| 1c | `IsSelfDeployingProvisioned`: self-deploying profile + Device Setup done + no explicit failure | **Succeeded** (reconcile) — no grace, no AwaitingUser: the profile has no user phase |
+| 2 | Account Setup all-succeeded, or `enrollment_complete`, or `whiteglove_complete` on a **non-resumed** session | Succeeded |
+| 3 | `HasAgentEmergencyBreak` | Incomplete immediately (agent self-destructed, nothing can arrive) |
+| 4 | `DesktopArrived` + (Hello terminal, or Hello disabled + SkipUserEsp) | Succeeded (user provably finished) |
+| 5 | Device Setup done | AwaitingUser within grace, Incomplete past grace |
+| 6 | otherwise | Incomplete |
+
+Grace = `ResolveGraceHours(SessionGraceHours, AbsoluteMaxSessionHours)` (~51h default),
+measured from `ResumedAt ?? StartedAt`. Every Succeeded-reconcile reason carries the
+silence-transparency clause (`AppendReconcileTiming`: last agent contact, silence, verdict time).
+
+## Why rule 1c exists (kiosk tenant audit 2026-08-23)
+
+On a self-deploying profile Device ESP all-succeeded IS the enrollment's end; the agent's
+own SelfDeploying terminal only adds a 5-min confirmation window
+(`DecisionEngine.SelfDeploying.s_deviceOnlyEspDetectionWindow`). Tenant `aebdce78` (school,
+Wi-Fi ThinkBooks, a language-change app forcing a reboot right after Device ESP) lost that
+window on 836 of ~970 sessions in one week: the agent died 0–4 min after
+`DeviceSetup fallback_confirmed`, the reboot relaunch never registered (30s registration
+budget on a Wi-Fi link still associating), and rule 5 labeled finished kiosks
+"awaiting user / Account Setup 0/5" → Incomplete 51h later. Nobody signs in on these devices
+(prepared during the holidays, boxed, in service) — AwaitingUser was factually wrong and
+Incomplete turned a working rollout into a red wall. Other self-deploying tenants completed
+normally in the same window, so this is a classification gap, not a profile-wide agent bug.
+Agent-side companion fix: `SessionRegistrationHelper` now waits up to 15s for a network link
+and retries 6× (~62s backoff) instead of 5× (~30s).
+
+## Callers (all pass both gate inputs)
+
+| Path | File | Notes |
+| --- | --- | --- |
+| Sweep stage 1 (2h agent-silent marker) | `MaintenanceService.MarkStalledSessionsAsTimedOutAsync` | Default: InProgress → Stalled. Gates run BEFORE the Stalled write: `TryMarkWhiteGloveAwaitingUserAsync` (AwaitingUser) and `TryReconcileSelfDeployingAsync` (Succeeded). Only flagged sessions pay the event read. |
+| Sweep stage 2 (SessionTimeoutHours, 5h default) | same | InProgress/Stalled/AwaitingUser past the window and past the 2h silence cutoff → full classification; AwaitingUser within grace is skipped without an event read. |
+| Agent max-lifetime (`enrollment_failed`/max_lifetime, `agent_shutting_down`/max_lifetime) | `EventIngestProcessor.ApplyMaxLifetimeVerdictAsync` | Watchdog = "agent stopped waiting", not a verdict; classified from evidence. Pending sessions exempt. |
+| Late-telemetry reconcile | `EventIngestProcessor.TryLateTelemetryReconcileAsync` | Heal-only: applies a Succeeded verdict to Incomplete/AwaitingUser rows when straggler events arrive. |
+| Agent stall probe (`session_stalled`) | `EventIngestProcessor.TryMapStallToWhiteGloveAwaitingUserAsync` | Only the WG gate (1b); a stall probe on a self-deploying device with Device Setup done means the agent is alive and something else blocks — stays Stalled. |
+| Admin retro-reconcile | `LegacyReclassificationService` | Modes `legacy_timeouts`, `pending_orphans`, `self_deploying_silent` (Incomplete/AwaitingUser/Stalled rows with the profile flag, heal-only to Succeeded, admin-marked rows skipped). `POST /api/maintenance/reclassify-legacy?mode=…&dryRun=false&tenantId=…&maxSessions=…`, dry-run by default. |
+
+## Known gap (decided, not built)
+
+After max-lifetime the agent writes `enrollment-complete.marker` and self-destructs, so an
+AwaitingUser verdict from that path can never heal — it is a 51h countdown to Incomplete.
+The emergency-break path (rule 3) already terminalizes immediately; max-lifetime still goes
+through the grace window by design ("user simply absent"). Revisit if customers read
+AwaitingUser as "still possible".
+
+# Examples
+
+* Session `195593e2` (tenant aebdce78): `DeviceSetup 4/4` + `fallback_confirmed`, `AccountSetup 0/5`
+  (IME false positive), Hello disabled, SkipUser=True, silent after 13:01 → rule 1c →
+  Succeeded "self-deploying profile — Device Setup fully provisioned … agent went silent
+  before confirming completion. Agent last reported 2026-08-21 13:01 UTC …".
+* Same event shape on a user-driven profile → rule 5 → AwaitingUser (within grace) /
+  Incomplete (past grace). Guarded by
+  `Classify_same_stream_without_self_deploying_flag_keeps_user_phase_verdicts`.
+* fairstone WG Part 2 reseal-reboot, technician powers off at logon screen → rule 1b →
+  AwaitingUser; user unboxes days later → ingest completion heals to Succeeded.
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Services/EnrollmentTimeoutClassifier.cs` — rollup + rule order, `IsWhiteGloveAwaitingUser`, `IsSelfDeployingProvisioned`, reason helpers.
+* `src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.cs` — `MarkStalledSessionsAsTimedOutAsync`, `TryMarkWhiteGloveAwaitingUserAsync`, `TryReconcileSelfDeployingAsync`.
+* `src/Backend/AutopilotMonitor.Functions/Services/EventIngestProcessor.Classification.cs` — max-lifetime verdict, late-telemetry reconcile, stall mapping.
+* `src/Backend/AutopilotMonitor.Functions/Services/LegacyReclassificationService.cs`, `Functions/Admin/ReclassifyLegacySessionsFunction.cs` — retro modes.
+* `src/Backend/AutopilotMonitor.Functions/Services/TableStorageService.Maintenance.cs` — `GetAgentSilentSessionsAsync`, `GetStalledSessionsAsync`, `GetSelfDeployingSilentSessionsAsync`.
+* `src/Agent/AutopilotMonitor.Agent.V2.Core/Security/SessionRegistrationHelper.cs` — link wait + 6-attempt registration budget.
+* `src/Shared/AutopilotMonitor.DecisionCore/Engine/DecisionEngine.SelfDeploying.cs` — the agent-side 5-min SelfDeploying terminal window.
+* `src/Backend/AutopilotMonitor.Functions.Tests/EnrollmentTimeoutClassifierTests.cs` — rule coverage incl. the 2026-08-23 self-deploying block.

docs/backend/table-schema-sentinel.md

@@ -0,0 +1,67 @@
+---
+type: Concept
+title: Table Schema Sentinel (startup table initialization)
+description: Why a cold start no longer issues one CreateTableIfNotExists per registered table - a hash over TableNames.All stored in AdminConfiguration gates the full pass, the SessionsIndex backfill only runs after a full pass and behind a conditional-insert claim, and daily maintenance repairs out-of-band deletions.
+resource: /src/Backend/AutopilotMonitor.Functions/Services/TableStorageService.cs
+tags:
+  - backend
+  - table-storage
+  - startup
+  - maintenance
+timestamp: 2026-08-23
+---
+
+# Schema
+
+## Problem
+
+`TableInitializerService` (hosted service) called `InitializeTablesAsync` on every process start, which issued `CreateTableIfNotExistsAsync` for every entry of `Constants.TableNames.All` (66 at the time of writing, DoP 8). Each call is a Create POST whose 409 is swallowed - on Flex Consumption with frequent cold starts that was ~9 serial round-trip waves before the first request. It was followed by a SessionsIndex emptiness query whose *error* path returned "empty", so a transient storage error triggered a full Sessions scan with serial upserts - on every scaled-out instance at once.
+
+## Sentinel
+
+* Row `AdminConfiguration` / PK `SchemaSentinel` / RK `tables`, property `TableSchemaHash` (+ `TableCount`, `UpdatedUtc`).
+* Value = `ComputeTableSchemaHash(TableNames.All)`: SHA-256 over the ordinal-sorted, newline-joined table names. Derived, never hand-maintained: adding a table to `TableNames.All` changes the hash by construction (`TableLifecycleBucketTests` already forces every constant into `All`).
+* `InitializeTablesAsync` (returns `bool fullPassRan`):
+  1. Point-read the sentinel. Match -> mark initialized, **zero** table calls, return `false`.
+  2. Missing / stale / unreadable (fresh storage has no AdminConfiguration table yet) -> `EnsureAllTablesAsync` (the old full pass), then upsert the sentinel **only if no table failed**, return `true`. A failed pass leaves `_tablesInitialized = false`, so the next call retries.
+* Scale-out: two instances that both see a stale sentinel both run the idempotent pass - harmless.
+
+## SessionsIndex backfill
+
+* Only evaluated when the full pass ran (`fullPassRan == true`): an empty index can only exist on fresh storage, which is exactly the case without a sentinel.
+* `IsSessionIndexEmptyAsync` now returns `false` on error (skip), not `true`.
+* `TryClaimSessionIndexBackfillAsync`: conditional `AddEntityAsync` on `SchemaSentinel` / `sessionIndexBackfill` - only the winner scans; 409 or any error -> skip. The manual maintenance backfill (`RunManualAsync`) remains the safety net.
+
+## Known gap and repair
+
+A table deleted out-of-band (cleanup skill, portal, tests) is not noticed at startup anymore. `IStorageInitializer.EnsureAllAsync` runs as the first step of the daily timer maintenance (`MaintenanceService.RunAllAsync`), recreating it within 24h. Storage helpers are fail-soft in the meantime.
+
+## Measuring it
+
+`StartupTelemetryService` (hosted service) emits two App Insights metrics on `IHostApplicationLifetime.ApplicationStarted` - i.e. after every hosted service finished `StartAsync`:
+
+* `BackendStartupMs` - process start (`Process.StartTime`) to host ready; includes runtime, DI and all hosted services.
+* `BackendTableInitMs` - the `InitializeTablesAsync` slice alone (`TableStorageService.LastInitialization`).
+
+Dimensions: `version`, `commit`, `tableInitFullPass`, `instance`. Metrics bypass adaptive sampling and the worker log level, so they arrive even though Information traces do not. The MCP server logs the same idea as one `{"type":"boot","readyMs":…}` stderr line per start (ContainerAppConsoleLogs_CL, see `.claude/commands/mcp-telemetry.md`).
+
+```kusto
+customMetrics
+| where name in ("BackendStartupMs", "BackendTableInitMs")
+| summarize p50 = percentile(value, 50), p95 = percentile(value, 95), n = count()
+  by name, version = tostring(customDimensions.version), fullPass = tostring(customDimensions.tableInitFullPass)
+```
+
+# Examples
+
+* Deploy that adds a table: first cold start after the deploy runs the full pass once (log: "Table schema sentinel missing or stale"), every later start logs "Table schema sentinel matches (...)".
+* Fresh storage account: sentinel read throws TableNotFound -> full pass -> index empty -> one instance claims and backfills.
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Services/TableStorageService.cs` - sentinel, `InitializeTablesAsync`, `EnsureAllTablesAsync`, `TryClaimSessionIndexBackfillAsync`
+* `src/Backend/AutopilotMonitor.Functions/Services/TableInitializerService.cs` - startup gating
+* `src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.cs` - daily `EnsureAllTablesAsync`
+* `src/Backend/AutopilotMonitor.Functions/Services/StartupTelemetryService.cs` - `BackendStartupMs` / `BackendTableInitMs`
+* `src/McpServer/autopilot-monitor-mcp/src/index.ts` - `{"type":"boot"}` line
+* `src/Backend/AutopilotMonitor.Functions.Tests/TableSchemaSentinelTests.cs`, `StartupTelemetryServiceTests.cs`

docs/backend/telemetry-ingest-shaping.md

@@ -0,0 +1,89 @@
+---
+type: Concept
+title: Telemetry Ingest Shaping (App Insights / Log Analytics)
+description: What the backend deliberately keeps out of Application Insights and why - one canonical request record (worker copy, per-item sampling bypass; host HTTP duplicate dropped by a workspace transformation), expected storage outcomes (404/412/409) and successful SignalR REST calls filtered in the worker, and the query rules that follow.
+resource: /src/Backend/AutopilotMonitor.Functions/Telemetry/StorageDependencyFilterProcessor.cs
+tags:
+  - backend
+  - telemetry
+  - application-insights
+  - cost
+timestamp: 2026-08-23
+---
+
+# Telemetry Ingest Shaping
+
+The backend is storage-I/O heavy and runs on the Functions isolated worker, which emits telemetry from TWO processes (host and worker). Left alone, Log Analytics ingestion is dominated by rows nobody reads. This document records what is removed, where, and what that means for queries.
+
+# Schema
+
+## Request telemetry — one canonical record
+
+| Source | Emitted by | Content | Fate |
+|---|---|---|---|
+| Host HTTP request (`Properties.Source` empty, `Url` set) | Functions host process | Name, duration, status — no tenant/user/correlation | **Dropped at ingestion** by the workspace transformation on `AppRequests` (`infra/appinsights-workspace-transforms.bicep`) |
+| Host non-HTTP invocation (`Url` empty) | Functions host process | Timer / queue executions | Kept (the worker middleware never sees them) |
+| Worker request (`Properties.Source == 'WorkerMiddleware'`) | `RequestTelemetryMiddleware` | TenantId, UserId, UserRole, CorrelationId, ClientSource, McpToolName, HttpPath | **Kept, unsampled** — the canonical record |
+
+Why two mechanisms:
+
+* The host request item cannot be suppressed from worker code or from `host.json` (`Host.Results` controls log traces, not the request item; verified 2026-06-09). The only lever is a Log Analytics workspace transformation: `source | where tostring(Properties.Source) == 'WorkerMiddleware' or isempty(Url)`.
+* `host.json` `samplingSettings.excludedTypes: "Request;Event"` only applies to the host process. The worker's `AddApplicationInsightsTelemetryWorkerService` runs its own adaptive sampling; before 2026-08-23 worker request rows carried `ItemCount` 2–5 on weekdays (≈30 % of individual requests missing, `count()` undercounting by the same share — `sum(ItemCount)` matched the host copy to 1.00 per day). Re-configuring the worker sampling chain did not take effect in the isolated worker (see the L4 note in `Program.cs`). The per-item opt-out does: the middleware sets `SamplingPercentage = 100` on its own `RequestTelemetry`, and the SDK sampling processor passes through every item whose sampling percentage is already set.
+
+## Dependency telemetry — keep only signal
+
+`StorageDependencyFilterProcessor` (worker pipeline) drops:
+
+* successful Azure Storage calls (Table/Queue/Blob) in both shapes — the HTTP shape (`Azure table`, target `*.table.core.windows.net`) and the Azure SDK ActivitySource shape (`InProc | Microsoft.Tables`, `InProc | Microsoft.Storage`);
+* storage calls with an **expected outcome**: `404` (point-read miss — EventTypeIndex, BlockedDevices lookups), `412` (ETag precondition), `409` (idempotent insert conflict). HTTP-shape rows carry the code in `ResultCode`; InProc rows only in `Properties["Error"]` (`Status: 404 (`). Before this filter these were 100 % of the remaining "failed" storage rows (~540 MB/week, two rows per call);
+* successful Azure SignalR Service REST calls the worker makes itself (`*.service.signalr.net`).
+
+The bulk of SignalR REST traffic (group add/remove per connection, ~470 MB/week) is emitted by the **host** process for the SignalR output binding (`SDKVersion rdddsc:*`) and never passes the worker pipeline — live-verified after the 2026-08-23 deploy (worker rows: 0, host rows: ~560/h). Those are removed by the second data flow of the same workspace transformation: `AppDependencies | where not(Success == true and Target endswith '.service.signalr.net')`.
+
+Everything else passes: storage 429/5xx/auth/timeouts, failed SignalR calls, all HTTP/Graph/NVD/MSRC dependencies, the worker's own `InProc Invoke` span, and every non-dependency item.
+
+## Measured baseline (7 days before the change, `_BilledSize`, `autopilotmonitor-api-eu`)
+
+| Table | Rows | MB | After |
+|---|---|---|---|
+| AppDependencies | 976k | 1248 | ≈ 240 (SignalR + expected-outcome rows gone) |
+| AppRequests host copy | 274k | 323 | ≈ 0 (only timer/queue rows) |
+| AppRequests worker copy | 189k | 208 | ≈ 300 (now 100 % of requests) |
+| AppTraces | 142k | 137 | unchanged |
+| AppEvents | 6k | 4 | unchanged — not worth touching |
+
+Net ≈ −1.3 GB/week with MORE usable data (every request now has tenant context).
+
+# Examples
+
+Counting requests (exact since 2026-08-23; for older data replace `count()` with `sum(itemCount)`):
+
+```kql
+requests
+| where customDimensions.Source == 'WorkerMiddleware'
+| summarize total=count(), failed=countif(success == false) by name
+```
+
+Real storage trouble only:
+
+```kql
+dependencies
+| where type startswith 'InProc | Microsoft.Tables' or target endswith '.table.core.windows.net'
+| summarize count() by resultCode, tostring(customDimensions.Error)
+```
+
+Post-deploy verification of the per-item sampling bypass (must be `ItemCount == 1` only):
+
+```kql
+AppRequests
+| where TimeGenerated > ago(1h) and tostring(Properties.Source) == 'WorkerMiddleware'
+| summarize rows=count() by ItemCount
+```
+
+# Citations
+
+* `src/Backend/AutopilotMonitor.Functions/Middleware/RequestTelemetryMiddleware.cs` — `SamplingPercentage = 100` before `TrackRequest`.
+* `src/Backend/AutopilotMonitor.Functions/Telemetry/StorageDependencyFilterProcessor.cs` + `StorageDependencyFilterProcessorTests.cs` — the filter contract.
+* `src/Backend/AutopilotMonitor.Functions/Program.cs` — the failed L4 worker-sampling attempt and why it must not be retried.
+* `infra/appinsights-workspace-transforms.bicep` — the AppRequests transformation DCR, apply/verify/rollback commands.
+* `.claude/commands/backend-logs.md`, `session-debug.md` — query conventions.

docs/backend/time-attribution.md

@@ -105,6 +105,11 @@ the partition).
 * 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.
+  Reads the tick's **shared window scan** (`MaintenanceService.LoadSweepWindowSessionsAsync`,
+  one projected 35-day cross-tenant drain via `MaintenanceSweepSessionProjection` that
+  also feeds the device-journey and verdict-calibration sweeps) — the StartedAt-only
+  filter is a full-table read in Table Storage, so it is done once per tick, not once per
+  sweep; `MaintenanceSweepProjectionEquivalenceTests` pins the column set.
 * `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

docs/backend/verdict-calibration.md

@@ -0,0 +1,164 @@
+---
+type: Concept
+title: Verdict Calibration (classifier thermometer)
+description: How every session status write declares its VerdictPath, how overriding a verdict preserves it as PriorStatus/PriorVerdictPath, how the daily calibration aggregate counts paths with the re-enrollment proxy and the correction stream, and how the drift radar (share / silence / evidence-gap) turns that into operator-only ops events.
+resource: src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.VerdictCalibration.cs
+tags:
+  - backend
+  - sessions
+  - classifier
+  - calibration
+  - regression-radar
+timestamp: 2026-08-23T18:00:00+02:00
+---
+
+# Verdict Calibration (classifier thermometer)
+
+A session's terminal status is written by ~20 code paths: the agent's own terminal events,
+the silence classifier (`EnrollmentTimeoutClassifier`, rules 1–6) reached via the
+maintenance sweep / the agent's max-lifetime shutdown / late-telemetry heal / retro
+reclassification, the analyze-rule engine, admin marks and session superseding. Verdict
+calibration makes every one of those paths **countable**, tracks how often a path's verdict
+was later **overridden**, attaches the delayed **re-enrollment proxy**, and raises an
+operator-only alarm when a path's share **drifts**. It is deterministic, fact-based tuning
+input for the rules — explicitly not a learned model, and never a customer-facing feature.
+
+# Schema
+
+## VerdictPath on the session row
+
+`SessionSummary.VerdictPath` (`VerdictPaths` vocabulary, `origin:detail`, append-only) is
+stamped by **every** status write — `UpdateSessionStatusAsync` takes it as a required
+parameter, `StoreSessionAsync` stamps `register:new` / `register:whiteglove_resume` and
+preserves an existing path through the re-registration Replace, `SetSessionPreProvisionedAsync`
+requires it whenever it writes a status. Unlike `FailureSource` (persisted only on `Failed`),
+the path survives for Incomplete, AwaitingUser, Stalled and Pending.
+
+| Origin | Paths |
+| --- | --- |
+| `agent` | `complete`, `complete_soft` (Continue-Anyway), `failed`, `esp_failure_fallback`, `gather_complete`, `whiteglove_pending`, `whiteglove_resumed`, `stall_probe`, `stall_heal` |
+| `ingest` | `wg_awaiting` (stall probe mapped to WhiteGlove AwaitingUser) |
+| `sweep` | `stalled`, `wg_awaiting`, `sd_reconcile`, `r{rule}` (stage-2 classifier) |
+| `maxlife` / `late` / `retro` | `r{rule}` (classifier via max-lifetime shutdown / late-telemetry heal / retro reclassification); `retro:sd_reconcile`, `retro:superseded` |
+| `register` | `new`, `whiteglove_resume`, `superseded` |
+| `rule` | `rule:{ruleId}` (analyze rule with MarkSessionAsFailed) |
+| `manual` | `failed`, `succeeded` |
+| `legacy` | read-side derivation only (`legacy:{rule}` when the rule is recoverable from the reason literal, `legacy:unknown` otherwise) — never written |
+
+Classifier rule ids (`ClassifierRules`): `r1`, `r1b_awaiting`, `r1b_succeeded`, `r1c`, `r2`,
+`r3`, `r4`, `r5_awaiting`, `r5_incomplete`, `r6`. `ClassifyTimedOutSession` returns
+`(Status, Reason, Rule)` so each `return` names its rule.
+
+## PriorStatus / PriorVerdictPath — the correction stream
+
+When a write replaces a prior **verdict** (existing status ∈ Succeeded/Failed/Incomplete/
+AwaitingUser with a stamped path, incoming status different), the storage seam preserves the
+pair (`TableStorageService.ComputePriorVerdict`, both the ETag path and the force path). A
+stall marker healing to InProgress, a same-status refresh, or an unstamped legacy row never
+sets it. Admin marks, late agent completions and retro reclassifications all flow through it.
+
+## Read-side derivation for pre-instrumentation rows
+
+`VerdictPathDerivation.Derive(session)` returns the stamped path verbatim, otherwise derives
+from `AdminMarkedAction`, `FailureSource`, `ReconcileReason` and `FailureReason` using the
+writers' fixed literals (classifier rule prefixes, the `max-lifetime watchdog` / `Retro-
+reclassified` suffixes, `Rule: `, `Superseded by session`, `Late completion report received`).
+Unambiguous rows get the real path (`agent:complete`, `manual:failed`, `rule:…`,
+`maxlife:r6`), ambiguous-but-rule-known rows `legacy:{rule}`, the rest `legacy:unknown`. No
+mass backfill: the aggregate counts derived rows with `DerivedCount` so the matrix shows
+history while flagging its weaker evidence. Tests pin the literals against the classifier
+output so a wording change fails loudly.
+
+## Daily aggregate — `VerdictCalibrationAggregates`
+
+PK = `TenantId` | `"global"`, RK = `yyyy-MM-dd` (session **StartedAt** date, like every daily
+aggregate). `VerdictCalibrationDailyAggregate { Version, SessionCount, TerminalSessionCount,
+Buckets[], ComputedAt }` with one bucket per (VerdictPath, Status):
+
+| Field | Meaning |
+| --- | --- |
+| `Count` / `DerivedCount` | sessions currently on the path; of those, derived read-side |
+| `Eligible7d` | terminal sessions whose end (`CompletedAt ?? LastEventAt ?? StartedAt`) lies ≥ 7 d before compute time |
+| `ReEnrolled7d` | of the eligible: the device's `DeviceHistories` chain shows the next terminal session starting < 7 d after this one's end |
+| `OverriddenByAdmin` / `OverriddenByLateCompletion` / `OverriddenOther` | sessions whose **PriorVerdictPath** is this bucket — attributed to the path that was overridden, so a bucket can carry overrides with `Count = 0` |
+
+`MaintenanceService.SweepVerdictCalibrationAsync` (2-hourly, after the device-journey sweep
+so the chains are fresh; manual maintenance parity; kill switch
+`VerdictCalibrationSweepDisabled`) slices the tick's shared projected window scan
+(`LoadSweepWindowSessionsAsync` — one 35-day cross-tenant drain feeds the time-attribution,
+device-journey and calibration sweeps plus the radar's tenant discovery; the StartedAt-only
+filter is a full-table read, so never per sweep), recomputes the rolling 30-day window whole (Replace),
+deletes stale date rows it did not regenerate, and the unbounded-tables cleanup applies the
+180-day retention. Sessions inside a deletion cascade are skipped; the table is regenerable
+and is wiped with the tenant on offboarding.
+
+## Read surfaces
+
+* `GET global/metrics/verdict-calibration?days=&tenantId=` (GlobalReadOrAdmin,
+  TenantScoping.QueryParam; `VerdictCalibrationResponse.Build`): per path — count, share of
+  all window sessions, derived count, eligible/re-enrolled with `reEnrollRatePct` **null below
+  20 eligible**, the three override counters, a today-anchored 7d-vs-28d share trend
+  (`window7`, `baseline28`, `lift` null without baseline) — plus totals, the trend basis and
+  the active `alerts[]`. Empty but well-formed before the first sweep.
+* MCP `get_verdict_calibration` (platform scope: GA + Global Reader). Same endpoint, LLM-shaped in
+  `verdict-calibration-shape.ts`: per-row `window7.sessions` / `baseline28.sessions` are dropped (the
+  denominators live once in `trend`), `reEnrollRatePct` / `lift` are always present (`null` = withheld),
+  and `minSharePct` / `top` trim the one-session tail — rows carrying overrides are never trimmed and
+  everything dropped is reported in `omitted`. The wire shape the admin page consumes is untouched.
+* Portal: Global Admin → Metrics → Verdict Calibration (`/admin/metrics/verdict-calibration`):
+  matrix grouped by origin, status pill, `derived n` gray pill, trend glyph (arrow only from
+  5 window hits and lift ≥ 2 / ≤ 0.5), re-enrollment rate or `— (n=…)`, overrides
+  `admin / late / other`, `↑ Drift` pill on rows with an active episode, and the active
+  episodes list.
+
+## Drift radar — `VerdictCalibrationRadar`
+
+Same statistics as the [rule regression radar](rule-regression-radar.md) (7-day window vs the
+prior 28 days, ≥ 5 window hits, ≥ 20 window sessions, lift ≥ 2, `MetricsMath.
+RateIncreaseSeparated`), anchored on yesterday, over every partition with rows (each tenant
+and `global`):
+
+| Kind | Signal | Re-arm |
+| --- | --- | --- |
+| `share_regression` | one **backend-decided** path's share of all sessions doubled (`sweep`/`maxlife`/`late`/`retro`/`rule`/`manual`/`ingest`; `agent:*` and `register:*` mirror customer workflow mix — first prod pass fired on `agent:whiteglove_pending` rollout weeks — and `legacy:*` is derived, all excluded) | share < 1.5× baseline, the path stops occurring, or the path is no longer eligible |
+| `silence_share_regression` | the `sweep:*` + `maxlife:*` share doubled — the backend had to decide more often because the agent went silent (a liveness signal, not a classifier one) | same |
+| `evidence_gap` | absolute: `r6` (pure fallthrough) decides ≥ 20 % of the window's classifier verdicts, ≥ 20 verdicts | < 15 % or < 20 verdicts |
+
+Episodes are tracker rows `verdictcalibration|{kind}|{path}|{status}` (register = `AddEntity`,
+409 = already burning, fail-closed; refresh carries `FirstNotifiedAt`; 30-day tracker
+retention re-arms long burners). A new episode fires **one** `VerdictCalibrationDrift` ops
+event (category Maintenance, Warning; dual-registered in the web `OPS_EVENT_TYPES`) — there
+is deliberately **no tenant bell**. Per-path findings on a tenant partition get the on-fire
+dimension correlation (`RuleRegressionRadar.ComputeDimensionConcentration` over the window's
+sessions whose derived path matches), reported with the shared wording contract "correlated,
+not necessarily causal"; group kinds and the global partition carry no dimension claim.
+Kill switch `VerdictCalibrationRadarDisabled`.
+
+# Examples
+
+* **Reading the thermometer (prod, 2026-08-23 hand analysis that motivated this):**
+  re-enrollment within 7 days after Succeeded 8.9 % (lab background), after Failed 24.3 %,
+  after Incomplete 10.6 % — Incomplete behaves like Succeeded, so rules 5/6 are probably too
+  cautious; the matrix now shows this per rule instead of per status.
+* **Correction stream:** a session swept to `sweep:r5_incomplete`, later completed by the
+  agent → row `agent:complete` with `PriorVerdictPath = sweep:r5_incomplete`; the aggregate
+  increments `OverriddenByLateCompletion` on the `sweep:r5_incomplete / Incomplete` bucket.
+  Four admin marks since launch were invisible before — now each one counts against the path
+  it corrected.
+* **Evidence gap:** a tenant whose agent build stopped emitting the Device Setup rollup line
+  drives its sweep verdicts into `r6`; the absolute gate fires at 20 % and names the
+  tenant — the fix is agent-side, the classifier is only the messenger.
+
+# Citations
+
+* `src/Shared/AutopilotMonitor.Shared/Models/VerdictPaths.cs` — vocabulary + `ClassifierRules`
+* `src/Backend/AutopilotMonitor.Functions/Services/TableStorageService.Sessions.cs` — `ComputePriorVerdict`, `ApplyVerdictPath`, re-registration preserve
+* `src/Backend/AutopilotMonitor.Functions/Helpers/VerdictPathDerivation.cs`
+* `src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.VerdictCalibration.cs` — sweep + `BuildVerdictCalibrationAggregates`
+* `src/Backend/AutopilotMonitor.Functions/Functions/Metrics/GetVerdictCalibrationFunction.cs`
+* `src/Backend/AutopilotMonitor.Functions/Helpers/VerdictCalibrationRadar.cs`, `Services/MaintenanceService.VerdictCalibrationRadar.cs`
+* `src/McpServer/autopilot-monitor-mcp/src/tools/admin.ts` — `get_verdict_calibration`
+* `src/McpServer/autopilot-monitor-mcp/src/verdict-calibration-shape.ts` — LLM-facing response shaping (`minSharePct` / `top`, explicit nulls)
+* `src/Web/autopilot-monitor-web/app/admin/metrics/sections/SectionVerdictCalibration.tsx`
+* Tests: `VerdictPathTests`, `VerdictCalibrationTests`, `VerdictCalibrationRadarTests`
+* Related: [Silent-Session Classification](silent-session-classification.md), [Device Journeys](device-journeys.md), [Rule Regression Radar](rule-regression-radar.md)

docs/index.md

@@ -25,6 +25,7 @@ https://docs.autopilotmonitor.com (separate repository).
 * [Gather Rule Debug Log](agent/gather-rule-debug-log.md) - EnableGatherRuleDebugLog (ConfigVersion 36): the per-tenant local trace file that explains every gather-rule evaluation decision — scope skips, on_change suppression, empty collector results, logparser position/parse-mode outcomes — for rules that produce no timeline events.
 * [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.
 * [ESP Completion Starvation — the sits-d Cloud-PC complex](agent/post-reboot-completion-recovery.md) - Why a SkipUser=true flow (no user ESP page → no Shell-Core 62407, no AccountSetup registry) left the completion gate satisfied but never evaluated — five Cloud PCs idled six hours at "waiting on: hello_resolution" — and the fix: the observed skip stands in for the final-exit evidence in the IME user-session completion attempt. Plus the two adjacent defects fixed alongside: the edge-only user-apps-settled synthesis (now level-triggered) and the never-wired Shell-Core restart replay (now Hello-wizard-only, with the reasoning why a replayed 62407 must never reach the reducer).
+* [Zero User-Targeted Apps — the starved user-phase completion gate](agent/zero-user-apps-completion-gate.md) - Why a device with no user-targeted Intune apps could never satisfy the IME user-session gate — IME skips its "Completed user session" line entirely in the zero-app branch — turning the 30-minute completion backstop into `esp_exit_without_completion_evidence` on a third of one tenant's enrollments, with the user at the desktop and RealmJoin still installing. The two fixes: read the line IME does write (IME-USER-SESSION-ZERO-APPS), and never fail a session while a completion gate with its own armed deadline is deliberately holding it.
 * [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.
 * [Continue-Anyway Observation Mode](agent/continue-anyway-observation.md) - EnableEspContinueAnywayObservation (ConfigVersion 37, operator-set): Device-phase ESP terminal failures on Continue-Anyway profiles get a bounded 60-min observation instead of an immediate hard fail; the DAD-validated real-user desktop + Hello gate detect the true end, the session completes Succeeded + EspSoftFailure (amber "with issues" badge), expiry un-defangs to the original esp_terminal_failure.
 * [CMTrace Time Resolution](agent/cmtrace-time-resolution.md) - Converting bias-less CMTrace local times to UTC without trusting any process's timezone belief: per-line self-anchoring on the 15-minute offset grid for provably fresh lines, uniform marked reader-zone fallback for backlog, and the provenance vocabulary (bias / line-anchored / reader-zone-fallback) that keeps every event re-derivable.
@@ -45,13 +46,17 @@ https://docs.autopilotmonitor.com (separate repository).
 
 # Backend
 
+* [Telemetry Ingest Shaping](backend/telemetry-ingest-shaping.md) - One canonical App Insights request record (worker copy with tenant context, per-item sampling bypass; host HTTP duplicate dropped by a workspace transformation DCR), expected storage outcomes 404/412/409 and successful SignalR REST calls filtered in the worker, and the count()/sum(itemCount) query rules that follow.
+* [Table Schema Sentinel](backend/table-schema-sentinel.md) - Startup no longer issues one CreateTableIfNotExists per table: a SHA-256 over TableNames.All in AdminConfiguration gates the full pass (derived, never hand-maintained), the SessionsIndex backfill runs only after a full pass behind a conditional-insert claim, and daily maintenance repairs out-of-band deletions.
 * [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.
 * [Send-Time Frame Separation](backend/send-time-frame-separation.md) - One device-clock send timestamp per upload batch (X-Send-Time-Utc → SentAt on Events) splits ReceivedAt − OccurredUtc into pure spool delay and pure device-vs-server clock offset; the clock_skew condition measures SentAt batches directly (spool-immune), the CMTrace tripwire deliberately stays on its differential math.
 * [OOBE Bootstrap go-Code Flow](backend/oobe-bootstrap-go-flow.md) - The bootstrap short-code lifecycle (portal-created session → `irm go.autopilotmonitor.com/CODE | iex` at Shift+F10 → GetBootstrapScriptFunction), the always-HTTP-200 text/plain contract, the two-line injection defense (GUID/URL/expiry validator + single-quoted PS literals, golden-pinned against the legacy TS template), and the per-IP rate-limit story behind Front Door.
+* [Silent-Session Classification](backend/silent-session-classification.md) - How a session whose agent went silent gets its honest status: the two-stage maintenance sweep (2h Stalled marker with the WhiteGlove Part-2 AwaitingUser gate and the self-deploying Succeeded gate, 5h timeout classification), the EnrollmentTimeoutClassifier rule order (explicit failure → WG Part 2 → self-deploying → completion evidence → emergency break → desktop+Hello → Device Setup/grace), the ingest paths sharing it (max-lifetime, late-telemetry heal, stall probe) and the admin retro-reconcile modes incl. self_deploying_silent.
 * [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.
 * [Attempt-Scoped App Install Durations](backend/app-install-duration-attempt.md) - AppInstallSummary.DurationSeconds = status-defining (last) install attempt via the LastAttemptStartedAt max-fold anchor instead of the first-observation→last-terminal span that billed multi-pass IME gaps to trivial apps; InstallPassCount, the weaker-terminal guard, the radar's cutover gate, and time attribution's union-of-attempt-segments occupancy.
 * [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.
+* [Verdict Calibration](backend/verdict-calibration.md) - How every session status write declares its VerdictPath (origin:detail vocabulary incl. classifier rule ids), how overriding a verdict preserves PriorStatus/PriorVerdictPath as the correction stream, the daily per-path calibration aggregate with the 7-day re-enrollment proxy and read-side legacy derivation, the GA endpoint / MCP tool / admin matrix, and the drift radar (share, silence-share, r6 evidence gap) with its operator-only ops events.
 * [App Version Duration Regression Radar](backend/app-version-regression-radar.md) - How app versions whose median install duration regressed vs. the previous version are detected (first-seen-ordered versions, median lift ≥2× plus a 5-minute absolute floor over measured installs only), how alert episodes live in the appversionregression tracker keyspace with drain/median re-arm semantics, and the bell/ops-event/analytics surfaces.
 * [Plan Downgrade Policy & the Retention Grace Period](backend/plan-downgrade-retention-grace.md) - What happens when a tenant loses Pro: every entitlement gates read-time and reversibly ("gate, don't destroy"), except retention — a 30-day grace keeps the Pro cap before the Community cap starts hard-deleting; the ProDowngradedUtc/trial-expiry anchor model, the TenantPlanDowngraded/RetentionGrace* ops events, and the deliberate non-grace of everything else.
 * [Tenant Activation & the Once-Only Welcome Mail](backend/tenant-activation-welcome-mail.md) - How a signup becomes an activated tenant (manual approve or the auto-approve queue), and the write-then-read + conditional-marker arbitration that sends the welcome mail exactly once even though activation and the user-typed notification address race each other.
@@ -72,6 +77,7 @@ https://docs.autopilotmonitor.com (separate repository).
 # 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.
+* [Client-Certificate Tenant Binding](backend/client-cert-tenant-binding.md) - Why a chain-valid Intune MDM certificate only proves "issued to some tenant", which extension actually carries the Entra tenant id (`…5.14`) versus the two that look like it (`…5.6` Intune Account id, `…5.4` device id, different encodings), why the Entra join certificate is not part of the auth path, and the shadow-then-enforce rollout with its volume-controlled telemetry.
 
 # Conventions for this bundle
 

docs/log.md

@@ -1,7 +1,30 @@
 # Log
 
+## 2026-08-25
+
+* **New**: `agent/zero-user-apps-completion-gate.md` — a Classic enrollment whose user has zero targeted Intune Win32 apps could not produce the evidence any user-phase completion path required: IME's zero-app branch (`ApplicationPoller`, verified against decompiled 1.104.102.0 and three older builds) `continue`s before its `Completed user session` statement, so `ImeUserSessionCompletedUtc` stayed null, all four arms of `ShouldTransitionToAwaitingHello` were dead, and the 30-min `advisory_completion` backstop resolved the session as `esp_exit_without_completion_evidence` — with a DAD-validated desktop, Hello provisioned and RealmJoin still installing. One tenant: 10 of 30 enrollments in a week, all false positives. Two fixes. (1) New pattern `IME-USER-SESSION-ZERO-APPS` reads the line IME *does* write and routes it through the same `OnUserSessionCompleted` callback, guarded to the AccountSetup phase and replayed if the observation predates the phase marker; `AreUserEspAppsSettled` now accepts an empty app list only with that explicit observation. (2) `HandleAdvisoryCompletionDeadlineFired` re-arms instead of failing while the RealmJoin gate is closed and its own deadline is still armed — session 75d6ae8e installed 25 RealmJoin packages inside the window, the last 2 min before the failure. `ime_user_session_completed` gains an `evidence` field (`zero_user_apps_enumerated` / `user_session_completed`); no new event type. The pattern reaches existing agents via the config API (an unknown action is a Debug line), the behaviour needs an agent release.
+
+## 2026-08-24
+
+* **Change**: `DeviceAssociationValidator` / `SecurityValidator` — the DevPrep association shadow now logs one line per real Graph lookup instead of one per request (`DeviceAssociationResult.ServedFromCache`, `AsCacheHit()` copies rather than marking the shared cached instance), the same gate the cert-device-binding shadow uses. Measured: one enrollment produced 263 identical lines off a handful of calls. Status of that preview, for the record: the `/beta/deviceManagement/tenantAssociatedDevices` endpoint answers (6 calls, all HTTP 200, `transient=False`), but every result so far is `matched=False` from a single Autopilot **v1** device — which is the correct answer for a device that was never part of a Device Preparation policy. The matching itself is therefore still unproven, and the endpoint remains undocumented in both v1.0 and beta Graph reference, so it is not a basis for enforcement.
+
+* **Update**: `backend/client-cert-tenant-binding.md` — stage 3 verified live (`ea6a375`, gktatooine): 136 requests, all `Match`, device object ~72 s older than the session start, so no enrollment race in that run. Measured Graph cost: **1 call for 136 requests** (30 min positive cache), versus 25 for the per-request Autopilot validator — the positive TTL needs no change. The shadow log is now gated on `ServedFromCache` (one line per real lookup instead of per request); never-cached outcomes still log every time.
+
+* **Update**: `backend/client-cert-tenant-binding.md` — stage 3, the cert-to-device binding: new `IntuneDeviceBindingValidator` resolves the Subject-CN managedDevice id against the tenant's own `deviceManagement/managedDevices` (point-GET, no filter literal), modelled on `CloudPcDeviceValidator` and wired fire-and-forget like the DevPrep shadow so a cold Graph round-trip never enters the agent request path. New GA-only tenant toggle `ValidateIntuneDeviceBinding` (shadow — never touches `deviceValidated`), new optional Graph feature `IntuneDeviceBinding` → `DeviceManagementManagedDevices.Read.All` in `GraphFeatureCatalog` + the customer grant script, and `enrolledAgeSeconds` on every log line so the enrollment race can be measured before this could ever become a gate. Grep marker: `CERT-DEVICE-BINDING-SHADOW`.
+
+* **Update**: `backend/client-cert-tenant-binding.md` — shadow telemetry moved off the dead `LogInformation` path: the `Match` outcome now rides on the request row as the `CertTenantBinding` dimension (`SecurityValidator` → `FunctionContext.Items` → `RequestTelemetryMiddleware`), which is unsampled and already exists per request, so the denominator costs nothing. Non-Match outcomes keep their Warning trace. Post-deploy verification of 249e39bc found zero rows across 32k requests — worker `LogInformation` never reaches App Insights (provider default rule Warning+), making "everything matched" indistinguishable from "never ran". `CertificateValidationResult.FromCache` removed with the sampling path it existed for.
+* **New**: `backend/client-cert-tenant-binding.md` — the agent's mTLS certificate chains to Microsoft's shared Intune roots, so certificate validation alone never proved tenant membership; tenant scoping rested entirely on the (non-secret) device serial in the downstream validators. The Intune MDM leaf carries the Entra tenant id in OID `1.2.840.113556.5.14` (nested OCTET STRING), which is NOT `…5.6` (Intune Account id, a different GUID) and NOT `…5.4` (device id, stored as 16 bare bytes) — all decoded by one shared `MsDeviceCertificateOids.TryParseGuid` now used by agent and backend. Stage 1 ships as shadow only: `CertTenantBinding.Evaluate` (pure) plus `SecurityValidator.ObserveCertTenantBinding` (returns void, swallows its own exceptions, cannot change an authorization outcome), with `Match` sampled via `CertificateValidationResult.FromCache` so the denominator survives without per-request volume. Grep marker for the stage-2 enforcement change: `CERT-TENANT-BINDING-SHADOW`.
+
 ## 2026-08-23
 
+* **New**: `backend/telemetry-ingest-shaping.md` — `RequestTelemetryMiddleware` sets `SamplingPercentage=100` on its own item (worker request rows were adaptively sampled, ItemCount 2–5 on weekdays, despite host.json `excludedTypes`), the redundant host HTTP request copy is dropped by a workspace transformation DCR (`infra/appinsights-workspace-transforms.bicep`, timer/queue rows kept), and `StorageDependencyFilterProcessor` now also drops expected storage outcomes (404/412/409, both InProc and HTTP shapes) and successful SignalR REST calls; the host-emitted SignalR binding calls (bulk of the volume, SDK `rdddsc`) are dropped by a second `AppDependencies` flow in the same DCR. Measured ≈ −1.3 GB/week.
+* **Update**: `backend/table-schema-sentinel.md` — "Measuring it": `StartupTelemetryService` emits `BackendStartupMs` (process start → ApplicationStarted) and `BackendTableInitMs` as App Insights custom metrics with version/commit/fullPass dimensions; the MCP server writes one `{"type":"boot","readyMs"}` stderr line per start (query in `.claude/commands/mcp-telemetry.md` §A2).
+* **New**: `backend/table-schema-sentinel.md` — startup table initialization gated by a schema sentinel (SHA-256 over `TableNames.All` in `AdminConfiguration`/`SchemaSentinel`/`tables`): matching sentinel ⇒ zero `CreateTableIfNotExists` calls on a cold start, missing/stale ⇒ one full pass + sentinel rewrite. SessionsIndex startup backfill only after a full pass, `IsSessionIndexEmptyAsync` no longer treats errors as "empty", conditional-insert claim so one scaled-out instance backfills. Daily maintenance runs `IStorageInitializer.EnsureAllAsync` to repair out-of-band table deletions.
+* **Update**: `backend/verdict-calibration.md` — MCP `get_verdict_calibration` response shaping: per-row trend denominators dropped, explicit nulls, `minSharePct` / `top` with an `omitted` report.
+* **Update**: `backend/verdict-calibration.md` — share-regression radar restricted to backend-decided paths; agent-declared/registration paths (customer workflow mix) and derived legacy paths no longer alert, stale episodes re-arm themselves.
+* **Update**: `backend/time-attribution.md`, `backend/device-journeys.md`, `backend/verdict-calibration.md` — the rolling maintenance sweeps (time attribution, device journeys, verdict calibration, calibration-radar discovery) now share ONE projected 35-day cross-tenant Sessions scan per tick (`LoadSweepWindowSessionsAsync` + `MaintenanceSweepSessionProjection`, pinned by `MaintenanceSweepProjectionEquivalenceTests`) instead of four full-table drains; each sweep slices its own window by StartedAt.
+* **New**: `backend/verdict-calibration.md` — the classifier thermometer: every status write now stamps `VerdictPath` (required parameter on `UpdateSessionStatusAsync`; `EnrollmentTimeoutClassifier` returns its rule id), overriding a verdict preserves `PriorStatus`/`PriorVerdictPath`, `VerdictPathDerivation` attributes pre-instrumentation rows read-side (`legacy:*` when ambiguous). New daily aggregate `VerdictCalibrationAggregates` (per-path count/derived/eligible7d/reEnrolled7d/overridden-by-admin|late|other, sweep after the device-journey pass, 180d retention), `GET global/metrics/verdict-calibration`, MCP `get_verdict_calibration`, admin metrics section, and the `VerdictCalibrationRadar` (share / silence-share / r6 evidence-gap episodes in the tracker keyspace `verdictcalibration|`, `VerdictCalibrationDrift` ops event, no tenant bell).
+* **New**: `backend/silent-session-classification.md` — one place for the silent-session verdict: `EnrollmentTimeoutClassifier` rule order incl. the new rule 1c `IsSelfDeployingProvisioned` (self-deploying profile + Device Setup all-succeeded + no explicit failure ⇒ Succeeded reconcile, no grace/AwaitingUser — kiosk tenant aebdce78 audit: 836 finished devices parked AwaitingUser/Incomplete), the sweep stage-1 gates (`TryMarkWhiteGloveAwaitingUserAsync`, new `TryReconcileSelfDeployingAsync`), the ingest callers (max-lifetime, late-telemetry heal, stall probe) and the new admin retro mode `self_deploying_silent` (`GetSelfDeployingSilentSessionsAsync`). Agent companion: `SessionRegistrationHelper` waits up to 15s for a network link and retries 6× (~62s) instead of 5× (~30s) so the reboot relaunch on Wi-Fi no longer exits 7 before the link is back.
 * **New**: `agent/diagnostics-package.md` — the diagnostics ZIP's built-in collection is now ONE catalog in Shared (`DiagnosticsBuiltInSections.All`): the agent iterates it in order and the backend serves the same list to the portal (`GET diagnostics/paths`, MemberRead), so what administrators see is exactly what the agent collects. New conditional sections: RealmJoin client/package/Chocolatey/per-user logs behind the tenant's RealmJoin Watcher toggle (mirrored onto `AgentConfiguration.EnableRealmJoinWatcher` by the merger) and the Device Preparation bootstrapper event log behind the deterministic WDP marker (probe once per build, fail toward Classic); every skip is a manifest line. Built-ins bypass `DiagnosticsPathGuards` by design (reviewed code, `%ProgramData%\RealmJoin` is not allowlisted); caps/reparse/evtx export still apply. `MergeDiagnosticsLogPaths` dedups global ∪ tenant paths. Portal: collapsed read-only catalog + one-line rows in both cards, tenant context dropped the GA-only global fetch.
 * **Update**: `rules/gather-rule-guardrails.md` — `DiagnosticsPathGuards` scope clarified: configured diagnostics paths only, never the built-in catalog.
 * **Update**: `backend/tenant-activation-welcome-mail.md` — new "Template overrides" section: Global Admins can replace the welcome/farewell HTML in the portal (Settings → Email Templates); `EmailTemplateService` resolves override-or-built-in per send (PreviewConfig table, partition `EmailTemplates`, 5-min cache, `{{domainName}}` placeholder, 30k limit, fail-soft send path), GA endpoints `global/email-templates/{kind}` (+`/test` to the caller's tenant contact address). Also: the mail header is image-free (CSS-drawn brand mark, gradient stripes and clip-path edge as progressive enhancement).

infra/appinsights-workspace-transforms.bicep

@@ -0,0 +1,78 @@
+// Autopilot-Monitor — Log Analytics workspace transformation (ingestion-time filter)
+//
+// Two ingestion-time filters. (2) Successful Azure SignalR REST dependencies emitted by the
+// HOST process for the SignalR output binding (SDKVersion rdddsc:*) — the worker-side
+// StorageDependencyFilterProcessor never sees those. Failed SignalR calls are kept.
+// (1) Drops the redundant HOST-side HTTP request record from AppRequests so that exactly one
+// canonical request row remains: the worker-side copy emitted by RequestTelemetryMiddleware
+// (Properties.Source == 'WorkerMiddleware', carries TenantId/UserId/CorrelationId and is
+// sampling-exempt per item). Non-HTTP host invocations (timer/queue triggers) have an empty
+// Url and are kept — the worker middleware never sees them.
+//
+// Why ingestion-time: the host request item is emitted by the Functions host process and
+// cannot be suppressed from worker code or host.json (verified 2026-06-09, see Program.cs).
+// Live numbers 2026-08-23 (7d): host copy 274k rows / 323 MB, worker copy 189k rows / 208 MB.
+//
+// Cost note: transformation filtering is free up to 50% of the table's incoming volume; the
+// portion beyond that (here ~11% of AppRequests) is billed at the (cheap) data-processing rate.
+//
+// Apply (two steps; the link is a CLI update = GET+PUT, which keeps retention/SKU intact):
+//   az deployment group create -g rg-autopilotmonitor-prd-gwc \
+//     --template-file infra/appinsights-workspace-transforms.bicep \
+//     --query properties.outputs.dcrId.value -o tsv
+//   az monitor log-analytics workspace update -g rg-autopilotmonitor-prd-gwc -n AutopilotMonitor \
+//     --data-collection-rule <dcrId-from-above>
+//
+// Verify after ~20 min (ingestion latency):
+//   AppRequests | where TimeGenerated > ago(15m) | where AppRoleName == 'autopilotmonitor-api-eu'
+//   | summarize rows=count() by source=tostring(Properties.Source), hasUrl=isnotempty(Url)
+//   → only (WorkerMiddleware, true) and ('', false) buckets remain.
+//
+// Rollback: az monitor log-analytics workspace update -g <rg> -n AutopilotMonitor
+//   --set properties.defaultDataCollectionRuleResourceId=null   (then delete the DCR)
+
+@description('Log Analytics workspace that backs the workspace-based Application Insights resources.')
+param workspaceName string = 'AutopilotMonitor'
+
+param location string = resourceGroup().location
+
+resource workspace 'Microsoft.OperationalInsights/workspaces@2023-09-01' existing = {
+  name: workspaceName
+}
+
+resource transforms 'Microsoft.Insights/dataCollectionRules@2023-03-11' = {
+  name: 'dcr-autopilotmonitor-ws-transforms'
+  location: location
+  kind: 'WorkspaceTransforms'
+  properties: {
+    description: 'AppRequests: keep the enriched worker request record (Properties.Source == WorkerMiddleware) and non-HTTP host invocations (empty Url); drop the host HTTP duplicate.'
+    destinations: {
+      logAnalytics: [
+        {
+          workspaceResourceId: workspace.id
+          name: 'ws'
+        }
+      ]
+    }
+    dataFlows: [
+      {
+        streams: [ 'Microsoft-Table-AppRequests' ]
+        destinations: [ 'ws' ]
+        transformKql: 'source | where tostring(Properties.Source) == \'WorkerMiddleware\' or isempty(Url)'
+      }
+      {
+        // Successful Azure SignalR REST calls (group add/remove per connection) are emitted by
+        // the HOST process for the SignalR output binding (SDKVersion rdddsc:*), so the worker's
+        // StorageDependencyFilterProcessor never sees them. Failed SignalR calls are kept.
+        streams: [ 'Microsoft-Table-AppDependencies' ]
+        destinations: [ 'ws' ]
+        transformKql: 'source | where not(Success == true and Target endswith \'.service.signalr.net\')'
+      }
+    ]
+  }
+}
+
+// A workspace has exactly one transformation DCR, activated by linking it as the workspace's
+// defaultDataCollectionRuleResourceId — deliberately NOT modelled here (see Apply above).
+
+output dcrId string = transforms.id

rules/dist/ime-log-patterns.json

@@ -675,6 +675,14 @@
       "description": "Detects when IME completes a user session, indicating enrollment has finished for that user.",
       "enabled": true
     },
+    {
+      "patternId": "IME-USER-SESSION-ZERO-APPS",
+      "category": "always",
+      "pattern": "\\[Win32App\\] Get 0 apps for user session (?<sessionId>\\d+), user id = (?<userId>[a-f0-9\\-]+)",
+      "action": "userSessionZeroApps",
+      "description": "Detects when IME enumerated a user session and found zero targeted Win32 apps. IME then skips its 'Completed user session' line entirely (the zero-app branch in ApplicationPoller continues before that log statement), so this line is the only evidence that user-phase enforcement finished on a device with no user-targeted apps. Treated as equivalent to a completed user session.",
+      "enabled": true
+    },
     {
       "patternId": "IME-WIN32-STATE-1",
       "category": "currentPhase",

rules/ime-log-patterns/IME-USER-SESSION-ZERO-APPS.json

@@ -0,0 +1,9 @@
+{
+  "$schema": "../schema/ime-log-pattern.schema.json",
+  "patternId": "IME-USER-SESSION-ZERO-APPS",
+  "category": "always",
+  "pattern": "\\[Win32App\\] Get 0 apps for user session (?<sessionId>\\d+), user id = (?<userId>[a-f0-9\\-]+)",
+  "action": "userSessionZeroApps",
+  "description": "Detects when IME enumerated a user session and found zero targeted Win32 apps. IME then skips its 'Completed user session' line entirely (the zero-app branch in ApplicationPoller continues before that log statement), so this line is the only evidence that user-phase enforcement finished on a device with no user-targeted apps. Treated as equivalent to a completed user session.",
+  "enabled": true
+}

rules/schema/ime-log-pattern.schema.json

@@ -63,6 +63,7 @@
             "imeImpersonation",
             "imeTokenFailure",
             "enrollmentCompleted",
+            "userSessionZeroApps",
             "updateDoTelemetry",
             "scriptStarted",
             "scriptContext",

scripts/CustomerSetup/Grant-AutopilotMonitorAddOn.ps1

@@ -1,4 +1,4 @@
-#Requires -Version 5.1
+#Requires -Version 5.1
 <#
 .SYNOPSIS
     Grants additional Microsoft Graph application permissions to the Autopilot Monitor
@@ -28,6 +28,11 @@
         (validates Windows 365 Cloud PCs against the tenant's Cloud PC inventory --
          Cloud PCs are never Autopilot-registered, so this enables agent monitoring
          of Cloud PC first-connect enrollment)
+      - IntuneDeviceBinding     -> DeviceManagementManagedDevices.Read.All
+        (cert-to-device binding: verifies that the Intune device id carried in the
+         agent's client certificate is a device THIS tenant actually enrolled --
+         currently a Global-Admin-only preview that records telemetry and does not
+         block enrollment)
       - All                    -> every optional permission above, in one run
         (also works with -Revoke to remove all optional grants at once)
 
@@ -93,7 +98,7 @@ param(
     [string]$ClientId,
 
     [Parameter(Mandatory = $true, ParameterSetName = 'ByFeatures')]
-    [ValidateSet('All', 'ScriptDisplayNames', 'W365CloudPcValidation')]
+    [ValidateSet('All', 'ScriptDisplayNames', 'W365CloudPcValidation', 'IntuneDeviceBinding')]
     [string[]]$Features,
 
     [Parameter(Mandatory = $true, ParameterSetName = 'ByPermissions')]
@@ -116,6 +121,7 @@ $ErrorActionPreference = 'Stop'
 $FeatureCatalog = @{
     'ScriptDisplayNames'    = @('DeviceManagementScripts.Read.All')
     'W365CloudPcValidation' = @('CloudPC.Read.All')
+    'IntuneDeviceBinding'   = @('DeviceManagementManagedDevices.Read.All')
 }
 
 # Translate Features -> Permissions when caller used the high-level form.

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Configuration/BuiltInConfigCatalogTests.cs

@@ -0,0 +1,62 @@
+using System.Linq;
+using AutopilotMonitor.Agent.V2.Core.Configuration;
+using Xunit;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
+{
+    /// <summary>
+    /// Pins the compiled-in fallback rule catalogs (<see cref="BuiltInConfigCatalog"/>):
+    /// the agent embeds the SAME CI-combined rules/dist artifacts the backend serves, so a
+    /// no-config/no-cache session still runs full IME tracking. These tests guard the
+    /// embedding + deserialization end-to-end — a broken LogicalName, a moved resource or a
+    /// shape change in the combined JSON must fail HERE, not silently strip the fallback
+    /// back down to zero patterns in the field.
+    /// </summary>
+    public class BuiltInConfigCatalogTests
+    {
+        [Fact]
+        public void ImeLogPatterns_LoadFromEmbeddedCatalog_EnabledOnly()
+        {
+            var patterns = BuiltInConfigCatalog.GetEnabledImeLogPatterns(logger: null);
+
+            Assert.NotEmpty(patterns);
+            Assert.All(patterns, p =>
+            {
+                Assert.True(p.Enabled);
+                Assert.False(string.IsNullOrWhiteSpace(p.PatternId));
+                Assert.False(string.IsNullOrWhiteSpace(p.Pattern));
+                Assert.False(string.IsNullOrWhiteSpace(p.Action));
+                Assert.False(string.IsNullOrWhiteSpace(p.Category));
+            });
+            Assert.Equal(patterns.Count, patterns.Select(p => p.PatternId).Distinct().Count());
+        }
+
+        [Fact]
+        public void ImeLogPatterns_CarryCoreTrackingActions()
+        {
+            // The app-tracking backbone: without these two actions the ImeLogTracker cannot
+            // attribute installs to apps or follow ESP phases. Their presence proves the
+            // deserialization actually populated the action field (a shape drift in the
+            // combined JSON would otherwise still pass the structural checks above).
+            var actions = BuiltInConfigCatalog.GetEnabledImeLogPatterns(logger: null)
+                .Select(p => p.Action).ToHashSet();
+
+            Assert.Contains("setCurrentApp", actions);
+            Assert.Contains("espPhaseDetected", actions);
+        }
+
+        [Fact]
+        public void GatherRules_LoadFromEmbeddedCatalog_EnabledOnly()
+        {
+            // The built-in gather-rule catalog ships mostly disabled rules — only the
+            // default-enabled ones may reach the fallback config.
+            var rules = BuiltInConfigCatalog.GetEnabledGatherRules(logger: null);
+
+            Assert.All(rules, r =>
+            {
+                Assert.True(r.Enabled);
+                Assert.False(string.IsNullOrWhiteSpace(r.RuleId));
+            });
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Configuration/RemoteConfigServiceRetryTests.cs

@@ -114,6 +114,22 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
             Assert.Equal(1, svc.LastFetchAttempts);
         }
 
+        [Fact]
+        public async Task FetchConfig_DefaultsFallback_CarriesBuiltInCatalogs()
+        {
+            // A no-config/no-cache session must still get the compiled-in rule catalogs —
+            // an empty ImeLogPatterns list kills the entire IME tracking pipeline (the
+            // pre-catalog behaviour this pins against regressing to).
+            var svc = NewService(BackendBehaviour.AlwaysTransientFailure);
+
+            var result = await svc.FetchConfigAsync(retryOnTransientErrors: true);
+
+            Assert.Equal(0, result.ConfigVersion);
+            Assert.NotEmpty(result.ImeLogPatterns);
+            Assert.All(result.ImeLogPatterns, p => Assert.True(p.Enabled));
+            Assert.All(result.GatherRules, r => Assert.True(r.Enabled));
+        }
+
         [Fact]
         public async Task FetchConfig_EmptyResponse_TreatedAsTransient()
         {
@@ -128,6 +144,94 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
             Assert.Equal(4, svc.LastFetchAttempts);
         }
 
+        // ── Network-link gate (tenant aebdce78 audit 2026-08-24) ────────────
+
+        [Fact]
+        public async Task FetchConfig_InitialFetch_WaitsForNetworkLink()
+        {
+            // The initial startup fetch must gate on the link-level signal so the retry
+            // budget isn't burned into a NIC that is provably still down (agent relaunched
+            // at boot, Wi-Fi still associating).
+            var svc = NewService(BackendBehaviour.SuccessImmediately);
+
+            await svc.FetchConfigAsync(retryOnTransientErrors: true);
+
+            Assert.Equal(1, svc.NetworkLinkWaits);
+        }
+
+        [Fact]
+        public async Task FetchConfig_RotatePath_SkipsNetworkLinkGate()
+        {
+            // rotate_config runs mid-session on a proven link and must stay latency-bounded.
+            var svc = NewService(BackendBehaviour.SuccessImmediately);
+
+            await svc.FetchConfigAsync(retryOnTransientErrors: false);
+
+            Assert.Equal(0, svc.NetworkLinkWaits);
+        }
+
+        // ── Post-registration recovery ──────────────────────────────────────
+
+        [Fact]
+        public async Task TryRecover_AfterFailedFetch_AppliesLiveConfig_KeepsFailureRecord()
+        {
+            // The core self-heal contract: registration proved the backend reachable, the
+            // recovery swaps in the live config — but the LastFetch* observability keeps
+            // describing the failed startup fetch so the wire event can report both.
+            var svc = NewService(BackendBehaviour.FailAllThenSucceed);
+            await svc.FetchConfigAsync(retryOnTransientErrors: true);
+            Assert.Equal(RemoteConfigFetchOutcome.UsedDefaults, svc.LastFetchOutcome);
+
+            var recovered = await svc.TryRecoverConfigAsync();
+
+            Assert.NotNull(recovered);
+            Assert.Equal(42, recovered.ConfigVersion);
+            Assert.Equal(42, svc.CurrentConfig.ConfigVersion);
+            Assert.True(svc.RecoveredAfterFetchFailure);
+            // Startup-failure record intact:
+            Assert.Equal(RemoteConfigFetchOutcome.UsedDefaults, svc.LastFetchOutcome);
+            Assert.Equal(1 + RemoteConfigService.InitialFetchRetryBackoff.Length, svc.LastFetchAttempts);
+            Assert.Equal(nameof(HttpRequestException), svc.LastFetchFailureType);
+        }
+
+        [Fact]
+        public async Task TryRecover_AfterSuccessfulFetch_IsNoOp()
+        {
+            var svc = NewService(BackendBehaviour.SuccessImmediately);
+            await svc.FetchConfigAsync(retryOnTransientErrors: true);
+
+            var recovered = await svc.TryRecoverConfigAsync();
+
+            Assert.Null(recovered);
+            Assert.False(svc.RecoveredAfterFetchFailure);
+        }
+
+        [Fact]
+        public async Task TryRecover_BeforeAnyFetch_IsNoOp()
+        {
+            var svc = NewService(BackendBehaviour.SuccessImmediately);
+
+            var recovered = await svc.TryRecoverConfigAsync();
+
+            Assert.Null(recovered);
+            Assert.False(svc.RecoveredAfterFetchFailure);
+        }
+
+        [Fact]
+        public async Task TryRecover_WhenBackendStillFailing_KeepsFallback_NeverThrows()
+        {
+            var svc = NewService(BackendBehaviour.AlwaysTransientFailure);
+            var fallback = await svc.FetchConfigAsync(retryOnTransientErrors: true);
+            Assert.Equal(0, fallback.ConfigVersion);
+
+            var recovered = await svc.TryRecoverConfigAsync();
+
+            Assert.Null(recovered);
+            Assert.False(svc.RecoveredAfterFetchFailure);
+            Assert.Equal(0, svc.CurrentConfig.ConfigVersion); // fallback config untouched
+            Assert.Equal(RemoteConfigFetchOutcome.UsedDefaults, svc.LastFetchOutcome);
+        }
+
         // ── Outcome state defaults ──────────────────────────────────────────
 
         [Fact]
@@ -145,6 +249,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
             SuccessImmediately,
             AlwaysTransientFailure,
             FailTwiceThenSucceed,
+            /// <summary>All 4 initial attempts fail; the NEXT call (the recovery fetch) succeeds.</summary>
+            FailAllThenSucceed,
             AlwaysAuthFailure401,
             AlwaysReturnNull,
         }
@@ -177,6 +283,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
                 _behaviour = behaviour;
             }
 
+            /// <summary>Number of times the network-link gate seam was awaited.</summary>
+            public int NetworkLinkWaits { get; private set; }
+
             protected override Task<AgentConfigResponse> CallBackendAsync()
             {
                 _calls++;
@@ -188,6 +297,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
                         if (_calls < 3)
                             throw new HttpRequestException("simulated cold-start timeout");
                         return Task.FromResult(new AgentConfigResponse { ConfigVersion = 42 });
+                    case BackendBehaviour.FailAllThenSucceed:
+                        if (_calls <= 1 + RemoteConfigService.InitialFetchRetryBackoff.Length)
+                            throw new HttpRequestException("simulated dead network link");
+                        return Task.FromResult(new AgentConfigResponse { ConfigVersion = 42 });
                     case BackendBehaviour.AlwaysTransientFailure:
                         throw new HttpRequestException("simulated network failure");
                     case BackendBehaviour.AlwaysAuthFailure401:
@@ -202,6 +315,13 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Configuration
             // Zero-delay so the suite finishes in ms instead of the live 10/30/60 s schedule.
             protected override Task DelayBetweenAttemptsAsync(TimeSpan delay) => Task.CompletedTask;
 
+            // Record + skip the real link poll so the suite is deterministic on offline CI.
+            protected override Task WaitForNetworkLinkAsync()
+            {
+                NetworkLinkWaits++;
+                return Task.CompletedTask;
+            }
+
             // In-memory + per-instance cache so a prior test's successful fetch can't
             // bleed into a later test that expects UsedDefaults.
             private AgentConfigResponse? _cached;

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Monitoring/Ime/ImeLogTrackerZeroUserAppsTests.cs

@@ -0,0 +1,199 @@
+using System.Collections.Generic;
+using AutopilotMonitor.Agent.V2.Core.Logging;
+using AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime;
+using AutopilotMonitor.Agent.V2.Core.Tests.Harness;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
+{
+    /// <summary>
+    /// Sessions 81daa77f / 75d6ae8e (2026-08-25) — zero user-targeted apps starved every
+    /// user-phase completion probe. The IME zero-app branch returns before it writes the
+    /// "Completed user session" line, so a device whose user has no targeted Win32 apps never
+    /// produced the evidence <c>IsImeUserSessionGenuine</c> requires, and the AdvisoryCompletion
+    /// window failed the enrollment 30 minutes later — with the user at the desktop, Hello
+    /// provisioned and RealmJoin still installing.
+    /// <para>
+    /// IME-USER-SESSION-ZERO-APPS reads the line IME does write in that branch and routes it
+    /// through the same <see cref="ImeLogTracker.OnUserSessionCompleted"/> callback.
+    /// </para>
+    /// </summary>
+    public sealed class ImeLogTrackerZeroUserAppsTests
+    {
+        // Mirrors rules/ime-log-patterns/IME-USER-SESSION-ZERO-APPS.json. The JSON itself is
+        // pinned against a real IME line by BuiltInRulesTests in the backend suite.
+        private const string ZeroAppsPattern =
+            @"\[Win32App\] Get 0 apps for user session (?<sessionId>\d+), user id = (?<userId>[a-f0-9\-]+)";
+
+        private const string EspPhasePattern =
+            @"\[Win32App\] (?:In|The) EspPhase: (?<espPhase>\w+)";
+
+        private const string ZeroAppsLine =
+            "[Win32App] Get 0 apps for user session 1, user id = 4a1ba5db-44ed-4e69-8477-47ed88e76020";
+
+        private static ImeLogTracker BuildTracker(TempDirectory tmp) =>
+            new ImeLogTracker(
+                logFolder: tmp.Path,
+                patterns: new List<ImeLogPattern>
+                {
+                    new ImeLogPattern
+                    {
+                        PatternId = "IME-USER-SESSION-ZERO-APPS",
+                        Category = "always",
+                        Pattern = ZeroAppsPattern,
+                        Action = "userSessionZeroApps",
+                    },
+                    new ImeLogPattern
+                    {
+                        PatternId = "IME-ESP-PHASE",
+                        Category = "always",
+                        Pattern = EspPhasePattern,
+                        Action = "espPhaseDetected",
+                    },
+                },
+                logger: new AgentLogger(tmp.Path, AgentLogLevel.Info));
+
+        [Fact]
+        public void ZeroApps_line_during_user_phase_fires_user_session_completed()
+        {
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+            var fired = 0;
+            tracker.OnUserSessionCompleted = () => fired++;
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+
+            Assert.Equal(1, fired);
+            Assert.True(tracker.UserSessionZeroAppsObserved);
+        }
+
+        [Fact]
+        public void ZeroApps_line_before_user_phase_does_not_fire()
+        {
+            // Firing here would burn the adapter fire-once flag on a pre-sign-in timestamp that
+            // IsImeUserSessionGenuine (>= AccountSetupEnteredUtc) can never accept.
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+            var fired = 0;
+            tracker.OnUserSessionCompleted = () => fired++;
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: DeviceSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+
+            Assert.Equal(0, fired);
+            Assert.True(tracker.UserSessionZeroAppsObserved);
+        }
+
+        [Fact]
+        public void ZeroApps_observed_early_is_replayed_on_the_user_phase_transition()
+        {
+            // IME writes its check-in lines independently of when the agent first parses the ESP
+            // phase marker, so the evidence can legitimately arrive first. Dropping it would
+            // leave a zero-app device with no completion evidence at all.
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+            var fired = 0;
+            tracker.OnUserSessionCompleted = () => fired++;
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: DeviceSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+            Assert.Equal(0, fired);
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+
+            Assert.Equal(1, fired);
+        }
+
+        [Fact]
+        public void Repeated_user_phase_matches_do_not_replay_again()
+        {
+            // IME re-emits the phase string as it re-evaluates app sets. Only a real transition
+            // replays; the adapter fire-once flag is the outer guard, but the tracker must not
+            // spam the callback either.
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+            var fired = 0;
+            tracker.OnUserSessionCompleted = () => fired++;
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: DeviceSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+
+            Assert.Equal(1, fired);
+        }
+
+        [Fact]
+        public void Nonzero_app_count_lines_do_not_match_the_pattern()
+        {
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+            var fired = 0;
+            tracker.OnUserSessionCompleted = () => fired++;
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+
+            tracker.ProcessLogMessageForTest(
+                "[Win32App] Get 3 apps for user session 1, user id = 4a1ba5db-44ed-4e69-8477-47ed88e76020");
+            tracker.ProcessLogMessageForTest(
+                "[Win32App] Get 10 apps for user session 1, user id = 4a1ba5db-44ed-4e69-8477-47ed88e76020");
+
+            Assert.Equal(0, fired);
+            Assert.False(tracker.UserSessionZeroAppsObserved);
+        }
+
+        [Fact]
+        public void AreUserEspAppsSettled_true_after_zero_apps_observation()
+        {
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+
+            Assert.True(tracker.AreUserEspAppsSettled());
+        }
+
+        [Fact]
+        public void AreUserEspAppsSettled_stays_false_on_an_empty_list_without_the_observation()
+        {
+            // Mutation proof for the branch above: an empty live list on its own still means
+            // "phase just cleared / apps not surfaced yet", never vacuously settled.
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+
+            Assert.False(tracker.AreUserEspAppsSettled());
+        }
+
+        [Fact]
+        public void AreUserEspAppsSettled_false_outside_the_user_phase_even_after_the_observation()
+        {
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: DeviceSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+
+            Assert.True(tracker.UserSessionZeroAppsObserved);
+            Assert.False(tracker.AreUserEspAppsSettled());
+        }
+
+        [Fact]
+        public void GetPendingRequiredUserEspInstallApps_is_empty_for_a_zero_app_user_session()
+        {
+            // The adapter deferral probe must not park the signal when there is nothing to wait
+            // for — otherwise the zero-app path would starve exactly like the old one.
+            using var tmp = new TempDirectory();
+            var tracker = BuildTracker(tmp);
+
+            tracker.ProcessLogMessageForTest("[Win32App] In EspPhase: AccountSetup");
+            tracker.ProcessLogMessageForTest(ZeroAppsLine);
+
+            Assert.Empty(tracker.GetPendingRequiredUserEspInstallApps());
+        }
+    }
+}

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

@@ -1,5 +1,6 @@
 using System;
 using System.Collections.Generic;
+using System.Linq;
 using AutopilotMonitor.Agent.V2.Core.Orchestration;
 using AutopilotMonitor.DecisionCore.Signals;
 
@@ -11,8 +12,18 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Orchestration
     internal sealed class FakeSignalIngressSink : ISignalIngressSink
     {
         private readonly List<PostedSignal> _posted = new List<PostedSignal>();
+        private readonly object _gate = new object();
 
-        public IReadOnlyList<PostedSignal> Posted => _posted;
+        /// <summary>
+        /// Snapshot, not the live list: producers post from timer / worker threads (e.g. the
+        /// parked-dwell tripwire) while tests poll this in a SpinWait loop. Enumerating the live
+        /// list across an Add throws "Collection was modified" — a rare CI-only failure
+        /// (run 32644425162) with no assertion text.
+        /// </summary>
+        public IReadOnlyList<PostedSignal> Posted
+        {
+            get { lock (_gate) return _posted.ToArray(); }
+        }
 
         /// <summary>When non-null, <see cref="Post"/> throws instead of capturing.</summary>
         public Exception? ThrowOnPost { get; set; }
@@ -27,7 +38,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Orchestration
             object? typedPayload = null)
         {
             if (ThrowOnPost != null) throw ThrowOnPost;
-            _posted.Add(new PostedSignal(kind, occurredAtUtc, sourceOrigin, evidence, payload, kindSchemaVersion, typedPayload));
+            lock (_gate)
+                _posted.Add(new PostedSignal(kind, occurredAtUtc, sourceOrigin, evidence, payload, kindSchemaVersion, typedPayload));
         }
 
         internal sealed class PostedSignal

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Program/LifecycleEmittersRemoteConfigTests.cs

@@ -182,6 +182,73 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Program
             Assert.Equal("FromCache", data["outcome"]);
         }
 
+        // ── Post-registration recovery (Phase 6.5, tenant aebdce78 audit 2026-08-24) ─
+
+        [Fact]
+        public void EmitFetchFailedIfAny_AfterRecovery_MarksRecoveredAndDemotesToInfo()
+        {
+            using var rig = new EmitterRig();
+            // Startup fetch failed (network link not up yet), then the post-registration
+            // recovery fetched ConfigVersion 39. The event must keep the original failure
+            // details AND mark the heal.
+            rig.RemoteConfig.SetFakeOutcome(
+                RemoteConfigFetchOutcome.UsedDefaults,
+                configVersion: 39, // CurrentConfig now holds the recovered config
+                attempts: 4,
+                failureType: "HttpRequestException",
+                failureMessage: "simulated dead network link",
+                recovered: true);
+
+            LifecycleEmitters.EmitRemoteConfigFetchFailedIfAny(
+                rig.Post, rig.AgentConfig, rig.RemoteConfig, rig.Logger);
+
+            var data = AssertSingleEmit(rig.Sink, expectedEventType: "remote_config_fetch_failed");
+            Assert.Equal("UsedDefaults", data["outcome"]); // startup failure record intact
+            Assert.Equal(4, data["attempts"]);
+            Assert.Equal(true, data["recovered"]);
+            Assert.Equal(39, data["recoveredConfigVersion"]);
+            Assert.Equal(nameof(EventSeverity.Info), rig.Sink.Posted[0].Payload!["severity"]);
+        }
+
+        [Fact]
+        public void EmitFetchFailedIfAny_WithoutRecovery_StaysWarning()
+        {
+            using var rig = new EmitterRig();
+            rig.RemoteConfig.SetFakeOutcome(
+                RemoteConfigFetchOutcome.UsedDefaults,
+                configVersion: 0,
+                attempts: 4,
+                failureType: "HttpRequestException",
+                failureMessage: "still down");
+
+            LifecycleEmitters.EmitRemoteConfigFetchFailedIfAny(
+                rig.Post, rig.AgentConfig, rig.RemoteConfig, rig.Logger);
+
+            var data = AssertSingleEmit(rig.Sink, expectedEventType: "remote_config_fetch_failed");
+            Assert.False(data.ContainsKey("recovered"));
+            Assert.False(data.ContainsKey("recoveredConfigVersion"));
+            Assert.Equal(nameof(EventSeverity.Warning), rig.Sink.Posted[0].Payload!["severity"]);
+        }
+
+        [Fact]
+        public void EmitAgentStarted_AfterRecovery_FlagsLiveConfig()
+        {
+            using var rig = new EmitterRig();
+            rig.RemoteConfig.SetFakeOutcome(
+                RemoteConfigFetchOutcome.UsedDefaults,
+                configVersion: 39,
+                recovered: true);
+
+            LifecycleEmitters.EmitAgentStarted(
+                rig.Post, rig.AgentConfig, rig.PreviousExit, "2.0.806", rig.RemoteConfig, rig.Logger);
+
+            var data = AssertSingleEmit(rig.Sink, expectedEventType: "agent_started");
+            Assert.Equal(39, data["configVersion"]);
+            Assert.Equal(true, data["remoteConfigFetched"]); // running on a live-fetched config
+            Assert.Equal("UsedDefaults", data["remoteConfigOutcome"]); // startup outcome stays honest
+            Assert.Equal(true, data["remoteConfigRecovered"]);
+        }
+
         // ── Helpers ─────────────────────────────────────────────────────────
 
         private static Dictionary<string, object> AssertSingleEmit(
@@ -258,8 +325,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Program
                 int attempts = 1,
                 string? failureType = null,
                 string? failureMessage = null,
-                int? authStatusCode = null)
+                int? authStatusCode = null,
+                bool recovered = false)
             {
+                typeof(RemoteConfigService).GetProperty(nameof(RecoveredAfterFetchFailure))?.SetValue(this, recovered);
                 // CurrentConfig is set via the protected SetConfig path — easiest:
                 // a real FetchConfigAsync replacement that just stashes a canned config.
                 typeof(RemoteConfigService).GetMethod("SetConfig",

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

@@ -0,0 +1,146 @@
+using System;
+using AutopilotMonitor.Agent.V2.Core.Security;
+using Xunit;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Security
+{
+    /// <summary>
+    /// Pins <see cref="EntraDeviceCertHelper.ParseGuidFromExtension"/>, the pure byte-level part of the
+    /// cert probe. Both encodings Microsoft actually ships are covered: the ASN.1 OCTET STRING wrapper
+    /// used by the MS-Organization-Access TenantId/DeviceId extensions (and Intune's
+    /// <c>1.2.840.113556.5.6</c> / <c>5.14</c>), and the BARE 16-byte payload used by Intune's DeviceId
+    /// extension <c>1.2.840.113556.5.4</c> — the latter was silently returning <c>null</c> before.
+    /// The store-iteration shell around it needs a real machine cert store and is not unit-tested.
+    /// </summary>
+    public sealed class EntraDeviceCertHelperParseTests
+    {
+        // Synthetic — never a real tenant/device id. ToByteArray() gives the MS little-endian layout
+        // that both cert encodings carry, so every expectation below is derived, not hand-transcribed.
+        private static readonly Guid Sample = new Guid("11223344-5566-7788-99aa-bbccddeeff00");
+
+        private static byte[] Wrap(byte[] payload, params byte[] header)
+        {
+            var buf = new byte[header.Length + payload.Length];
+            Buffer.BlockCopy(header, 0, buf, 0, header.Length);
+            Buffer.BlockCopy(payload, 0, buf, header.Length, payload.Length);
+            return buf;
+        }
+
+        [Fact]
+        public void OctetString_ShortForm_Parses()
+        {
+            var raw = Wrap(Sample.ToByteArray(), 0x04, 0x10);
+
+            Assert.Equal(Sample, EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        [Fact]
+        public void OctetString_LongForm_Parses()
+        {
+            var raw = Wrap(Sample.ToByteArray(), 0x04, 0x81, 0x10);
+
+            Assert.Equal(Sample, EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        /// <summary>
+        /// The regression this class exists for: Intune's <c>1.2.840.113556.5.4</c> extension value is
+        /// exactly 16 bytes with no <c>04 10</c> prefix, so the strict OCTET-STRING parser bailed on the
+        /// tag check and the DeviceId was unreadable.
+        /// </summary>
+        [Fact]
+        public void BareSixteenBytes_Parses()
+        {
+            var raw = Sample.ToByteArray();
+
+            Assert.Equal(Sample, EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        /// <summary>
+        /// A bare GUID may legitimately start with 0x04 (the OCTET STRING tag). That must not be
+        /// mistaken for a wrapper: a wrapped 16-byte GUID is always >= 18 bytes, so the two forms are
+        /// disjoint by length and the bare path has to win here.
+        /// </summary>
+        [Fact]
+        public void BareSixteenBytes_StartingWithOctetStringTag_StillParses()
+        {
+            var guid = new Guid("aabbcc04-0110-4020-8030-405060708090");
+            var raw = guid.ToByteArray();
+
+            Assert.Equal(0x04, raw[0]); // guard: the byte layout really does start with the ASN.1 tag
+            Assert.Equal(16, raw.Length);
+            Assert.Equal(guid, EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        [Fact]
+        public void OctetString_IsLittleEndianMsLayout()
+        {
+            // 04 10 + the canonical MS binary layout of {11223344-...} => first group byte-reversed.
+            var raw = new byte[]
+            {
+                0x04, 0x10,
+                0x44, 0x33, 0x22, 0x11,
+                0x66, 0x55,
+                0x88, 0x77,
+                0x99, 0xAA, 0xBB, 0xCC, 0xDD, 0xEE, 0xFF, 0x00
+            };
+
+            Assert.Equal(Sample, EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        [Fact]
+        public void Null_ReturnsNull()
+        {
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(null));
+        }
+
+        [Fact]
+        public void Empty_ReturnsNull()
+        {
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(new byte[0]));
+        }
+
+        [Theory]
+        [InlineData(15)]
+        [InlineData(17)]
+        public void BarePayload_OfWrongLength_ReturnsNull(int length)
+        {
+            var raw = new byte[length];
+            for (var i = 0; i < length; i++) raw[i] = (byte)(0x40 + i); // no 0x04 tag, no valid wrapper
+
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        [Fact]
+        public void NonOctetStringTag_ReturnsNull()
+        {
+            // e.g. Intune's 1.2.840.113556.5.15 => ASN.1 INTEGER 2.
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(new byte[] { 0x02, 0x01, 0x02 }));
+        }
+
+        [Fact]
+        public void OctetString_WithNonGuidContentLength_ReturnsNull()
+        {
+            var raw = Wrap(new byte[8], 0x04, 0x08);
+
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        [Fact]
+        public void OctetString_TruncatedContent_ReturnsNull()
+        {
+            // Declares 16 content bytes but only carries 10.
+            var raw = Wrap(new byte[10], 0x04, 0x10);
+
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+
+        [Fact]
+        public void OctetString_LongFormWithOversizedLengthHeader_ReturnsNull()
+        {
+            // 0x85 => 5 length bytes, above the 4 the parser accepts.
+            var raw = Wrap(Sample.ToByteArray(), 0x04, 0x85, 0x00, 0x00, 0x00, 0x00, 0x10);
+
+            Assert.Null(EntraDeviceCertHelper.ParseGuidFromExtension(raw));
+        }
+    }
+}

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

@@ -97,7 +97,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Security
         }
 
         [Fact]
-        public async Task Fails_after_five_exceptions_without_auth_error()
+        public async Task Fails_after_six_exceptions_without_auth_error()
         {
             using var tmp = new TempDirectory();
             var logger = NewLogger(tmp);
@@ -111,8 +111,50 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Security
                 backoffDelay: _ => Task.CompletedTask);
 
             Assert.Equal(SessionRegistrationOutcome.Failed, result.Outcome);
-            Assert.Equal(5, apiClient.CallCount);
-            Assert.Contains("transient attempt 5", result.ErrorMessage);
+            Assert.Equal(6, apiClient.CallCount);
+            Assert.Contains("transient attempt 6", result.ErrorMessage);
+        }
+
+        [Fact]
+        public async Task Waits_for_network_link_before_first_attempt()
+        {
+            using var tmp = new TempDirectory();
+            var logger = NewLogger(tmp);
+            var config = NewConfig();
+
+            var order = new System.Collections.Generic.List<string>();
+            var apiClient = new FakeApiClient(_ =>
+            {
+                order.Add("register");
+                return new RegisterSessionResponse { Success = true, SessionId = config.SessionId };
+            });
+
+            var result = await SessionRegistrationHelper.RegisterWithRetryAsync(
+                apiClient, config, agentVersion: "0.0.0", logger: logger,
+                backoffDelay: _ => Task.CompletedTask,
+                networkLinkWait: _ => { order.Add("link-wait"); return Task.CompletedTask; });
+
+            Assert.Equal(SessionRegistrationOutcome.Succeeded, result.Outcome);
+            Assert.Equal(new[] { "link-wait", "register" }, order);
+        }
+
+        [Fact]
+        public async Task Network_link_wait_failure_does_not_block_registration()
+        {
+            using var tmp = new TempDirectory();
+            var logger = NewLogger(tmp);
+            var config = NewConfig();
+
+            var apiClient = new FakeApiClient(_ =>
+                new RegisterSessionResponse { Success = true, SessionId = config.SessionId });
+
+            var result = await SessionRegistrationHelper.RegisterWithRetryAsync(
+                apiClient, config, agentVersion: "0.0.0", logger: logger,
+                backoffDelay: _ => Task.CompletedTask,
+                networkLinkWait: _ => throw new InvalidOperationException("probe broken"));
+
+            Assert.Equal(SessionRegistrationOutcome.Succeeded, result.Outcome);
+            Assert.Equal(1, apiClient.CallCount);
         }
 
         [Fact]

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

@@ -163,5 +163,45 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.SignalAdapters
             Assert.Single(f.DecisionSignals(DecisionSignalKind.ImeUserSessionCompleted));
             Assert.Single(f.InfoEvents(SharedEventTypes.ImeUserSessionCompleted));
         }
+
+        // ==================================================== zero-app evidence label ====
+
+        [Fact]
+        public void Zero_app_user_session_labels_the_evidence_on_signal_and_timeline()
+        {
+            // Sessions 81daa77f / 75d6ae8e: with no user-targeted apps IME never writes its
+            // "Completed user session" line at all, so the verdict rests on a different
+            // observation. Whoever reads the timeline later must be able to tell which.
+            using var f = new ImeLogTrackerAdapterFixture();
+            using var adapter = new ImeLogTrackerAdapter(f.Tracker, f.Ingress, f.Clock);
+            f.Tracker.SeedCurrentPhaseForTesting("AccountSetup");
+            f.Tracker.SeedUserSessionZeroAppsForTesting();
+
+            adapter.TriggerUserSessionCompletedFromTest();
+
+            Assert.Single(f.DecisionSignals(DecisionSignalKind.ImeUserSessionCompleted));
+            var evt = Assert.Single(f.InfoEvents(SharedEventTypes.ImeUserSessionCompleted));
+            Assert.Equal(
+                ImeLogTrackerAdapter.EvidenceZeroUserApps,
+                evt.Payload!["evidence"]);
+        }
+
+        [Fact]
+        public void Ordinary_user_session_completion_keeps_the_default_evidence_label()
+        {
+            // Mutation proof for the branch above.
+            using var f = new ImeLogTrackerAdapterFixture();
+            using var adapter = new ImeLogTrackerAdapter(f.Tracker, f.Ingress, f.Clock);
+            f.Tracker.SeedCurrentPhaseForTesting("AccountSetup");
+            AddApp(f.Tracker, "app-done", AppIntent.Install, AppInstallationState.Installed);
+
+            adapter.TriggerUserSessionCompletedFromTest();
+
+            var evt = Assert.Single(f.InfoEvents(SharedEventTypes.ImeUserSessionCompleted));
+            Assert.Equal(
+                ImeLogTrackerAdapter.EvidenceUserSessionCompleted,
+                evt.Payload!["evidence"]);
+        }
+
     }
 }

src/Agent/AutopilotMonitor.Agent.V2.Core/AutopilotMonitor.Agent.V2.Core.csproj

@@ -52,6 +52,11 @@
          carries the V2 assembly namespace so GatherRuleGuards locates it under
          "AutopilotMonitor.Agent.V2.Core.Resources.guardrails.json". -->
     <EmbeddedResource Include="..\..\..\rules\guardrails.json" LogicalName="AutopilotMonitor.Agent.V2.Core.Resources.guardrails.json" />
+    <!-- Built-in fallback rule catalogs (BuiltInConfigCatalog): the SAME CI-combined
+         rules/dist artifacts the backend embeds and serves. Snapshotted per agent build so
+         the no-config/no-cache fallback still runs full IME tracking instead of nothing. -->
+    <EmbeddedResource Include="..\..\..\rules\dist\ime-log-patterns.json" LogicalName="AutopilotMonitor.Agent.V2.Core.Resources.ime-log-patterns.json" />
+    <EmbeddedResource Include="..\..\..\rules\dist\gather-rules.json" LogicalName="AutopilotMonitor.Agent.V2.Core.Resources.gather-rules.json" />
   </ItemGroup>
 
   <ItemGroup>

src/Agent/AutopilotMonitor.Agent.V2.Core/Configuration/BuiltInConfigCatalog.cs

@@ -0,0 +1,74 @@
+using System;
+using System.Collections.Generic;
+using System.IO;
+using System.Linq;
+using AutopilotMonitor.Agent.V2.Core.Logging;
+using AutopilotMonitor.Shared.Models;
+using Newtonsoft.Json;
+
+namespace AutopilotMonitor.Agent.V2.Core.Configuration
+{
+    /// <summary>
+    /// Compiled-in fallback copies of the global rule catalogs, embedded from the same
+    /// CI-generated <c>rules/dist</c> artifacts the backend seeds and serves — single
+    /// source by construction: each agent release snapshots the catalog state of its own
+    /// build, so the offline fallback can never drift further than the running agent
+    /// version itself (no manually maintained default to forget).
+    /// <para>
+    /// Consumed only by <see cref="RemoteConfigService"/>'s built-in default config,
+    /// i.e. when a session has NO live config and NO cache. Before this existed, that
+    /// path shipped an empty <c>ImeLogPatterns</c> list, which left the entire IME
+    /// tracking pipeline (app installs, platform/remediation scripts, health-script
+    /// compliance, IME ESP-phase signals) dead for the whole session — the core of
+    /// enrollment monitoring. A live fetch (startup, post-registration recovery or
+    /// rotate_config) always wins over this catalog.
+    /// </para>
+    /// Fail-soft: any load/parse error returns an empty list (the pre-catalog behaviour).
+    /// </summary>
+    internal static class BuiltInConfigCatalog
+    {
+        /// <summary>Enabled built-in IME log patterns from the embedded catalog.</summary>
+        public static List<ImeLogPattern> GetEnabledImeLogPatterns(AgentLogger logger)
+            => LoadEnabled<ImeLogPattern>("ime-log-patterns.json", p => p.Enabled, logger);
+
+        /// <summary>Enabled built-in gather rules from the embedded catalog.</summary>
+        public static List<GatherRule> GetEnabledGatherRules(AgentLogger logger)
+            => LoadEnabled<GatherRule>("gather-rules.json", r => r.Enabled, logger);
+
+        private static List<T> LoadEnabled<T>(string name, Func<T, bool> isEnabled, AgentLogger logger)
+        {
+            try
+            {
+                var assembly = typeof(BuiltInConfigCatalog).Assembly;
+                using (var stream = assembly.GetManifestResourceStream(
+                    "AutopilotMonitor.Agent.V2.Core.Resources." + name))
+                {
+                    if (stream == null)
+                    {
+                        logger?.Warning($"Built-in config catalog resource '{name}' not found — fallback runs without it.");
+                        return new List<T>();
+                    }
+
+                    using (var reader = new StreamReader(stream))
+                    {
+                        var wrapper = JsonConvert.DeserializeObject<RulesFile<T>>(reader.ReadToEnd());
+                        var all = wrapper?.Rules ?? new List<T>();
+                        return all.Where(isEnabled).ToList();
+                    }
+                }
+            }
+            catch (Exception ex)
+            {
+                logger?.Warning($"Built-in config catalog '{name}' failed to load ({ex.Message}) — fallback runs without it.");
+                return new List<T>();
+            }
+        }
+
+        /// <summary>Shape of the CI-combined catalog files: <c>{ "rules": [...] }</c>.</summary>
+        private sealed class RulesFile<T>
+        {
+            [JsonProperty("rules")]
+            public List<T> Rules { get; set; }
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core/Configuration/RemoteConfigService.cs

@@ -42,6 +42,15 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
             TimeSpan.FromSeconds(60),
         };
 
+        /// <summary>
+        /// Upper bound for the pre-fetch network-link wait on the initial startup fetch.
+        /// Deliberately longer than the registration gate's 15s: the fetch runs FIRST after a
+        /// boot-time relaunch, when Wi-Fi may still be associating (802.1X, OOBE), and a fetch
+        /// burned into a dead link strands the whole session on built-in defaults — tenant aebdce78
+        /// audit 2026-08-24: ~50% of sessions lost their tenant config exactly this way.
+        /// </summary>
+        internal static readonly TimeSpan InitialFetchNetworkLinkWaitMax = TimeSpan.FromSeconds(60);
+
         /// <summary>
         /// Gets the current configuration (thread-safe)
         /// </summary>
@@ -76,6 +85,13 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
         /// <summary>HTTP status for auth failures (401/403), null otherwise.</summary>
         public int? LastFetchAuthStatusCode { get; private set; }
 
+        /// <summary>
+        /// True once <see cref="TryRecoverConfigAsync"/> replaced a failed startup fetch with a
+        /// live config. The <c>LastFetch*</c> properties intentionally keep describing the failed
+        /// startup fetch so the wire event can report both the original failure and the recovery.
+        /// </summary>
+        public bool RecoveredAfterFetchFailure { get; private set; }
+
         public RemoteConfigService(BackendApiClient apiClient, string tenantId, AgentLogger logger, EmergencyReporter emergencyReporter = null, DistressReporter distressReporter = null, AuthFailureTracker authFailureTracker = null)
         {
             _apiClient = apiClient ?? throw new ArgumentNullException(nameof(apiClient));
@@ -116,6 +132,18 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
             LastFetchFailureMessage = null;
             LastFetchAuthStatusCode = null;
 
+            // Initial startup fetch only: give a still-associating NIC a chance to come up
+            // before the first attempt. Without this the retry budget (10/30/60s) is burned
+            // into a provably dead link — the agent starts at boot, Wi-Fi associates minutes
+            // later, and the session runs on defaults although the backend was never the
+            // problem. The rotate_config path (retryOnTransientErrors:false) runs mid-session
+            // on a proven link and must stay latency-bounded, so it skips the gate.
+            if (retryOnTransientErrors)
+            {
+                try { await WaitForNetworkLinkAsync().ConfigureAwait(false); }
+                catch (Exception ex) { _logger.Debug($"Pre-fetch network wait failed ({ex.Message}) — fetching anyway."); }
+            }
+
             var maxAttempts = retryOnTransientErrors ? InitialFetchRetryBackoff.Length + 1 : 1;
             Exception lastException = null;
 
@@ -203,14 +231,63 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
                 return cached;
             }
 
-            // Fall back to defaults
-            _logger.Info("No cached config available, using defaults (all optional collectors disabled)");
+            // Fall back to built-in defaults (collector/analyzer defaults from the shared
+            // model initializers + the compiled-in rule catalogs).
             var defaults = CreateDefaultConfig();
+            _logger.Info($"No cached config available, using built-in defaults (ConfigVersion=0, " +
+                         $"{defaults.ImeLogPatterns?.Count ?? 0} built-in IME patterns, " +
+                         $"{defaults.GatherRules?.Count ?? 0} built-in gather rules)");
             SetConfig(defaults);
             LastFetchOutcome = RemoteConfigFetchOutcome.UsedDefaults;
             return defaults;
         }
 
+        /// <summary>
+        /// Single-attempt recovery fetch for a session whose startup fetch failed. Called right
+        /// after session registration succeeded — registration proves the backend is reachable,
+        /// so the config that Phase 4 could not get is one cheap call away instead of the whole
+        /// session being stranded on cache/defaults (only a manual <c>rotate_config</c> action
+        /// could rescue it before this existed).
+        /// <para>
+        /// Unlike <see cref="FetchConfigAsync"/> this does NOT reset the <c>LastFetch*</c>
+        /// observability state — those keep describing the failed startup fetch so
+        /// <c>remote_config_fetch_failed</c> can report the original failure AND the recovery.
+        /// Returns the recovered config, or null when there is nothing to recover (startup fetch
+        /// succeeded / never ran) or the recovery attempt itself failed. Never throws.
+        /// </para>
+        /// </summary>
+        public async Task<AgentConfigResponse> TryRecoverConfigAsync()
+        {
+            if (LastFetchOutcome != RemoteConfigFetchOutcome.FromCache &&
+                LastFetchOutcome != RemoteConfigFetchOutcome.UsedDefaults)
+                return null;
+
+            try
+            {
+                _logger.Info("Retrying remote config fetch after successful session registration...");
+                var config = await CallBackendAsync().ConfigureAwait(false);
+                if (config == null)
+                {
+                    _logger.Warning("Config recovery fetch returned empty response — keeping fallback config.");
+                    return null;
+                }
+
+                _logger.Info($"Remote config recovered: ConfigVersion={config.ConfigVersion}, " +
+                             $"GatherRules={config.GatherRules?.Count ?? 0}");
+                LogCollectorSettings(config.Collectors);
+                SetConfig(config);
+                CacheConfig(config);
+                _authFailureTracker?.RecordSuccess();
+                RecoveredAfterFetchFailure = true;
+                return config;
+            }
+            catch (Exception ex)
+            {
+                _logger.Warning($"Config recovery fetch failed ({ex.Message}) — keeping fallback config.");
+                return null;
+            }
+        }
+
         private void SetConfig(AgentConfigResponse config)
         {
             lock (_configLock)
@@ -319,6 +396,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
 
         private AgentConfigResponse CreateDefaultConfig()
         {
+            // Everything not set here falls to the shared AgentConfigResponse property
+            // initializers — the same values the backend uses as its `??` fallbacks, so the
+            // scalar knobs cannot drift. The rule catalogs are the one thing initializers
+            // cannot supply: without them the whole IME tracking pipeline is dead, so they
+            // come from the compiled-in snapshot of the served catalogs (BuiltInConfigCatalog).
             return new AgentConfigResponse
             {
                 ConfigVersion = 0,
@@ -330,8 +412,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
                 EnableGatherRuleDebugLog = false,
                 EnableEspContinueAnywayObservation = false,
                 Collectors = CollectorConfiguration.CreateDefault(),
-                GatherRules = new System.Collections.Generic.List<GatherRule>(),
-                ImeLogPatterns = new System.Collections.Generic.List<ImeLogPattern>()
+                GatherRules = BuiltInConfigCatalog.GetEnabledGatherRules(_logger),
+                ImeLogPatterns = BuiltInConfigCatalog.GetEnabledImeLogPatterns(_logger)
             };
         }
 
@@ -358,6 +440,15 @@ namespace AutopilotMonitor.Agent.V2.Core.Configuration
         protected virtual Task DelayBetweenAttemptsAsync(TimeSpan delay)
             => Task.Delay(delay);
 
+        /// <summary>
+        /// Awaits the pre-fetch network-link gate (initial startup fetch only). Production
+        /// polls the link-level signal via <see cref="NetworkLinkWait"/> bounded by
+        /// <see cref="InitialFetchNetworkLinkWaitMax"/>; tests subclass to record the call
+        /// and return immediately.
+        /// </summary>
+        protected virtual Task WaitForNetworkLinkAsync()
+            => NetworkLinkWait.WaitAsync(_logger, InitialFetchNetworkLinkWaitMax, "Remote config fetch");
+
         public void Dispose()
         {
             // Nothing to dispose

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/Ime/ImeLogTracker.Handlers.cs

@@ -1042,6 +1042,56 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
         /// </summary>
         internal string LastEspPhaseDetected => _lastEspPhaseDetected;
 
+        /// <summary>
+        /// True while the user-scope ESP phase is the tracker's current phase. Every user-phase
+        /// probe and the zero-apps evidence gate on this — outside AccountSetup the IME lines
+        /// belong to the device phase or the pre-sign-in (defaultuser0) frame.
+        /// </summary>
+        private bool IsUserEspPhaseActive =>
+            string.Equals(_lastEspPhaseDetected, "AccountSetup", StringComparison.OrdinalIgnoreCase);
+
+        /// <summary>
+        /// True once IME reported zero targeted Win32 apps for a user session while the user ESP
+        /// phase was active. Read by the adapter to label the evidence behind
+        /// <c>ime_user_session_completed</c>.
+        /// </summary>
+        internal bool UserSessionZeroAppsObserved => _userSessionZeroAppsObserved;
+
+        /// <summary>
+        /// IME-USER-SESSION-ZERO-APPS: IME enumerated the user session and found no targeted
+        /// Win32 apps. Its zero-app branch then <c>continue</c>s past the "Completed user
+        /// session" log statement, so this line is the only completion evidence such a device
+        /// ever produces — and IME itself treats the same condition as finished
+        /// (<c>ProviderResult.ProvisioningComplete</c> in the user check-in path, and an ESP
+        /// lock-in with an empty application list in the poller path).
+        /// <para>
+        /// Deliberately routed through <see cref="OnUserSessionCompleted"/> rather than a
+        /// separate callback: downstream this IS a completed user session, and the adapter's
+        /// pending-apps deferral plus its fire-once flag then apply unchanged.
+        /// </para>
+        /// <para>
+        /// Phase guard: firing before AccountSetup would burn the adapter's fire-once flag on a
+        /// pre-sign-in timestamp that <c>IsImeUserSessionGenuine</c> can never accept. The
+        /// observation is remembered instead and replayed by
+        /// <see cref="HandleEspPhaseDetected"/> on the transition into the user phase.
+        /// </para>
+        /// </summary>
+        private void HandleUserSessionZeroApps()
+        {
+            _userSessionZeroAppsObserved = true;
+
+            if (!IsUserEspPhaseActive)
+            {
+                _logger.Debug(
+                    "ImeLogTracker: zero-app user session observed before the user ESP phase " +
+                    $"(current phase: {_lastEspPhaseDetected ?? "(none)"}) - deferring until AccountSetup");
+                return;
+            }
+
+            _logger.Info("ImeLogTracker: user session enumerated with zero targeted apps - treating as user session completed");
+            OnUserSessionCompleted?.Invoke();
+        }
+
         private void HandleEspPhaseDetected(string espPhaseString)
         {
             // Validate phase and get its ordinal for forward-only enforcement
@@ -1120,6 +1170,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
             ActivatePatterns(logPhaseIsCurrentPhase);
 
             OnEspPhaseChanged?.Invoke(espPhaseString);
+
+            // Replay a zero-app user session observed before this phase was known. IME logs its
+            // check-in lines independently of when the agent first parses the ESP phase marker,
+            // so the evidence can legitimately arrive first; without this the only completion
+            // evidence a zero-app device produces would be dropped on the floor.
+            if (isFirstOrChangedPhase && IsUserEspPhaseActive && _userSessionZeroAppsObserved)
+            {
+                _logger.Info("ImeLogTracker: replaying zero-app user session evidence observed before the user ESP phase");
+                OnUserSessionCompleted?.Invoke();
+            }
         }
 
         private void HandleEspTrackStatus(Match match)

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Enrollment/Ime/ImeLogTracker.LogProcessing.cs

@@ -387,6 +387,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
                         OnUserSessionCompleted?.Invoke();
                         break;
 
+                    case "usersessionzeroapps":
+                        // IME found no targeted user apps — its zero-app branch skips the
+                        // "Completed user session" line, so this is the equivalent evidence.
+                        HandleUserSessionZeroApps();
+                        break;
+
                     case "updatestateinstalled":
                         if (!string.IsNullOrEmpty(id))
                             UpdateStateWithCallback(id, AppInstallationState.Installed);

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

@@ -129,6 +129,14 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
         private CancellationTokenSource _cts;
         private bool _allAppsCompletedFired;
 
+        // Sticky: IME enumerated a user session and reported ZERO targeted Win32 apps
+        // (IME-USER-SESSION-ZERO-APPS). IME's zero-app branch returns before it writes the
+        // "Completed user session" line, so on a device with no user-targeted apps that line
+        // never appears and every user-phase completion probe would starve forever. This flag
+        // IS the user-phase completion evidence for that shape. Set only while the user ESP
+        // phase is active, or replayed on the transition into it.
+        private bool _userSessionZeroAppsObserved;
+
         // State persistence: saves tracker state to disk so agent restart continues
         // from the exact log position without re-parsing or re-building ignore lists.
         private readonly ImeTrackerStatePersistence _statePersistence;
@@ -341,6 +349,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
             _lastEspPhaseDetected = phaseName;
         }
 
+        /// <summary>
+        /// Test-only seam: marks the zero-targeted-apps user session as observed without driving
+        /// the IME-USER-SESSION-ZERO-APPS pattern through a log line. The pattern-to-flag path
+        /// itself is covered end-to-end by <c>ImeLogTrackerZeroUserAppsTests</c>.
+        /// </summary>
+        internal void SeedUserSessionZeroAppsForTesting()
+        {
+            _userSessionZeroAppsObserved = true;
+        }
+
         /// <summary>
         /// Returns a deduped union of phase-snapshotted apps plus the live
         /// <see cref="PackageStates"/> (current phase). Live entries win on Id collision.
@@ -395,11 +413,18 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
         {
             try
             {
-                if (!string.Equals(_lastEspPhaseDetected, "AccountSetup", StringComparison.OrdinalIgnoreCase))
+                if (!IsUserEspPhaseActive)
                     return false;
 
                 var apps = new List<AppPackageState>(_packageStates);
-                if (apps.Count == 0) return false;
+
+                // Zero user-targeted apps is the MOST settled state there is, but the
+                // conservative "at least one required app" rule below reads it as "not
+                // settled" and starves the AccountSetup synthesis. Only an explicit IME
+                // observation lifts that — never a merely empty list, which also means
+                // "phase just cleared" or "apps not surfaced yet".
+                if (apps.Count == 0)
+                    return _userSessionZeroAppsObserved;
 
                 var sawRequired = false;
                 foreach (var pkg in apps)
@@ -746,6 +771,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
             _currentPhaseOrder = state.CurrentPhaseOrder;
             _lastEspPhaseDetected = state.LastEspPhaseDetected;
             _allAppsCompletedFired = state.AllAppsCompletedFired;
+            _userSessionZeroAppsObserved = state.UserSessionZeroAppsObserved;
             _logPhaseIsCurrentPhase = state.LogPhaseIsCurrentPhase;
 
             // Restore seen app IDs
@@ -861,6 +887,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
                 CurrentPhaseOrder = _currentPhaseOrder,
                 LastEspPhaseDetected = _lastEspPhaseDetected,
                 AllAppsCompletedFired = _allAppsCompletedFired,
+                UserSessionZeroAppsObserved = _userSessionZeroAppsObserved,
                 LogPhaseIsCurrentPhase = _logPhaseIsCurrentPhase,
                 SeenAppIds = _seenAppIds.ToList(),
                 IgnoreList = _packageStates.IgnoreList.ToList(),

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

@@ -109,6 +109,13 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
         public int CurrentPhaseOrder { get; set; }
         public string LastEspPhaseDetected { get; set; }
         public bool AllAppsCompletedFired { get; set; }
+
+        /// <summary>
+        /// IME reported zero targeted Win32 apps for a user session. Absent in state files
+        /// written before this field existed — deserializing to false there is correct: the
+        /// tracker simply re-observes the line on the next IME check-in.
+        /// </summary>
+        public bool UserSessionZeroAppsObserved { get; set; }
         public bool LogPhaseIsCurrentPhase { get; set; }
         public List<string> SeenAppIds { get; set; }
         public List<string> IgnoreList { get; set; }

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Transport/NetworkLinkWait.cs

@@ -0,0 +1,38 @@
+using System;
+using System.Net.NetworkInformation;
+using System.Threading.Tasks;
+using AutopilotMonitor.Agent.V2.Core.Logging;
+
+namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Transport
+{
+    /// <summary>
+    /// Boot-time NIC grace shared by every first-contact backend call (remote-config fetch,
+    /// session registration, emergency-break reporter pattern): polls the cheap link-level
+    /// signal once per second for at most <paramref name="maxWait"/>. It cannot prove backend
+    /// reachability — the callers' retry loops handle that — it only keeps retry budgets from
+    /// being burned into a link that is provably still down (BootTrigger relaunch after a
+    /// mid-enrollment reboot, Wi-Fi still associating; tenant aebdce78 audits 2026-08-23/24).
+    /// Free on the normal path — a live link returns immediately. Probe errors end the wait,
+    /// never the caller's operation.
+    /// </summary>
+    internal static class NetworkLinkWait
+    {
+        internal static async Task WaitAsync(AgentLogger logger, TimeSpan maxWait, string context)
+        {
+            if (NetworkInterface.GetIsNetworkAvailable()) return;
+
+            logger?.Info($"{context}: no network link yet — waiting up to {maxWait.TotalSeconds:F0}s.");
+            var deadline = DateTime.UtcNow + maxWait;
+            while (DateTime.UtcNow < deadline)
+            {
+                await Task.Delay(1000).ConfigureAwait(false);
+                if (NetworkInterface.GetIsNetworkAvailable())
+                {
+                    logger?.Info($"{context}: network link is up.");
+                    return;
+                }
+            }
+            logger?.Warning($"{context}: still no network link after {maxWait.TotalSeconds:F0}s — attempting anyway.");
+        }
+    }
+}

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

@@ -2,6 +2,7 @@ using System;
 using System.Linq;
 using System.Security.Cryptography.X509Certificates;
 using AutopilotMonitor.Agent.V2.Core.Logging;
+using AutopilotMonitor.Shared.Security;
 
 namespace AutopilotMonitor.Agent.V2.Core.Security
 {
@@ -23,6 +24,24 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
     /// <item><description><c>1.2.840.113556.1.5.284.2</c> → DeviceId (same encoding)</description></item>
     /// </list>
     /// <para>
+    /// The sibling <c>Microsoft Intune MDM Device CA</c> client cert (the one the agent authenticates
+    /// with, see <see cref="CertificateHelper"/>) carries a DIFFERENT OID arc that identifies the same
+    /// device from Intune's side. Observed on a live Intune-enrolled Windows 11 device:
+    /// </para>
+    /// <list type="table">
+    /// <listheader><term>OID</term><description>Meaning / encoding</description></listheader>
+    /// <item><term><c>1.2.840.113556.5.4</c></term><description>Intune DeviceId — BARE 16-byte GUID, no ASN.1 wrapper. Equals the cert's Subject CN.</description></item>
+    /// <item><term><c>1.2.840.113556.5.6</c></term><description>Intune AccountId (the Intune tenant account, NOT the Entra TenantId) — OCTET STRING GUID.</description></item>
+    /// <item><term><c>1.2.840.113556.5.14</c></term><description>Entra TenantId — OCTET STRING GUID. This is the one to compare against an Entra tenant id.</description></item>
+    /// <item><term><c>1.2.840.113556.5.11</c></term><description>Further GUID, OCTET STRING; semantics not documented by Microsoft and not verified here.</description></item>
+    /// <item><term><c>1.2.840.113556.5.15/.16/.17/.18/.19/.23</c></term><description>ASN.1 INTEGER enums (platform / enrollment flavour); semantics undocumented.</description></item>
+    /// </list>
+    /// <para>
+    /// The bare-GUID encoding of <c>1.2.840.113556.5.4</c> is why
+    /// <see cref="ParseGuidFromExtension"/> accepts an unwrapped 16-byte payload as well —
+    /// a strict OCTET-STRING parser silently returns <c>null</c> for that extension.
+    /// </para>
+    /// <para>
     /// All probes are best-effort: never throw, return <c>null</c> on miss, log a Warning
     /// with the specific failure reason (no matching cert / extension missing / ASN.1
     /// parse failed) so the agent log explains why the fallback didn't help.
@@ -31,8 +50,8 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
     public static class EntraDeviceCertHelper
     {
         public const string MsOrganizationAccessIssuer = "MS-Organization-Access";
-        public const string TenantIdOid = "1.2.840.113556.1.5.284.5";
-        public const string DeviceIdOid = "1.2.840.113556.1.5.284.2";
+        public const string TenantIdOid = MsDeviceCertificateOids.EntraJoinCertTenantIdOid;
+        public const string DeviceIdOid = MsDeviceCertificateOids.EntraJoinCertDeviceIdOid;
 
         /// <summary>
         /// Reads the Entra TenantId from the MS-Organization-Access device cert.
@@ -88,7 +107,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
                             continue;
                         }
 
-                        var guid = ParseGuidFromAsn1OctetString(ext.RawData);
+                        var guid = ParseGuidFromExtension(ext.RawData);
                         if (guid != null)
                         {
                             logger?.Info(
@@ -98,7 +117,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
                         }
 
                         logger?.Warning(
-                            $"EntraDeviceCertHelper: failed to parse {label} ASN.1 OCTET STRING in cert thumbprint={cert.Thumbprint} " +
+                            $"EntraDeviceCertHelper: failed to parse {label} GUID extension in cert thumbprint={cert.Thumbprint} " +
                             $"(raw {ext.RawData.Length} bytes); trying next candidate.");
                     }
 
@@ -116,46 +135,27 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
 
         /// <summary>
         /// Parses a 16-byte MS-format binary GUID (little-endian for the first three groups,
-        /// per <see cref="Guid(byte[])"/>) wrapped in an ASN.1 OCTET STRING. Handles both
-        /// short-form (<c>04 10 ...</c>) and long-form (<c>04 81 10 ...</c>) length encoding.
+        /// per <see cref="Guid(byte[])"/>) out of a certificate extension's raw value.
+        /// Two encodings are accepted, because Microsoft uses both:
+        /// <list type="number">
+        /// <item><description>Wrapped in an ASN.1 OCTET STRING — short-form (<c>04 10 ...</c>) or
+        /// long-form (<c>04 81 10 ...</c>) length. Used by the MS-Organization-Access TenantId/DeviceId
+        /// extensions and by the Intune <c>…5.6</c> / <c>…5.14</c> extensions.</description></item>
+        /// <item><description>A bare, unwrapped 16-byte payload. Used by the Intune DeviceId extension
+        /// <c>1.2.840.113556.5.4</c> — verified against a live Intune MDM device cert, whose raw value is
+        /// exactly 16 bytes with no <c>04 10</c> prefix.</description></item>
+        /// </list>
+        /// A valid OCTET-STRING wrapper around 16 bytes is at least 18 bytes long, so the two forms
+        /// never collide for well-formed input. A 16-byte value is always read as the bare form —
+        /// a truncated wrapper of exactly that length is indistinguishable from it.
         /// Returns <c>null</c> on any structural mismatch — never throws.
+        /// <para>
+        /// The decoding itself lives in <see cref="MsDeviceCertificateOids.TryParseGuid"/> so the agent
+        /// and the backend's client-cert tenant binding share one implementation; this stays as the
+        /// agent-side nullable-returning façade.
+        /// </para>
         /// </summary>
-        public static Guid? ParseGuidFromAsn1OctetString(byte[] raw)
-        {
-            if (raw == null || raw.Length < 2) return null;
-
-            // ASN.1 OCTET STRING tag.
-            if (raw[0] != 0x04) return null;
-
-            int contentStart;
-            int contentLen;
-
-            if (raw[1] < 0x80)
-            {
-                // short-form length
-                contentLen = raw[1];
-                contentStart = 2;
-            }
-            else
-            {
-                // long-form length: lower 7 bits = number of following length bytes.
-                int numLenBytes = raw[1] & 0x7F;
-                if (numLenBytes < 1 || numLenBytes > 4 || raw.Length < 2 + numLenBytes) return null;
-
-                contentLen = 0;
-                for (int i = 0; i < numLenBytes; i++)
-                    contentLen = (contentLen << 8) | raw[2 + i];
-
-                contentStart = 2 + numLenBytes;
-            }
-
-            if (contentLen != 16 || contentStart + 16 > raw.Length) return null;
-
-            var guidBytes = new byte[16];
-            Buffer.BlockCopy(raw, contentStart, guidBytes, 0, 16);
-
-            try { return new Guid(guidBytes); }
-            catch { return null; }
-        }
+        public static Guid? ParseGuidFromExtension(byte[] raw) =>
+            MsDeviceCertificateOids.TryParseGuid(raw, out var value) ? value : (Guid?)null;
     }
 }

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

@@ -19,7 +19,15 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
     /// <b>Retry contract (V1 <c>MonitoringService.RegisterSessionAsync</c>):</b>
     /// </para>
     /// <list type="bullet">
-    ///   <item>5 attempts total; between attempts <c>2^attempt</c> seconds (2s, 4s, 8s, 16s).</item>
+    ///   <item>Boot-time NIC grace before the first attempt: when no network link is up yet
+    ///     (BootTrigger relaunch after a mid-enrollment reboot, Wi-Fi still associating) wait
+    ///     up to <see cref="NetworkLinkWaitMax"/> for the link-level signal. Best-effort and
+    ///     free on the normal path — a live link returns immediately.</item>
+    ///   <item>6 attempts total; between attempts <c>2^attempt</c> seconds (2s, 4s, 8s, 16s, 32s —
+    ///     ~62 s of backoff). The original V1 budget of 5 attempts / ~30 s was too short for the
+    ///     reboot-relaunch on Wi-Fi kiosks: the relaunched agent gave up and exited 7 before
+    ///     the network came back, leaving the session silent forever (tenant aebdce78,
+    ///     2026-08-23 audit). A relaunch that misses here is lost — nothing retries later.</item>
     ///   <item>On <c>response.Success == true</c> → stop immediately.</item>
     ///   <item>On <see cref="BackendAuthException"/> (401/403) → report to the
     ///     <see cref="AuthFailureTracker"/> (which fires the first-failure distress + may trip
@@ -37,16 +45,23 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
     /// </summary>
     public static class SessionRegistrationHelper
     {
-        private const int MaxAttempts = 5;
+        private const int MaxAttempts = 6;
+
+        /// <summary>Upper bound for the pre-registration link wait (see class remarks).</summary>
+        internal static readonly TimeSpan NetworkLinkWaitMax = TimeSpan.FromSeconds(15);
 
         /// <summary>
-        /// Executes the 5-retry register-session handshake against the backend. Returns the
+        /// Executes the retrying register-session handshake against the backend. Returns the
         /// response + classified outcome. Never throws.
         /// </summary>
         /// <param name="backoffDelay">
         /// Optional delay-provider used by tests to avoid real-time waits. Production callers
         /// pass <c>null</c> → V1 parity <c>Task.Delay(2^attempt * 1000)</c> between attempts.
         /// </param>
+        /// <param name="networkLinkWait">
+        /// Optional pre-registration link wait used by tests. Production callers pass
+        /// <c>null</c> → <see cref="WaitForNetworkLinkAsync"/> (bounded by <see cref="NetworkLinkWaitMax"/>).
+        /// </param>
         public static async Task<SessionRegistrationResult> RegisterWithRetryAsync(
             BackendApiClient apiClient,
             AgentConfiguration agentConfig,
@@ -56,14 +71,19 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
             EmergencyReporter emergencyReporter = null,
             Func<int, Task> backoffDelay = null,
             Func<Exception, Task> onTerminalTransportFailure = null,
-            (string Manufacturer, string Model, string SerialNumber)? deviceHardware = null)
+            (string Manufacturer, string Model, string SerialNumber)? deviceHardware = null,
+            Func<AgentLogger, Task> networkLinkWait = null)
         {
             if (apiClient == null) throw new ArgumentNullException(nameof(apiClient));
             if (agentConfig == null) throw new ArgumentNullException(nameof(agentConfig));
             if (logger == null) throw new ArgumentNullException(nameof(logger));
 
             backoffDelay ??= DefaultBackoffDelay;
+            networkLinkWait ??= WaitForNetworkLinkAsync;
             var registration = BuildRegistration(agentConfig, agentVersion, deviceHardware);
+
+            try { await networkLinkWait(logger).ConfigureAwait(false); }
+            catch (Exception ex) { logger.Debug($"Pre-registration network wait failed ({ex.Message}) — registering anyway."); }
             string lastError = null;
             Exception lastException = null;
 
@@ -113,7 +133,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
 
                 if (attempt < MaxAttempts)
                 {
-                    var delaySeconds = (int)Math.Pow(2, attempt); // 2, 4, 8, 16
+                    var delaySeconds = (int)Math.Pow(2, attempt); // 2, 4, 8, 16, 32
                     logger.Info($"Retrying session registration in {delaySeconds}s");
                     await backoffDelay(attempt).ConfigureAwait(false);
                 }
@@ -143,6 +163,14 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
         private static Task DefaultBackoffDelay(int attempt)
             => Task.Delay(((int)Math.Pow(2, attempt)) * 1000);
 
+        /// <summary>
+        /// Boot-time NIC grace before the first registration attempt — shared polling loop in
+        /// <see cref="NetworkLinkWait"/>, bounded by <see cref="NetworkLinkWaitMax"/>. Probe
+        /// errors end the wait, never the registration.
+        /// </summary>
+        internal static Task WaitForNetworkLinkAsync(AgentLogger logger)
+            => NetworkLinkWait.WaitAsync(logger, NetworkLinkWaitMax, "Session registration");
+
         private static SessionRegistration BuildRegistration(
             AgentConfiguration agentConfig,
             string agentVersion,
@@ -194,7 +222,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Security
         /// <summary>401/403 from the backend — cert/token rejected. No retry was attempted.</summary>
         AuthFailed = 1,
 
-        /// <summary>All 5 attempts failed with non-auth errors (network, 5xx, malformed response).</summary>
+        /// <summary>All attempts failed with non-auth errors (network, 5xx, malformed response).</summary>
         Failed = 2,
     }
 

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

@@ -56,6 +56,13 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
     {
         private const string SourceLabel = "ImeLogTracker";
 
+        /// <summary>Timeline label: IME wrote its "Completed user session" line and no required user-scope app was pending.</summary>
+        internal const string EvidenceUserSessionCompleted = "user_session_completed";
+
+        /// <summary>Timeline label: IME reported zero targeted apps for the user session, so it never wrote a
+        /// "Completed user session" line at all. The user phase had nothing to enforce.</summary>
+        internal const string EvidenceZeroUserApps = "zero_user_apps_enumerated";
+
         private readonly ImeLogTracker _tracker;
         private readonly ISignalIngressSink _ingress;
         private readonly IClock _clock;
@@ -591,9 +598,24 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
 
             var now = ResolveOccurredAt(out var derivedFromClock, out var rawSourceTs);
 
+            // Which observation carried the completion. The zero-app variant matters to whoever
+            // reads the timeline later: the verdict then rests on IME reporting that this user
+            // had no targeted apps at all, not on watching enforcement finish.
+            var zeroApps = _tracker.UserSessionZeroAppsObserved;
+            var evidenceLabel = zeroApps ? EvidenceZeroUserApps : EvidenceUserSessionCompleted;
+            var summary = zeroApps
+                ? "IME enumerated the user session with zero targeted apps"
+                : "IME user session completed (no required user-scope apps pending)";
+            var message = zeroApps
+                ? "IME enumerated the user session with zero targeted apps - no user-phase enforcement to wait for"
+                : "IME user session completed";
+
             var derivationInputs = new Dictionary<string, string>(StringComparer.Ordinal)
             {
-                ["detectionSource"] = "IME log pattern (UserSessionCompleted)",
+                ["detectionSource"] = zeroApps
+                    ? "IME log pattern (UserSessionZeroApps)"
+                    : "IME log pattern (UserSessionCompleted)",
+                ["evidence"] = evidenceLabel,
             };
             TagDerivedTimestamp(derivationInputs, derivedFromClock, rawSourceTs);
 
@@ -604,12 +626,13 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
                 evidence: new Evidence(
                     kind: EvidenceKind.Derived,
                     identifier: "ime-log-tracker-v1",
-                    summary: "IME user session completed (no required user-scope apps pending)",
+                    summary: summary,
                     derivationInputs: derivationInputs));
 
             var data = new Dictionary<string, string>(StringComparer.Ordinal)
             {
                 ["detectedAt"] = now.ToString("o"),
+                ["evidence"] = evidenceLabel,
             };
             var patternId = _tracker.LastMatchedPatternId;
             if (!string.IsNullOrEmpty(patternId))
@@ -619,12 +642,12 @@ namespace AutopilotMonitor.Agent.V2.Core.SignalAdapters
             _post.Emit(
                 eventType: SharedEventTypes.ImeUserSessionCompleted,
                 source: SourceLabel,
-                message: "IME user session completed",
+                message: message,
                 data: data,
                 occurredAtUtc: now);
 
             // PR3-D4: terminal-ish lifecycle marker — INFO so it's visible at the default level.
-            _logger?.Info($"ImeAdapter: user session completed -> posting ImeUserSessionCompleted (patternId={patternId ?? "(none)"})");
+            _logger?.Info($"ImeAdapter: user session completed ({evidenceLabel}) -> posting ImeUserSessionCompleted (patternId={patternId ?? "(none)"})");
         }
 
         /// <summary>

src/Agent/AutopilotMonitor.Agent.V2/Program.cs

@@ -274,6 +274,15 @@ namespace AutopilotMonitor.Agent.V2
             {
                 return registration.ExitCode;
             }
+
+            // Phase 6.5 (config self-heal): when the Phase 4 fetch ran before the network link
+            // was up, registration has just proven the backend reachable — retry the fetch once
+            // so the session runs on the live tenant config instead of cache/defaults.
+            // See AgentRuntimeConfig.TryRecoverAfterRegistration for what is deliberately
+            // NOT re-run here (kill-signal check, endpoint migration).
+            Runtime.AgentRuntimeConfig.TryRecoverAfterRegistration(
+                bootstrap.AgentConfig, auth, runtimeConfig, GetAgentVersion(), consoleMode, logger);
+
             // Phase 7+8 (orchestrator + shutdown lifecycle + onIngressReady wiring +
             // initial signal posts + analyzer/probes startup + shutdown.Wait + finally)
             // is encapsulated in AgentRuntimeHost. See Runtime/AgentRuntimeHost.cs.

src/Agent/AutopilotMonitor.Agent.V2/Runtime/AgentRuntimeConfig.cs

@@ -69,6 +69,63 @@ namespace AutopilotMonitor.Agent.V2.Runtime
                 }
             }
 
+            var configMergeResult = ApplyFetchedConfig(agentConfig, remoteConfig, auth, agentVersion, consoleMode, logger);
+
+            return new RuntimeConfigBundle(remoteConfigService, remoteConfig, configMergeResult, pendingMigrationTarget: null);
+        }
+
+        /// <summary>
+        /// Phase 6.5 — config self-heal. When the Phase 4 fetch ran before the network link was
+        /// up (agent relaunched at boot, Wi-Fi still associating — tenant aebdce78 audit 2026-08-24:
+        /// ~50% of sessions ran on ConfigVersion=0 exactly this way), a successful session
+        /// registration has just proven the backend reachable. Retry the fetch once and apply
+        /// the live tenant config instead of stranding the whole session on cache/defaults.
+        /// <para>
+        /// Deliberately NOT re-run here: the kill-signal check (post-registration kills arrive
+        /// via the telemetry-channel control signals on every upload response) and endpoint
+        /// migration (registration already succeeded against the current base URL; a migration
+        /// signal is honoured at the next agent start). Never throws — recovery is strictly
+        /// best-effort on top of an already-degraded-but-working fallback.
+        /// </para>
+        /// </summary>
+        public static void TryRecoverAfterRegistration(
+            AgentConfiguration agentConfig,
+            BackendAuthBundle auth,
+            RuntimeConfigBundle bundle,
+            string agentVersion,
+            bool consoleMode,
+            AgentLogger logger)
+        {
+            try
+            {
+                var service = bundle?.RemoteConfigService;
+                if (service == null) return;
+
+                var recovered = service.TryRecoverConfigAsync().GetAwaiter().GetResult();
+                if (recovered == null) return;
+
+                var mergeResult = ApplyFetchedConfig(agentConfig, recovered, auth, agentVersion, consoleMode, logger);
+                bundle.ApplyRecovered(recovered, mergeResult);
+            }
+            catch (Exception ex)
+            {
+                logger.Warning($"Post-registration config recovery failed: {ex.Message}");
+            }
+        }
+
+        /// <summary>
+        /// Applies a live-fetched config to the runtime: merge, tracker ceilings, log level,
+        /// integrity SHA + check, bootstrap-config cleanup. Shared by the Phase 4 startup path
+        /// and the post-registration recovery path so both apply identical semantics.
+        /// </summary>
+        private static RemoteConfigMergeResult ApplyFetchedConfig(
+            AgentConfiguration agentConfig,
+            AgentConfigResponse remoteConfig,
+            BackendAuthBundle auth,
+            string agentVersion,
+            bool consoleMode,
+            AgentLogger logger)
+        {
             // Project remote tenant-controlled knobs onto the runtime AgentConfiguration so that
             // downstream consumers (CleanupService, SummaryDialogLauncher, StartupEnvironmentProbes,
             // DiagnosticsPackageService, EnrollmentTerminationHandler, the logger, the watchdog)
@@ -91,7 +148,7 @@ namespace AutopilotMonitor.Agent.V2.Runtime
 
             TryCleanupBootstrapConfig(agentConfig, agentVersion, logger);
 
-            return new RuntimeConfigBundle(remoteConfigService, remoteConfig, configMergeResult, pendingMigrationTarget: null);
+            return configMergeResult;
         }
 
         /// <summary>
@@ -175,10 +232,21 @@ namespace AutopilotMonitor.Agent.V2.Runtime
     internal sealed class RuntimeConfigBundle
     {
         public RemoteConfigService RemoteConfigService { get; }
-        public AgentConfigResponse RemoteConfig { get; }
-        public RemoteConfigMergeResult MergeResult { get; }
+        public AgentConfigResponse RemoteConfig { get; private set; }
+        public RemoteConfigMergeResult MergeResult { get; private set; }
         public string PendingMigrationTarget { get; }
 
+        /// <summary>
+        /// Swaps in the config recovered by <see cref="AgentRuntimeConfig.TryRecoverAfterRegistration"/>
+        /// so Phase 7+8 (orchestrator, collectors, gather rules, IME patterns, the
+        /// unrestricted-mode audit) run on the live tenant config instead of the fallback.
+        /// </summary>
+        public void ApplyRecovered(AgentConfigResponse recoveredConfig, RemoteConfigMergeResult mergeResult)
+        {
+            RemoteConfig = recoveredConfig;
+            MergeResult = mergeResult;
+        }
+
         public RuntimeConfigBundle(
             RemoteConfigService remoteConfigService,
             AgentConfigResponse remoteConfig,

src/Agent/AutopilotMonitor.Agent.V2/Runtime/BackendSessionRegistration.cs

@@ -11,7 +11,7 @@ namespace AutopilotMonitor.Agent.V2.Runtime
 {
     /// <summary>
     /// Phase 6 of <see cref="Program"/>'s <c>RunAgent</c>: blocking
-    /// <c>POST /api/agent/register-session</c> with 5-retry exponential backoff before the
+    /// <c>POST /api/agent/register-session</c> with link-wait + 6-retry exponential backoff before the
     /// orchestrator starts. V1 parity (MonitoringService.RegisterSessionAsync) — without
     /// this call the backend's Sessions table never gets a row for this session, so
     /// <c>IncrementSessionEventCountAsync</c> / <c>UpdateSessionStatusAsync</c> silently

src/Agent/AutopilotMonitor.Agent.V2/Runtime/LifecycleEmitters.cs

@@ -52,6 +52,7 @@ namespace AutopilotMonitor.Agent.V2.Runtime
                 // prior to this, ConfigVersion=0 was only deducible by inferring from
                 // the absence of tenant-controlled knobs in the DataJson.
                 var outcome = remoteConfigService?.LastFetchOutcome ?? RemoteConfigFetchOutcome.NotAttempted;
+                var recovered = remoteConfigService?.RecoveredAfterFetchFailure == true;
                 var configVersion = remoteConfigService?.CurrentConfig?.ConfigVersion ?? 0;
 
                 var data = new Dictionary<string, object>
@@ -66,8 +67,12 @@ namespace AutopilotMonitor.Agent.V2.Runtime
                     { "diagnosticsUploadMode", agentConfig.DiagnosticsUploadMode ?? "Off" },
                     { "previousExitType", previousExit?.ExitType ?? "unknown" },
                     { "unrestrictedMode", agentConfig.UnrestrictedMode },
+                    // remoteConfigFetched answers "is the agent running on a live-fetched
+                    // config" — true also when the post-registration recovery replaced a
+                    // failed startup fetch (remoteConfigOutcome then still names the startup
+                    // failure; remoteConfigRecovered marks the heal).
                     { "configVersion", configVersion },
-                    { "remoteConfigFetched", outcome == RemoteConfigFetchOutcome.Succeeded },
+                    { "remoteConfigFetched", outcome == RemoteConfigFetchOutcome.Succeeded || recovered },
                     { "remoteConfigOutcome", outcome.ToString() },
                     // WinRT OOBE state sampled at this start. Together with previousExitType
                     // this makes late starts provable: "completed" on a first_run means OOBE
@@ -75,6 +80,9 @@ namespace AutopilotMonitor.Agent.V2.Runtime
                     { "oobeStateAtAgentStart", OobeStateReader.Read() },
                 };
 
+                if (recovered)
+                    data["remoteConfigRecovered"] = true;
+
                 if (!string.IsNullOrEmpty(previousExit?.CrashExceptionType))
                     data["previousCrashException"] = previousExit.CrashExceptionType;
 
@@ -135,12 +143,25 @@ namespace AutopilotMonitor.Agent.V2.Runtime
                     ? $"Remote config fetch failed after {remoteConfigService.LastFetchAttempts} attempt(s); using cached config."
                     : $"Remote config fetch failed after {remoteConfigService.LastFetchAttempts} attempt(s); running with built-in defaults (ConfigVersion=0).";
 
+                // Post-registration recovery (Phase 6.5) replaced the fallback with a live
+                // fetch before this emitter ran: keep the failure details (the network-late
+                // condition is worth surfacing) but mark the heal and demote to Info — the
+                // session is NOT degraded.
+                var recovered = remoteConfigService.RecoveredAfterFetchFailure;
+                if (recovered)
+                {
+                    var recoveredVersion = remoteConfigService.CurrentConfig?.ConfigVersion ?? 0;
+                    data["recovered"] = true;
+                    data["recoveredConfigVersion"] = recoveredVersion;
+                    msg += $" Recovered after session registration (ConfigVersion={recoveredVersion}).";
+                }
+
                 post.Emit(new EnrollmentEvent
                 {
                     SessionId = agentConfig.SessionId,
                     TenantId = agentConfig.TenantId,
                     EventType = SharedConstants.EventTypes.RemoteConfigFetchFailed,
-                    Severity = EventSeverity.Warning,
+                    Severity = recovered ? EventSeverity.Info : EventSeverity.Warning,
                     Source = "Agent",
                     Phase = EnrollmentPhase.Unknown,
                     Message = msg,

src/Agent/AutopilotMonitor.DecisionCore.Tests/AdvisoryCompletionGateHoldingTests.cs

@@ -0,0 +1,242 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using AutopilotMonitor.DecisionCore.Engine;
+using AutopilotMonitor.DecisionCore.Signals;
+using AutopilotMonitor.DecisionCore.State;
+using Xunit;
+
+namespace AutopilotMonitor.DecisionCore.Tests
+{
+    /// <summary>
+    /// Sessions 81daa77f / 75d6ae8e (2026-08-25) — the 30-min <c>advisory_completion</c> window
+    /// overran a completion gate that was deliberately holding the session. The RealmJoin gate
+    /// blocks completion until phase 110 and carries its own bounded resolution (60 min from
+    /// detection, re-armed on deployment activity, hard-capped at 4 h), so on every device whose
+    /// first deployment outlives 30 minutes the enrollment was failed while the engine own
+    /// <c>completion_waiting</c> still read "waiting on: realmjoin_resolution". 75d6ae8e
+    /// installed 25 RealmJoin packages successfully inside the window, the last one 2 min before
+    /// the failure; the customer tenant hit this on 10 of 30 enrollments in one week.
+    /// <para>
+    /// The guard re-arms while the gate holds. Convergence comes from the gate itself: Resolved,
+    /// FirstDeploymentIncomplete and Timeout all set <c>RealmJoinFacts.Outcome</c>, which opens
+    /// <c>RealmJoinGateOpen</c> and stops the guard applying.
+    /// </para>
+    /// </summary>
+    public sealed class AdvisoryCompletionGateHoldingTests
+    {
+        private static readonly DateTime T0 = new DateTime(2026, 8, 25, 6, 10, 0, DateTimeKind.Utc);
+
+        private static DecisionSignal MakeSignal(
+            long ordinal,
+            DecisionSignalKind kind,
+            DateTime occurredAtUtc,
+            IReadOnlyDictionary<string, string>? payload = null)
+        {
+            return new DecisionSignal(
+                sessionSignalOrdinal: ordinal,
+                sessionTraceOrdinal: ordinal,
+                kind: kind,
+                kindSchemaVersion: 1,
+                occurredAtUtc: occurredAtUtc,
+                sourceOrigin: "test",
+                evidence: new Evidence(EvidenceKind.Synthetic, $"{kind}-{ordinal}", "test"),
+                payload: payload);
+        }
+
+        private static DecisionSignal DeadlineFired(long ordinal, DateTime occurredAtUtc, string deadlineName) =>
+            MakeSignal(ordinal, DecisionSignalKind.DeadlineFired, occurredAtUtc,
+                new Dictionary<string, string> { [SignalPayloadKeys.Deadline] = deadlineName });
+
+        private static ActiveDeadline? FindDeadline(DecisionState state, string name) =>
+            state.Deadlines.FirstOrDefault(d => d.Name == name);
+
+        private static DecisionEffect SingleTimelineEffect(DecisionStep step, string eventType) =>
+            step.Effects.Single(e =>
+                e.Kind == DecisionEffectKind.EmitEventTimelineEntry
+                && e.Parameters != null
+                && e.Parameters.TryGetValue("eventType", out var et)
+                && et == eventType);
+
+        /// <summary>
+        /// Replays the shape both field sessions shared: Classic user-driven, full ESP, Hello
+        /// policy enabled, AccountSetup entered, then a guard-blocked post-AccountSetup ESP exit
+        /// (Shell-Core 62407, "Exiting page normally", errorCode 0) that arms the 30-min window.
+        /// The ESP registry froze at "1 of 5" and the device had zero user-targeted Intune apps,
+        /// so no arm of ShouldTransitionToAwaitingHello could ever open.
+        /// </summary>
+        private static DecisionState SetupEspExitDeadEnd(DecisionEngine engine)
+        {
+            var state = DecisionState.CreateInitial("sess-81daa77f", "tenant-81daa77f", T0);
+            state = engine.Reduce(state, MakeSignal(0, DecisionSignalKind.SessionStarted, T0)).NewState;
+            state = engine.Reduce(state, MakeSignal(
+                5, DecisionSignalKind.EspConfigDetected, T0.AddMinutes(1),
+                new Dictionary<string, string>
+                {
+                    [SignalPayloadKeys.SkipUserEsp] = "false",
+                    [SignalPayloadKeys.SkipDeviceEsp] = "false",
+                    [SignalPayloadKeys.EspAllowContinueAnyway] = "false",
+                })).NewState;
+            state = engine.Reduce(state, MakeSignal(
+                10, DecisionSignalKind.EspPhaseChanged, T0.AddMinutes(2),
+                new Dictionary<string, string> { [SignalPayloadKeys.EspPhase] = "DeviceSetup" })).NewState;
+            state = engine.Reduce(state, MakeSignal(
+                20, DecisionSignalKind.EspPhaseChanged, T0.AddMinutes(2.5),
+                new Dictionary<string, string> { [SignalPayloadKeys.EspPhase] = "AccountSetup" })).NewState;
+
+            // Guard-blocked post-AccountSetup exit — arms advisory_completion at exit + 30 min.
+            state = engine.Reduce(state, MakeSignal(30, DecisionSignalKind.EspExiting, T0.AddMinutes(3))).NewState;
+            Assert.NotNull(FindDeadline(state, DeadlineNames.AdvisoryCompletion));
+            Assert.Null(state.Outcome);
+            return state;
+        }
+
+        /// <summary>Desktop (DAD-validated real user), RealmJoin detected, Hello provisioned.</summary>
+        private static DecisionState AddDesktopRealmJoinAndHello(DecisionEngine engine, DecisionState state)
+        {
+            state = engine.Reduce(state, MakeSignal(40, DecisionSignalKind.DesktopArrived, T0.AddMinutes(3.5))).NewState;
+            state = engine.Reduce(state, MakeSignal(
+                50, DecisionSignalKind.RealmJoinDetected, T0.AddMinutes(6),
+                new Dictionary<string, string> { [DecisionEngine.RealmJoinPayloadKeys.DeploymentPhase] = "0" })).NewState;
+            state = engine.Reduce(state, MakeSignal(
+                60, DecisionSignalKind.HelloResolved, T0.AddMinutes(14),
+                new Dictionary<string, string> { [SignalPayloadKeys.HelloOutcome] = "Success" })).NewState;
+            return state;
+        }
+
+        // ==================================================== the gate-holding guard ====
+
+        [Fact]
+        public void DeadlineFired_WhileRealmJoinGateHolds_RearmsInsteadOfFailing()
+        {
+            var engine = new DecisionEngine();
+            var state = SetupEspExitDeadEnd(engine);
+            state = AddDesktopRealmJoinAndHello(engine, state);
+
+            // Preconditions: the gate is closed and owns an armed deadline of its own.
+            Assert.False(DecisionEngine.RealmJoinGateOpen(state));
+            Assert.NotNull(FindDeadline(state, DeadlineNames.RealmJoinTimeout));
+
+            var fireAt = T0.AddMinutes(33);
+            var step = engine.Reduce(state, DeadlineFired(70, fireAt, DeadlineNames.AdvisoryCompletion));
+
+            Assert.True(step.Transition.Taken);
+            Assert.Null(step.NewState.Outcome);
+            Assert.NotEqual(SessionStage.Failed, step.NewState.Stage);
+
+            var rearmed = FindDeadline(step.NewState, DeadlineNames.AdvisoryCompletion);
+            Assert.NotNull(rearmed);
+            Assert.Equal(fireAt.AddMinutes(30), rearmed!.DueAtUtc);
+
+            var schedule = step.Effects.Single(e => e.Kind == DecisionEffectKind.ScheduleDeadline);
+            Assert.Equal(DeadlineNames.AdvisoryCompletion, schedule.Deadline!.Name);
+
+            var waiting = SingleTimelineEffect(step, "completion_waiting");
+            Assert.Contains("CompletionGateHolding", waiting.Parameters!["trigger"]);
+            Assert.Contains("realmjoin_resolution", waiting.Parameters!["missingPrerequisites"]);
+            Assert.Equal(rearmed.DueAtUtc.ToString("o"), waiting.Parameters!["resolutionDeadlineDueAtUtc"]);
+        }
+
+        [Fact]
+        public void Session81daa77f_RealmJoinResolvesAfterTheWindow_CompletesInsteadOfFailing()
+        {
+            // End-to-end proof against the real field shape: before this guard the session was
+            // stamped Failed at exit+30 min with esp_exit_without_completion_evidence, 34 min
+            // before the RealmJoin gate would have released it.
+            var engine = new DecisionEngine();
+            var state = SetupEspExitDeadEnd(engine);
+            state = AddDesktopRealmJoinAndHello(engine, state);
+
+            var fired = engine.Reduce(state, DeadlineFired(70, T0.AddMinutes(33), DeadlineNames.AdvisoryCompletion));
+            Assert.Null(fired.NewState.Outcome);
+
+            // RealmJoin reaches phase 110 — the gate opens and the session completes.
+            var resolved = engine.Reduce(fired.NewState, MakeSignal(
+                80, DecisionSignalKind.RealmJoinResolved, T0.AddMinutes(50),
+                new Dictionary<string, string> { [DecisionEngine.RealmJoinPayloadKeys.DeploymentPhase] = "110" }));
+
+            Assert.Equal(SessionStage.Finalizing, resolved.NewState.Stage);
+            Assert.Equal("Resolved", resolved.NewState.RealmJoinFacts.Outcome!.Value);
+
+            var grace = FindDeadline(resolved.NewState, DeadlineNames.FinalizingGrace);
+            Assert.NotNull(grace);
+            var completed = engine.Reduce(resolved.NewState,
+                DeadlineFired(90, grace!.DueAtUtc, DeadlineNames.FinalizingGrace));
+            Assert.Equal(SessionStage.Completed, completed.NewState.Stage);
+            Assert.Equal(SessionOutcome.EnrollmentComplete, completed.NewState.Outcome);
+        }
+
+        // ============================================================== convergence ====
+
+        [Fact]
+        public void DeadlineFired_AfterTheGateOpened_ResolvesNormallyAgain()
+        {
+            // Convergence proof: the guard is not an escape hatch. Once RealmJoin timed out the
+            // gate is open, the guard stops applying, and a session with no completion evidence
+            // fails exactly as before.
+            var engine = new DecisionEngine();
+            var state = SetupEspExitDeadEnd(engine);
+            state = engine.Reduce(state, MakeSignal(
+                50, DecisionSignalKind.RealmJoinDetected, T0.AddMinutes(6),
+                new Dictionary<string, string> { [DecisionEngine.RealmJoinPayloadKeys.DeploymentPhase] = "0" })).NewState;
+
+            var rjDeadline = FindDeadline(state, DeadlineNames.RealmJoinTimeout);
+            Assert.NotNull(rjDeadline);
+            state = engine.Reduce(state,
+                DeadlineFired(60, rjDeadline!.DueAtUtc, DeadlineNames.RealmJoinTimeout)).NewState;
+
+            Assert.True(DecisionEngine.RealmJoinGateOpen(state));
+            Assert.Null(FindDeadline(state, DeadlineNames.RealmJoinTimeout));
+
+            // No desktop, no Hello, no IME evidence — the genuine dead-end shape.
+            var step = engine.Reduce(state,
+                DeadlineFired(70, T0.AddMinutes(80), DeadlineNames.AdvisoryCompletion));
+
+            Assert.Equal(SessionStage.Failed, step.NewState.Stage);
+            Assert.Equal(SessionOutcome.EnrollmentFailed, step.NewState.Outcome);
+            var failed = SingleTimelineEffect(step, "enrollment_failed");
+            Assert.Equal("esp_exit_without_completion_evidence", failed.Parameters!["reason"]);
+        }
+
+        // =========================================================== mutation proofs ====
+
+        [Fact]
+        public void DeadlineFired_WithoutRealmJoin_StillFails()
+        {
+            // The guard must key on an actually-held gate, never on the absence of evidence.
+            var engine = new DecisionEngine();
+            var state = SetupEspExitDeadEnd(engine);
+
+            var step = engine.Reduce(state,
+                DeadlineFired(70, T0.AddMinutes(33), DeadlineNames.AdvisoryCompletion));
+
+            Assert.Equal(SessionStage.Failed, step.NewState.Stage);
+            Assert.Equal(SessionOutcome.EnrollmentFailed, step.NewState.Outcome);
+            var failed = SingleTimelineEffect(step, "enrollment_failed");
+            Assert.Equal("esp_exit_without_completion_evidence", failed.Parameters!["reason"]);
+        }
+
+        [Fact]
+        public void DeadlineFired_GateClosedButItsDeadlineNoLongerArmed_StillFails()
+        {
+            // A gate that can no longer resolve itself must not park the session forever. The
+            // hand-built state drops only the RealmJoin deadline, leaving the gate closed.
+            var engine = new DecisionEngine();
+            var state = SetupEspExitDeadEnd(engine);
+            state = AddDesktopRealmJoinAndHello(engine, state);
+
+            var builder = state.ToBuilder();
+            builder.CancelDeadline(DeadlineNames.RealmJoinTimeout);
+            state = builder.Build();
+
+            Assert.False(DecisionEngine.RealmJoinGateOpen(state));
+            Assert.Null(FindDeadline(state, DeadlineNames.RealmJoinTimeout));
+
+            var step = engine.Reduce(state,
+                DeadlineFired(70, T0.AddMinutes(33), DeadlineNames.AdvisoryCompletion));
+
+            Assert.Equal(SessionStage.Failed, step.NewState.Stage);
+            Assert.Equal(SessionOutcome.EnrollmentFailed, step.NewState.Outcome);
+        }
+    }
+}

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

@@ -1,3 +1,4 @@
+using System.Text.RegularExpressions;
 using AutopilotMonitor.Functions.Services;
 using AutopilotMonitor.Shared.Models;
 
@@ -172,7 +173,7 @@ public class BuiltInRulesTests
             "policiesDiscovered", "ignoreCompletedApp", "imeAgentVersion",
             "espTrackStatus", "updateName", "updateWin32AppState",
             "cancelStuckAndSetCurrent", "imeSessionChange", "imeImpersonation", "imeTokenFailure",
-            "enrollmentCompleted", "updateDoTelemetry", "scriptStarted",
+            "enrollmentCompleted", "userSessionZeroApps", "updateDoTelemetry", "scriptStarted",
             "scriptContext", "scriptExitCode", "scriptOutput", "scriptCompleted",
             "resetPlatformScriptContext",
             "healthScriptResult", "healthScriptDetectionResult",
@@ -201,4 +202,38 @@ public class BuiltInRulesTests
                 $"Pattern {pattern.PatternId} has invalid category '{pattern.Category}'");
         }
     }
+
+    /// <summary>
+    /// Sessions 81daa77f / 75d6ae8e (2026-08-25). IME writes this line and then returns before
+    /// its "Completed user session" statement, so on a device with no user-targeted Win32 apps
+    /// it is the ONLY evidence that user-phase enforcement finished. The exact wording is
+    /// pinned here against the real IME line (verified against decompiled IME 1.104.102.0,
+    /// ApplicationPoller and AppWorkloadAbstraction) because a silent drift would bring back
+    /// the 30-min esp_exit_without_completion_evidence false positive with no test failing.
+    /// </summary>
+    [Fact]
+    public void BuiltInImeLogPatterns_ZeroUserAppsPattern_MatchesTheRealImeLine()
+    {
+        var pattern = Assert.Single(
+            BuiltInImeLogPatterns.GetAll(),
+            p => p.PatternId == "IME-USER-SESSION-ZERO-APPS");
+
+        Assert.Equal("userSessionZeroApps", pattern.Action);
+        Assert.Equal("always", pattern.Category);
+
+        var regex = new Regex(pattern.Pattern, RegexOptions.CultureInvariant);
+
+        var real = "[Win32App] Get 0 apps for user session 1, user id = 4a1ba5db-44ed-4e69-8477-47ed88e76020";
+        var match = regex.Match(real);
+        Assert.True(match.Success, "zero-app user session line must match");
+        Assert.Equal("1", match.Groups["sessionId"].Value);
+        Assert.Equal("4a1ba5db-44ed-4e69-8477-47ed88e76020", match.Groups["userId"].Value);
+
+        // Must not swallow a session that DOES have apps — that would fabricate completion
+        // evidence while enforcement is still running.
+        Assert.DoesNotMatch(regex,
+            "[Win32App] Get 3 apps for user session 1, user id = 4a1ba5db-44ed-4e69-8477-47ed88e76020");
+        Assert.DoesNotMatch(regex,
+            "[Win32App] Get 10 apps for user session 1, user id = 4a1ba5db-44ed-4e69-8477-47ed88e76020");
+    }
 }

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

@@ -0,0 +1,358 @@
+using System.Net;
+using System.Security.Cryptography.X509Certificates;
+using AutopilotMonitor.Functions.Security;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.DataAccess;
+using AutopilotMonitor.Shared.Models;
+using AutopilotMonitor.Shared.Security;
+using Microsoft.Azure.Functions.Worker.Http;
+using Microsoft.Extensions.Caching.Memory;
+using Microsoft.Extensions.Logging;
+using Moq;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// CERT-TENANT-BINDING-SHADOW — stage 1 coverage.
+///
+/// The agent authenticates with the Intune MDM device certificate, whose chain is pinned to
+/// Microsoft's Intune roots. Those roots are shared by every Intune tenant, so a valid chain alone
+/// does not prove the caller belongs to the tenant it claims. The CA stamps the Entra tenant id
+/// into OID 1.2.840.113556.5.14, which lets the backend bind the two.
+///
+/// These tests pin three things:
+///   - the tenant id is actually extracted from a REAL field certificate (not a synthetic fixture),
+///   - the pure comparison covers every outcome the telemetry can report,
+///   - shadow mode observes without deciding: a cross-tenant certificate must still pass the
+///     certificate stage in stage 1, or we would be enforcing before we measured.
+/// </summary>
+public class CertTenantBindingTests
+{
+    private const string DeviceSamplePem = "device-cert-sample.pem";
+
+    /// <summary>
+    /// Tenant stamped into <c>device-cert-sample.pem</c> by the Intune MDM Device CA. The sample is
+    /// a public-key-only cert from an internal test device; the GUID is the test tenant's own id.
+    /// </summary>
+    private static readonly Guid SampleCertTenantId = Guid.Parse("b54dc1af-5320-4f60-b5d4-821e0cf2a359");
+
+    private static string SampleCertBase64()
+    {
+        var assemblyDir = Path.GetDirectoryName(typeof(CertTenantBindingTests).Assembly.Location)!;
+        var pem = File.ReadAllText(Path.Combine(assemblyDir, DeviceSamplePem));
+        using var cert = X509Certificate2.CreateFromPem(pem);
+        return Convert.ToBase64String(cert.Export(X509ContentType.Cert));
+    }
+
+    // ---------------------------------------------------------------- extraction
+
+    [Fact]
+    public void ValidateCertificate_RealIntuneCert_ExtractsEntraTenantId()
+    {
+        // The load-bearing test for the whole feature: if Microsoft ever stops stamping OID
+        // 1.2.840.113556.5.14, or changes its encoding, enforcement would lock every device out.
+        var result = CertificateValidator.ValidateCertificate(SampleCertBase64());
+
+        Assert.True(result.IsValid, $"Real device cert rejected: {result.ErrorMessage}");
+        Assert.Equal(CertTenantIdStatus.Present, result.CertTenantIdStatus);
+        Assert.Equal(SampleCertTenantId, result.CertTenantId);
+    }
+
+    [Fact]
+    public void ValidateCertificate_CachedResult_KeepsTenantId()
+    {
+        // The tenant id is a pure function of the cert and is cached with the validation result,
+        // so a cache hit must not silently drop it and turn every repeat request into
+        // ExtensionMissing.
+        var b64 = SampleCertBase64();
+
+        _ = CertificateValidator.ValidateCertificate(b64);
+        var second = CertificateValidator.ValidateCertificate(b64);
+
+        Assert.True(second.IsValid);
+        Assert.Equal(CertTenantIdStatus.Present, second.CertTenantIdStatus);
+        Assert.Equal(SampleCertTenantId, second.CertTenantId);
+    }
+
+    // ---------------------------------------------------------------- encoding
+
+    [Fact]
+    public void TryParseGuid_PinsRealFieldEncodings()
+    {
+        // Exact bytes lifted from the sample cert. OID 5.14 nests the GUID in an OCTET STRING,
+        // OID 5.4 stores the 16 bytes bare — both shapes must decode, and both are little-endian.
+        var wrapped = Convert.FromHexString("0410afc14db52053604fb5d4821e0cf2a359");
+        var bare = Convert.FromHexString("6701bc076190b743b2430bb6aeda736e");
+
+        Assert.True(MsDeviceCertificateOids.TryParseGuid(wrapped, out var tenantId));
+        Assert.Equal(SampleCertTenantId, tenantId);
+
+        Assert.True(MsDeviceCertificateOids.TryParseGuid(bare, out var deviceId));
+        // The Intune device id is also the certificate's Subject CN.
+        Assert.Equal(Guid.Parse("07bc0167-9061-43b7-b243-0bb6aeda736e"), deviceId);
+    }
+
+    [Fact]
+    public void TryParseGuid_AccountIdOidIsNotTheTenantId()
+    {
+        // Guards the mistake this feature was nearly built on: OID 5.6 carries the Intune ACCOUNT
+        // id, a different GUID. Comparing it against the tenant id would reject every request.
+        var accountIdExt = Convert.FromHexString("04103d85dede74e26b409d0d46b6fe65b440");
+
+        Assert.True(MsDeviceCertificateOids.TryParseGuid(accountIdExt, out var accountId));
+        Assert.NotEqual(SampleCertTenantId, accountId);
+    }
+
+    [Theory]
+    [InlineData("")]                                            // empty
+    [InlineData("0400")]                                        // OCTET STRING, zero length
+    [InlineData("020102")]                                      // not an OCTET STRING
+    [InlineData("0411afc14db52053604fb5d4821e0cf2a35900")]      // wrapper claims 17 bytes
+    [InlineData("0410afc14db52053604fb5d4821e0cf2a3")]          // wrapper claims 16, only 15 present
+    public void TryParseGuid_RejectsMalformed(string hex)
+    {
+        Assert.False(MsDeviceCertificateOids.TryParseGuid(Convert.FromHexString(hex), out _));
+    }
+
+    [Fact]
+    public void TryParseGuid_AnySixteenBytesDecodeAsBareGuid()
+    {
+        // Documents a deliberate limit rather than a gap: the bare form (OID 5.4) is exactly 16
+        // bytes with no tag, so a 16-byte input is indistinguishable from a truncated wrapper and
+        // is always read as a GUID. Harmless here — the value is only ever compared for equality
+        // against a known tenant id, never trusted as a claim in its own right.
+        Assert.True(MsDeviceCertificateOids.TryParseGuid(
+            Convert.FromHexString("0410afc14db52053604fb5d4821e0cf2"), out _));
+    }
+
+    [Fact]
+    public void TryParseGuid_RejectsNull()
+    {
+        Assert.False(MsDeviceCertificateOids.TryParseGuid(null!, out _));
+    }
+
+    // ---------------------------------------------------------------- comparison
+
+    [Fact]
+    public void Evaluate_MatchingTenant_ReturnsMatch()
+    {
+        var outcome = CertTenantBinding.Evaluate(
+            SampleCertTenantId, CertTenantIdStatus.Present, SampleCertTenantId.ToString());
+
+        Assert.Equal(CertTenantBinding.Outcome.Match, outcome);
+        Assert.False(CertTenantBinding.WouldRejectUnderEnforcement(outcome));
+    }
+
+    [Fact]
+    public void Evaluate_MatchingTenant_IsCaseInsensitive()
+    {
+        // Tenant ids travel as strings in headers/routes and casing must not decide access.
+        var outcome = CertTenantBinding.Evaluate(
+            SampleCertTenantId, CertTenantIdStatus.Present, SampleCertTenantId.ToString().ToUpperInvariant());
+
+        Assert.Equal(CertTenantBinding.Outcome.Match, outcome);
+    }
+
+    [Fact]
+    public void Evaluate_ForeignTenantCert_ReturnsMismatch()
+    {
+        // The attack this feature exists for: a valid Intune cert from the attacker's own tenant,
+        // replayed against a victim tenant whose device serial the attacker knows.
+        var outcome = CertTenantBinding.Evaluate(
+            SampleCertTenantId, CertTenantIdStatus.Present, "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee");
+
+        Assert.Equal(CertTenantBinding.Outcome.Mismatch, outcome);
+        Assert.True(CertTenantBinding.WouldRejectUnderEnforcement(outcome));
+    }
+
+    [Theory]
+    [InlineData(CertTenantIdStatus.ExtensionMissing, CertTenantBinding.Outcome.ExtensionMissing)]
+    [InlineData(CertTenantIdStatus.Unparseable, CertTenantBinding.Outcome.Unparseable)]
+    public void Evaluate_UndecodableCert_ReportsWhyWithoutClaimingMismatch(
+        CertTenantIdStatus status, string expected)
+    {
+        // These must stay distinguishable from Mismatch: they measure rollout readiness
+        // (how many field certs can be enforced against), not an attack.
+        var outcome = CertTenantBinding.Evaluate(null, status, SampleCertTenantId.ToString());
+
+        Assert.Equal(expected, outcome);
+        Assert.False(CertTenantBinding.WouldRejectUnderEnforcement(outcome));
+    }
+
+    [Fact]
+    public void Evaluate_PresentButNullGuid_FallsBackToExtensionMissing()
+    {
+        var outcome = CertTenantBinding.Evaluate(null, CertTenantIdStatus.Present, SampleCertTenantId.ToString());
+
+        Assert.Equal(CertTenantBinding.Outcome.ExtensionMissing, outcome);
+    }
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    [InlineData("not-a-guid")]
+    public void Evaluate_NonGuidRequestTenant_IsReportedSeparately(string? requestedTenantId)
+    {
+        var outcome = CertTenantBinding.Evaluate(
+            SampleCertTenantId, CertTenantIdStatus.Present, requestedTenantId);
+
+        Assert.Equal(CertTenantBinding.Outcome.RequestTenantNotAGuid, outcome);
+        Assert.False(CertTenantBinding.WouldRejectUnderEnforcement(outcome));
+    }
+
+    // ---------------------------------------------------------------- shadow mode
+
+    [Fact]
+    public async Task ValidateRequestAsync_ForeignTenantCert_IsObservedButNotRejected()
+    {
+        // Stage 1 contract: a cross-tenant certificate is logged and still passes the certificate
+        // stage. If this test ever fails, enforcement was switched on without the telemetry review.
+        var logger = new CapturingLogger();
+        var validator = BuildValidator(logger);
+
+        var result = await validator.ValidateRequestAsync(
+            BuildRequestWithCert(SampleCertBase64()),
+            tenantId: "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee");
+
+        // Deliberately accepted: stage 1 measures, it does not enforce. This assertion is the one
+        // that must be inverted (to a 403) when stage 2 turns enforcement on.
+        Assert.True(result.IsValid, $"Shadow mode changed the outcome: {result.ErrorMessage} / {result.Details}");
+        Assert.NotEqual(HttpStatusCode.Unauthorized, result.StatusCode);
+
+        var observed = Assert.Single(logger.Entries, e => e.Message.Contains("AgentCertTenantBinding"));
+        Assert.Equal(LogLevel.Warning, observed.Level);
+        Assert.Contains(CertTenantBinding.Outcome.Mismatch, observed.Message);
+        Assert.Contains("enforced=False", observed.Message);
+        Assert.Contains("wouldReject=True", observed.Message);
+    }
+
+    [Fact]
+    public async Task ValidateRequestAsync_MatchingTenantCert_IsAccepted()
+    {
+        var logger = new CapturingLogger();
+        var validator = BuildValidator(logger);
+
+        var result = await validator.ValidateRequestAsync(
+            BuildRequestWithCert(SampleCertBase64()),
+            tenantId: SampleCertTenantId.ToString());
+
+        Assert.True(result.IsValid, $"Matching-tenant request rejected: {result.ErrorMessage}");
+
+        // Match is deliberately NOT a trace line — worker-side LogInformation never reaches App
+        // Insights, so a Match trace would be dead code that silently costs us the denominator.
+        Assert.DoesNotContain(logger.Entries, e => e.Message.Contains("AgentCertTenantBinding"));
+    }
+
+    [Fact]
+    public async Task ValidateRequestAsync_StampsOutcomeOnTheRequestRow()
+    {
+        // The denominator lives here: RequestTelemetryMiddleware copies this item onto the request
+        // row, which already exists per request and is unsampled. Without it, "Match" is invisible
+        // and the shadow telemetry cannot say how often the binding held.
+        var items = new Dictionary<object, object>();
+        var validator = BuildValidator(new CapturingLogger());
+
+        var result = await validator.ValidateRequestAsync(
+            BuildRequestWithCert(SampleCertBase64(), items),
+            tenantId: SampleCertTenantId.ToString());
+
+        Assert.True(result.IsValid, $"Matching-tenant request rejected: {result.ErrorMessage}");
+        Assert.Equal(
+            CertTenantBinding.Outcome.Match,
+            Assert.Contains(CertTenantBinding.RequestItemKey, items));
+    }
+
+    [Fact]
+    public async Task ValidateRequestAsync_StampsMismatchOnTheRequestRowToo()
+    {
+        var items = new Dictionary<object, object>();
+        var validator = BuildValidator(new CapturingLogger());
+
+        await validator.ValidateRequestAsync(
+            BuildRequestWithCert(SampleCertBase64(), items),
+            tenantId: "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee");
+
+        Assert.Equal(
+            CertTenantBinding.Outcome.Mismatch,
+            Assert.Contains(CertTenantBinding.RequestItemKey, items));
+    }
+
+    // ---------------------------------------------------------------- harness
+
+    private static SecurityValidator BuildValidator(ILogger logger)
+    {
+        var configRepo = Mock.Of<IConfigRepository>();
+        var cache = new MemoryCache(new MemoryCacheOptions());
+
+        // AllowInsecureAgentRequests clears the "at least one device validator" gate so the request
+        // reaches — and passes — the certificate stage without needing Graph mocks.
+        var config = new TenantConfiguration
+        {
+            AllowInsecureAgentRequests = true,
+            ManufacturerWhitelist = "*",
+            ModelWhitelist = "*",
+        };
+
+        var configServiceMock = new Mock<TenantConfigurationService>(
+            configRepo, Mock.Of<ILogger<TenantConfigurationService>>(), cache)
+        { CallBase = false };
+        configServiceMock
+            .Setup(x => x.TryGetConfigurationAsync(It.IsAny<string>()))
+            .ReturnsAsync((config, true));
+
+        var adminConfigServiceMock = new Mock<AdminConfigurationService>(
+            configRepo, Mock.Of<ILogger<AdminConfigurationService>>(), cache)
+        { CallBase = false };
+        adminConfigServiceMock
+            .Setup(x => x.GetConfigurationAsync())
+            .ReturnsAsync(new AdminConfiguration());
+
+        return new SecurityValidator(
+            configService: configServiceMock.Object,
+            adminConfigService: adminConfigServiceMock.Object,
+            rateLimitService: new RateLimitService(cache, Mock.Of<ILogger<RateLimitService>>()),
+            autopilotDeviceValidator: null!,
+            corporateIdentifierValidator: null!,
+            logger: logger,
+            bootstrapSessionService: null,
+            deviceAssociationValidator: null);
+    }
+
+    private static HttpRequestData BuildRequestWithCert(
+        string certBase64,
+        IDictionary<object, object>? items = null)
+    {
+        var contextMock = new Mock<Microsoft.Azure.Functions.Worker.FunctionContext>();
+        contextMock.SetupGet(c => c.Items).Returns(items ?? new Dictionary<object, object>());
+        var reqMock = new Mock<HttpRequestData>(contextMock.Object);
+
+        var headers = new HttpHeadersCollection
+        {
+            { "X-ARR-ClientCert", certBase64 },
+            // The hardware whitelist stage runs after the certificate stage and rejects requests
+            // with no hardware headers, which would mask what these tests are actually asserting.
+            { "X-Device-Manufacturer", "Contoso" },
+            { "X-Device-Model", "TestBook" },
+            { "X-Device-SerialNumber", "TESTSERIAL01" },
+        };
+        reqMock.SetupGet(r => r.Headers).Returns(headers);
+        reqMock.SetupGet(r => r.Url).Returns(new Uri("https://example.invalid/api/telemetry"));
+
+        return reqMock.Object;
+    }
+
+    private sealed record LogEntry(LogLevel Level, string Message);
+
+    private sealed class CapturingLogger : ILogger
+    {
+        public List<LogEntry> Entries { get; } = new();
+
+        public IDisposable? BeginScope<TState>(TState state) where TState : notnull => null;
+
+        public bool IsEnabled(LogLevel logLevel) => true;
+
+        public void Log<TState>(LogLevel logLevel, EventId eventId, TState state, Exception? exception,
+            Func<TState, Exception?, string> formatter)
+            => Entries.Add(new LogEntry(logLevel, formatter(state, exception)));
+    }
+}

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

@@ -0,0 +1,99 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using AutopilotMonitor.Functions.Services;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// <see cref="TableStorageService.BuildCveIndexEntities"/> collapses report findings to one
+/// CveIndex entity per CVE. Production reports routinely list the same CVE under many
+/// findings (one RD CVE across 14 installed RD components), which previously produced
+/// 14 concurrent upserts racing on the same PK/RK.
+/// </summary>
+public class CveIndexEntityMergeTests
+{
+    private const string Tenant = "11111111-1111-1111-1111-111111111111";
+    private const string Session = "22222222-2222-2222-2222-222222222222";
+
+    private static Dictionary<string, object> Finding(string software, string risk, params Dictionary<string, object>[] vulns)
+        => new()
+        {
+            ["softwareName"] = software,
+            ["riskLevel"] = risk,
+            ["vulnerabilities"] = vulns.ToList(),
+        };
+
+    private static Dictionary<string, object> Vuln(string cve, double score, string severity, bool kev = false)
+        => new()
+        {
+            ["cveId"] = cve,
+            ["cvssScore"] = score,
+            ["cvssSeverity"] = severity,
+            ["isKev"] = kev,
+        };
+
+    [Fact]
+    public void SameCveAcrossManyFindings_CollapsesToOneEntity()
+    {
+        var findings = Enumerable.Range(0, 14)
+            .Select(i => Finding($"RD Component {i}", "high", Vuln("CVE-2025-29966", 8.8, "HIGH")))
+            .ToList();
+
+        var entities = TableStorageService.BuildCveIndexEntities(Tenant, Session, findings);
+
+        var e = Assert.Single(entities);
+        Assert.Equal($"{Tenant}_CVE-2025-29966", e.PartitionKey);
+        Assert.Equal(Session, e.RowKey);
+        Assert.Equal("RD Component 0", e.GetString("SoftwareName")); // tie keeps report order
+    }
+
+    [Fact]
+    public void Merge_HighestScoreWins_KevOrs_RiskMaxes()
+    {
+        var findings = new List<Dictionary<string, object>>
+        {
+            Finding("App A", "medium", Vuln("CVE-1", 5.0, "MEDIUM", kev: true)),
+            Finding("App B", "critical", Vuln("CVE-1", 9.1, "CRITICAL")),
+            Finding("App C", "low", Vuln("CVE-1", 3.0, "LOW")),
+        };
+
+        var e = Assert.Single(TableStorageService.BuildCveIndexEntities(Tenant, Session, findings));
+
+        Assert.Equal(9.1, e.GetDouble("CvssScore"));
+        Assert.Equal("CRITICAL", e.GetString("CvssSeverity"));
+        Assert.Equal("App B", e.GetString("SoftwareName"));
+        Assert.True(e.GetBoolean("IsKev"));
+        Assert.Equal("critical", e.GetString("OverallRisk"));
+    }
+
+    [Fact]
+    public void DistinctCves_StayDistinct_AndInvalidEntriesAreSkipped()
+    {
+        var findings = new List<Dictionary<string, object>>
+        {
+            Finding("App A", "high", Vuln("CVE-1", 7.0, "HIGH"), Vuln("CVE-2", 6.0, "MEDIUM")),
+            Finding("App B", "high", new Dictionary<string, object> { ["cvssScore"] = 9.0 }), // no cveId
+            new() { ["softwareName"] = "No vulns" },
+        };
+
+        var entities = TableStorageService.BuildCveIndexEntities(Tenant, Session, findings);
+
+        Assert.Equal(2, entities.Count);
+        Assert.Contains(entities, x => x.GetString("CveId") == "CVE-1");
+        Assert.Contains(entities, x => x.GetString("CveId") == "CVE-2");
+    }
+
+    [Fact]
+    public void CveIdComparison_IsCaseInsensitive()
+    {
+        var findings = new List<Dictionary<string, object>>
+        {
+            Finding("A", "high", Vuln("CVE-2024-1", 7.0, "HIGH")),
+            Finding("B", "high", Vuln("cve-2024-1", 7.0, "HIGH")),
+        };
+
+        Assert.Single(TableStorageService.BuildCveIndexEntities(Tenant, Session, findings));
+    }
+}

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

@@ -127,4 +127,47 @@ public class DeviceAssociationValidatorTests
         Assert.True(result.IsValid);
     }
 
+    [Fact]
+    public void AsCacheHit_FlagsTheCopyAndLeavesTheCachedInstanceAlone()
+    {
+        // The cached instance is shared across every request for that serial. Marking it in place
+        // would make the first (real) lookup look cached too, and the shadow log gate would then
+        // drop the one line per lookup that carries the finding.
+        var cached = new DeviceAssociationResult
+        {
+            IsValid = true,
+            SerialNumber = Serial,
+            AssociationState = "preassociated",
+            DevicePreparationPolicyId = "11111111-2222-3333-4444-555555555555",
+            AssignedToUserPrincipalName = "someone@example.invalid",
+            ManagedDeviceId = "66666666-7777-8888-9999-000000000000",
+        };
+
+        var hit = cached.AsCacheHit();
+
+        Assert.True(hit.ServedFromCache);
+        Assert.False(cached.ServedFromCache);
+        Assert.NotSame(cached, hit);
+
+        // Every field the shadow log reports must survive the copy.
+        Assert.Equal(cached.IsValid, hit.IsValid);
+        Assert.Equal(cached.IsTransient, hit.IsTransient);
+        Assert.Equal(cached.SerialNumber, hit.SerialNumber);
+        Assert.Equal(cached.AssociationState, hit.AssociationState);
+        Assert.Equal(cached.DevicePreparationPolicyId, hit.DevicePreparationPolicyId);
+        Assert.Equal(cached.AssignedToUserPrincipalName, hit.AssignedToUserPrincipalName);
+        Assert.Equal(cached.ManagedDeviceId, hit.ManagedDeviceId);
+    }
+
+    [Fact]
+    public void FreshlyParsedResult_IsNotFlaggedAsCached()
+    {
+        // A result straight from a Graph response must be loggable — that is the one line per lookup.
+        var body = $@"{{""value"":[{{""serialNumber"":""{Serial}"",""associationState"":""preassociated""}}]}}";
+
+        var result = DeviceAssociationValidator.ParseTenantAssociatedDevicesResponse(body, Serial);
+
+        Assert.False(result.ServedFromCache);
+    }
+
 }

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

@@ -163,6 +163,12 @@ public class EnrollmentTimeoutClassifierTests
     // -------- ClassifyTimedOutSession --------
 
     private static (SessionStatus, string) Classify(IReadOnlyList<EnrollmentEvent> events, double hoursSinceStart = 6, int grace = 72)
+    {
+        var (status, reason, _) = ClassifyWithRule(events, hoursSinceStart, grace);
+        return (status, reason);
+    }
+
+    private static (SessionStatus, string, string) ClassifyWithRule(IReadOnlyList<EnrollmentEvent> events, double hoursSinceStart = 6, int grace = 72)
     {
         var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
         var now = Start.AddHours(hoursSinceStart);
@@ -220,7 +226,7 @@ public class EnrollmentTimeoutClassifierTests
             Esp(DeviceSetup44), Esp(AccountSetup15),
             Evt("desktop_arrived"), Evt("hello_provisioning_completed"), Evt("realmjoin_detected"),
         });
-        var (status, reason) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+        var (status, reason, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
             rollup, Start, now, graceHours: 72, lastEventAtUtc: lastEvent);
 
         Assert.Equal(SessionStatus.Succeeded, status);
@@ -481,8 +487,9 @@ public class EnrollmentTimeoutClassifierTests
     {
         var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
         var now = Start.AddHours(hoursSinceStart);
-        return EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+        var (status, reason, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
             rollup, Start, now, grace, isPreProvisioned: true, resumedAt: Resumed);
+        return (status, reason);
     }
 
     /// <summary>The fairstone event shape: Part 1 sealed, Part 2 resumed, nobody signed in.</summary>
@@ -611,8 +618,89 @@ public class EnrollmentTimeoutClassifierTests
         {
             Esp(DeviceSetup44), Evt("whiteglove_complete"), Esp(AccountSetup05),
         });
-        var (status, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+        var (status, _, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
             rollup, Start, Start.AddHours(6), 51, isPreProvisioned: true, resumedAt: null);
         Assert.Equal(SessionStatus.Succeeded, status);
     }
+
+    // -------- Self-deploying profile gate (kiosk tenant aebdce78, audit 2026-08-23) --------
+
+    private static (SessionStatus, string) ClassifySelfDeploying(
+        IReadOnlyList<EnrollmentEvent> events, double hoursSinceStart = 3, int grace = 51)
+    {
+        var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
+        var (status, reason, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+            rollup, Start, Start.AddHours(hoursSinceStart), grace,
+            lastEventAtUtc: Start.AddMinutes(4), isSelfDeployingProfile: true);
+        return (status, reason);
+    }
+
+    // Session 195593e2 replay: DeviceSetup 4/4 (fallback-confirmed), AccountSetup registry
+    // at 0/5 (the IME false positive — user ESP never runs on this profile), Hello disabled,
+    // SkipUser=True, then silence after the post-ESP reboot. No user will ever sign in.
+    private static readonly EnrollmentEvent[] KioskSilentStream =
+    {
+        Esp(DeviceSetup44), Esp(AccountSetup05), Esp(DeviceSetupFallback),
+        HelloPolicy(enabled: false), EspConfig(skipUser: true),
+    };
+
+    [Fact]
+    public void IsSelfDeployingProvisioned_requires_flag_device_setup_and_no_failure()
+    {
+        var provisioned = EnrollmentTimeoutClassifier.ExtractRollup(KioskSilentStream);
+        Assert.True(EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned(provisioned, isSelfDeployingProfile: true));
+        Assert.False(EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned(provisioned, isSelfDeployingProfile: false));
+
+        var beforeDeviceSetup = EnrollmentTimeoutClassifier.ExtractRollup(new[] { Esp(AccountSetup05) });
+        Assert.False(EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned(beforeDeviceSetup, isSelfDeployingProfile: true));
+
+        var failed = EnrollmentTimeoutClassifier.ExtractRollup(new[] { Esp(DeviceSetup44), Evt("enrollment_failed") });
+        Assert.False(EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned(failed, isSelfDeployingProfile: true));
+    }
+
+    [Fact]
+    public void Classify_self_deploying_silent_after_device_setup_reconciles_to_Succeeded_within_grace()
+    {
+        var (status, reason) = ClassifySelfDeploying(KioskSilentStream, hoursSinceStart: 3);
+        Assert.Equal(SessionStatus.Succeeded, status);
+        Assert.Contains("self-deploying profile", reason);
+        Assert.Contains("no user / Account Setup phase", reason);
+        Assert.Contains("Agent last reported 2026-07-06 15:04 UTC", reason);
+        Assert.DoesNotContain("Account Setup completed", reason);
+    }
+
+    [Fact]
+    public void Classify_self_deploying_silent_after_device_setup_reconciles_to_Succeeded_past_grace()
+    {
+        // Grace is a user-phase concept; it does not apply when there is no user phase.
+        var (status, _) = ClassifySelfDeploying(KioskSilentStream, hoursSinceStart: 60, grace: 51);
+        Assert.Equal(SessionStatus.Succeeded, status);
+    }
+
+    [Fact]
+    public void Classify_same_stream_without_self_deploying_flag_keeps_user_phase_verdicts()
+    {
+        // Guard against the gate leaking into user-driven sessions: the identical event shape
+        // on a user-driven profile is still "awaiting user" within grace / Incomplete past it.
+        var rollup = EnrollmentTimeoutClassifier.ExtractRollup(KioskSilentStream);
+        var (within, _, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(rollup, Start, Start.AddHours(3), 51);
+        var (past, _, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(rollup, Start, Start.AddHours(60), 51);
+        Assert.Equal(SessionStatus.AwaitingUser, within);
+        Assert.Equal(SessionStatus.Incomplete, past);
+    }
+
+    [Fact]
+    public void Classify_self_deploying_explicit_failure_stays_Failed()
+    {
+        var (status, _) = ClassifySelfDeploying(new[] { Esp(DeviceSetup44), Evt("enrollment_failed") });
+        Assert.Equal(SessionStatus.Failed, status);
+    }
+
+    [Fact]
+    public void Classify_self_deploying_silent_before_device_setup_is_Incomplete()
+    {
+        // The profile flag alone proves nothing — Device ESP must have finished.
+        var (status, _) = ClassifySelfDeploying(new[] { Esp(AccountSetup05), Evt("agent_started") });
+        Assert.Equal(SessionStatus.Incomplete, status);
+    }
 }

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

@@ -1,4 +1,4 @@
-using System.Text.RegularExpressions;
+using System.Text.RegularExpressions;
 using AutopilotMonitor.Shared.Models.Graph;
 
 namespace AutopilotMonitor.Functions.Tests;
@@ -63,6 +63,18 @@ public class GraphFeatureCatalogSyncTests
             scriptFeatures.Where(f => f != "All").OrderBy(f => f, StringComparer.OrdinalIgnoreCase));
     }
 
+    [Fact]
+    public void GrantScript_Help_DocumentsEveryFeature()
+    {
+        // The -Features help block is how an admin discovers what they can grant. A feature that
+        // is grantable but undocumented effectively does not exist for them.
+        var script = ReadGrantScript();
+        var help = script.Substring(0, script.IndexOf("$FeatureCatalog", StringComparison.Ordinal));
+
+        foreach (var feature in GraphFeatureCatalog.Features)
+            Assert.Contains(feature, help, StringComparison.Ordinal);
+    }
+
     private static string ReadGrantScript()
     {
         var path = Path.Combine(FindRepoRoot(), "scripts", "CustomerSetup", "Grant-AutopilotMonitorAddOn.ps1");

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

@@ -0,0 +1,223 @@
+using AutopilotMonitor.Functions.Security;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Tests for <see cref="IntuneDeviceBindingValidator"/> — the pure pieces: the managedDevice
+/// JSON-to-DTO mapping (incl. the exact-match guard on the returned id), the enrolledDateTime
+/// parsing that the enrollment-race measurement depends on, and the cache key shape.
+/// The HTTP/cache/retry resilience mirrors <see cref="CloudPcDeviceValidator"/> 1:1.
+/// </summary>
+public class IntuneDeviceBindingValidatorTests
+{
+    private const string TenantId = "11111111-1111-1111-1111-111111111111";
+    private const string DeviceId = "07bc0167-9061-43b7-b243-0bb6aeda736e";
+
+    // -- BuildCacheKey --
+
+    [Fact]
+    public void BuildCacheKey_StableShape()
+    {
+        var key = IntuneDeviceBindingValidator.BuildCacheKey(TenantId, DeviceId);
+        Assert.Equal($"intune-device-binding:{TenantId}:{DeviceId}", key);
+    }
+
+    [Fact]
+    public void BuildCacheKey_DistinctFromOtherValidatorKeys()
+    {
+        // A shared prefix would let a positive hit in one validator satisfy a lookup in another.
+        var key = IntuneDeviceBindingValidator.BuildCacheKey(TenantId, DeviceId);
+        Assert.StartsWith("intune-device-binding:", key);
+        Assert.DoesNotContain("cloudpc", key);
+        Assert.DoesNotContain("autopilot", key);
+    }
+
+    [Fact]
+    public void BuildCacheKey_IsTenantScoped()
+    {
+        // The whole point of the check is that the same device id must not resolve across
+        // tenants — a tenant-blind cache key would defeat it.
+        var a = IntuneDeviceBindingValidator.BuildCacheKey(TenantId, DeviceId);
+        var b = IntuneDeviceBindingValidator.BuildCacheKey("22222222-2222-2222-2222-222222222222", DeviceId);
+        Assert.NotEqual(a, b);
+    }
+
+    // -- ParseManagedDeviceResponse --
+
+    [Fact]
+    public void Parse_MatchingDevice_ReturnsMatchWithDiagnostics()
+    {
+        var body = $$"""
+        {
+          "id": "{{DeviceId}}",
+          "deviceName": "TEST-DEVICE-01",
+          "enrolledDateTime": "2026-08-24T15:20:00Z",
+          "azureADDeviceId": "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee",
+          "managementState": "managed"
+        }
+        """;
+
+        var result = IntuneDeviceBindingValidator.ParseManagedDeviceResponse(body, DeviceId);
+
+        Assert.Equal(IntuneDeviceBindingOutcome.Match, result.Outcome);
+        Assert.True(result.IsValid);
+        Assert.False(result.IsTransient);
+        Assert.Equal("TEST-DEVICE-01", result.DeviceName);
+        Assert.Equal("managed", result.ManagementState);
+        Assert.Equal(
+            new DateTimeOffset(2026, 8, 24, 15, 20, 0, TimeSpan.Zero),
+            result.EnrolledDateTime);
+    }
+
+    [Fact]
+    public void Parse_DifferentIdInBody_IsNotAMatch()
+    {
+        // Guards against a redirected or widened lookup being taken as confirmation of the id
+        // we actually asked about.
+        var body = """
+        { "id": "99999999-9999-9999-9999-999999999999", "deviceName": "OTHER" }
+        """;
+
+        var result = IntuneDeviceBindingValidator.ParseManagedDeviceResponse(body, DeviceId);
+
+        Assert.Equal(IntuneDeviceBindingOutcome.NotFound, result.Outcome);
+        Assert.False(result.IsValid);
+    }
+
+    [Fact]
+    public void Parse_IdCasingDiffers_StillMatches()
+    {
+        // Graph echoes GUIDs lowercase, but casing must never decide a security outcome.
+        var body = $$"""
+        { "id": "{{DeviceId.ToUpperInvariant()}}", "deviceName": "TEST-DEVICE-01" }
+        """;
+
+        var result = IntuneDeviceBindingValidator.ParseManagedDeviceResponse(body, DeviceId);
+
+        Assert.Equal(IntuneDeviceBindingOutcome.Match, result.Outcome);
+    }
+
+    [Theory]
+    [InlineData("")]
+    [InlineData("not json at all")]
+    [InlineData("{}")]
+    [InlineData("""{ "id": "" }""")]
+    [InlineData("""{ "deviceName": "no id field" }""")]
+    public void Parse_MalformedOrEmpty_IsNotFoundNotAnException(string body)
+    {
+        var result = IntuneDeviceBindingValidator.ParseManagedDeviceResponse(body, DeviceId);
+
+        Assert.Equal(IntuneDeviceBindingOutcome.NotFound, result.Outcome);
+        Assert.Equal(DeviceId, result.IntuneDeviceId);
+    }
+
+    [Fact]
+    public void Parse_MissingEnrolledDateTime_LeavesItNull()
+    {
+        var body = $$"""
+        { "id": "{{DeviceId}}", "deviceName": "TEST-DEVICE-01" }
+        """;
+
+        var result = IntuneDeviceBindingValidator.ParseManagedDeviceResponse(body, DeviceId);
+
+        Assert.Equal(IntuneDeviceBindingOutcome.Match, result.Outcome);
+        Assert.Null(result.EnrolledDateTime);
+    }
+
+    // -- ParseEnrolledDateTime --
+
+    [Fact]
+    public void ParseEnrolledDateTime_IntuneZeroDate_IsAbsenceNotADate()
+    {
+        // Intune returns 0001-01-01T00:00:00Z when it has no enrollment timestamp. Treating that
+        // as a real date would report an object age of ~2000 years and poison the race analysis.
+        Assert.Null(IntuneDeviceBindingValidator.ParseEnrolledDateTime("0001-01-01T00:00:00Z"));
+    }
+
+    [Theory]
+    [InlineData(null)]
+    [InlineData("")]
+    [InlineData("   ")]
+    [InlineData("not-a-date")]
+    public void ParseEnrolledDateTime_UnusableInput_IsNull(string? raw)
+    {
+        Assert.Null(IntuneDeviceBindingValidator.ParseEnrolledDateTime(raw));
+    }
+
+    [Fact]
+    public void ParseEnrolledDateTime_NormalizesToUtc()
+    {
+        // A device object created minutes ago is the signature of an enrollment race, so the
+        // instant has to be comparable regardless of the offset Graph reports it in.
+        var parsed = IntuneDeviceBindingValidator.ParseEnrolledDateTime("2026-08-24T17:20:00+02:00");
+
+        Assert.NotNull(parsed);
+        Assert.Equal(TimeSpan.Zero, parsed!.Value.Offset);
+        Assert.Equal(new DateTimeOffset(2026, 8, 24, 15, 20, 0, TimeSpan.Zero), parsed.Value);
+    }
+
+    // -- result semantics --
+
+    [Theory]
+    [InlineData(IntuneDeviceBindingOutcome.NotFound)]
+    [InlineData(IntuneDeviceBindingOutcome.NoDeviceIdInCert)]
+    [InlineData(IntuneDeviceBindingOutcome.PermissionMissing)]
+    public void DefinitiveNegatives_AreNotTransient(IntuneDeviceBindingOutcome outcome)
+    {
+        // Only Transient may be retried / left uncached. A missing grant or a foreign device is a
+        // state, not an outage — retrying cannot change either.
+        var result = new IntuneDeviceBindingResult { Outcome = outcome };
+
+        Assert.False(result.IsTransient);
+        Assert.False(result.IsValid);
+    }
+
+    [Fact]
+    public void AsCacheHit_FlagsTheCopyAndLeavesTheCachedInstanceAlone()
+    {
+        // The cached instance is shared across every request for that device. Marking it in place
+        // would make the first (real) lookup look cached too, and the log gate would then drop the
+        // one line per lookup we actually want to keep.
+        var cached = new IntuneDeviceBindingResult
+        {
+            Outcome = IntuneDeviceBindingOutcome.Match,
+            IntuneDeviceId = DeviceId,
+            DeviceName = "TEST-DEVICE-01",
+            EnrolledDateTime = new DateTimeOffset(2026, 8, 24, 16, 51, 0, TimeSpan.Zero),
+        };
+
+        var hit = cached.AsCacheHit();
+
+        Assert.True(hit.ServedFromCache);
+        Assert.False(cached.ServedFromCache);
+        Assert.NotSame(cached, hit);
+
+        // Everything the log line reports must survive the copy.
+        Assert.Equal(cached.Outcome, hit.Outcome);
+        Assert.Equal(cached.IntuneDeviceId, hit.IntuneDeviceId);
+        Assert.Equal(cached.DeviceName, hit.DeviceName);
+        Assert.Equal(cached.EnrolledDateTime, hit.EnrolledDateTime);
+    }
+
+    [Fact]
+    public void FreshResult_IsNotFlaggedAsCached()
+    {
+        // A freshly parsed Graph response must be loggable — that is the one line per lookup.
+        var body = $$"""
+        { "id": "{{DeviceId}}", "deviceName": "TEST-DEVICE-01" }
+        """;
+
+        var result = IntuneDeviceBindingValidator.ParseManagedDeviceResponse(body, DeviceId);
+
+        Assert.False(result.ServedFromCache);
+    }
+
+    [Fact]
+    public void TransientOutcome_IsNeitherValidNorDefinitive()
+    {
+        var result = new IntuneDeviceBindingResult { Outcome = IntuneDeviceBindingOutcome.Transient };
+
+        Assert.True(result.IsTransient);
+        Assert.False(result.IsValid);
+    }
+}

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

@@ -0,0 +1,166 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using Azure.Data.Tables;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+using Microsoft.Extensions.Logging.Abstractions;
+using Moq;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Pins that the ONE projected window scan the maintenance tick shares across its rolling
+/// sweeps (<c>GetMaintenanceWindowSessionsAsync</c> / <c>MaintenanceSweepSessionProjection</c>)
+/// carries every field the time-attribution, device-journey and verdict-calibration sweeps
+/// read. In Azure Table Storage a <c>$select</c> returns ONLY the listed properties, so the
+/// projected row here is built from the production array — dropping a column there fails the
+/// sweep that consumes it. Mirrors <see cref="UsageMetricsProjectionEquivalenceTests"/>.
+/// </summary>
+public class MaintenanceSweepProjectionEquivalenceTests
+{
+    private const string TenantId = "00000000-0000-0000-0000-000000000abc";
+    private static readonly DateTime Started = new(2026, 8, 10, 9, 0, 0, DateTimeKind.Utc);
+
+    private static readonly TableStorageService Sut =
+        new(new Mock<TableServiceClient>().Object, NullLogger<TableStorageService>.Instance);
+
+    /// <summary>Full wide Sessions row with every column the sweeps read set to a distinct non-default value, plus noise.</summary>
+    private static TableEntity FullRow() => new(TenantId, "11111111-1111-1111-1111-111111111111")
+    {
+        ["Status"] = "Succeeded",
+        ["StartedAt"] = new DateTimeOffset(Started),
+        ["CompletedAt"] = new DateTimeOffset(Started.AddHours(1)),
+        ["LastEventAt"] = new DateTimeOffset(Started.AddHours(2)),
+        ["DurationSeconds"] = 3600,
+        ["EventCount"] = 321,
+        ["DeletionState"] = "None",
+        ["SerialNumber"] = "SN-1234",
+        ["Manufacturer"] = "Contoso",
+        ["Model"] = "Model-X",
+        ["EnrollmentType"] = "v2",
+        ["IsPreProvisioned"] = true,
+        ["IsSelfDeployingProfile"] = true,
+        ["IsUserDriven"] = false,
+        ["ResumedAt"] = new DateTimeOffset(Started.AddMinutes(30)),
+        ["AdminMarkedAction"] = "Succeeded",
+        ["VerdictPath"] = "manual:succeeded",
+        ["PriorStatus"] = "Incomplete",
+        ["PriorVerdictPath"] = "sweep:r6",
+        ["FailureReason"] = "No Device Setup completion or explicit failure signal observed before timeout",
+        ["ReconcileReason"] = "Reconciled at timeout: user completed setup (desktop + Windows Hello)",
+        ["FailureSource"] = "manual",
+        ["EspSoftFailure"] = true,
+        // Noise the sweeps never read — present on the wide row, dropped by the projection.
+        ["DeviceName"] = "PC-FULL",
+        ["OsName"] = "Windows 11",
+        ["OsBuild"] = "26200.1234",
+        ["GeoCountry"] = "DE",
+        ["AgentVersion"] = "2.9.0",
+        ["DiagnosticsBlobName"] = "diag.zip",
+        ["FailureSnapshotJson"] = "{\"big\":\"" + new string('x', 2000) + "\"}",
+    };
+
+    private static TableEntity Project(TableEntity full)
+    {
+        var keep = new HashSet<string>(TableStorageService.MaintenanceSweepSessionProjection, StringComparer.Ordinal);
+        var projected = new TableEntity(full.PartitionKey, full.RowKey);
+        foreach (var kv in full)
+        {
+            if (keep.Contains(kv.Key)) projected[kv.Key] = kv.Value;
+        }
+        return projected;
+    }
+
+    [Fact]
+    public void Projection_drops_the_wide_columns()
+    {
+        var projected = Project(FullRow());
+        Assert.False(projected.ContainsKey("FailureSnapshotJson"));
+        Assert.False(projected.ContainsKey("GeoCountry"));
+        Assert.False(projected.ContainsKey("OsBuild"));
+        Assert.True(projected.Count < FullRow().Count);
+    }
+
+    [Fact]
+    public void Every_field_the_sweeps_read_survives_the_projection()
+    {
+        var full = Sut.MapToSessionSummary(FullRow());
+        var proj = Sut.MapToSessionSummary(Project(FullRow()));
+
+        // Shared / structural
+        Assert.Equal(full.TenantId, proj.TenantId);
+        Assert.Equal(full.SessionId, proj.SessionId);
+        Assert.Equal(full.Status, proj.Status);
+        Assert.Equal(full.StartedAt, proj.StartedAt);
+        Assert.Equal(full.CompletedAt, proj.CompletedAt);
+        Assert.Equal(full.DeletionState, proj.DeletionState);
+
+        // Time attribution: Status/DurationSeconds/DeletionState/EventCount + enrollment class
+        Assert.Equal(full.DurationSeconds, proj.DurationSeconds);
+        Assert.Equal(full.EventCount, proj.EventCount);
+        Assert.Equal(TimeAttributionCalculator.GetEnrollmentClass(full), TimeAttributionCalculator.GetEnrollmentClass(proj));
+
+        // Device journeys: BuildSessionRef + BuildDeviceHistoryRow display fields
+        Assert.Equal(full.SerialNumber, proj.SerialNumber);
+        Assert.Equal(full.Manufacturer, proj.Manufacturer);
+        Assert.Equal(full.Model, proj.Model);
+        Assert.Equal(full.EnrollmentType, proj.EnrollmentType);
+        Assert.Equal(full.IsPreProvisioned, proj.IsPreProvisioned);
+        Assert.Equal(full.AdminMarkedAction, proj.AdminMarkedAction);
+        var refFull = DeviceJourneyCalculator.BuildSessionRef(full)!;
+        var refProj = DeviceJourneyCalculator.BuildSessionRef(proj)!;
+        Assert.Equal(System.Text.Json.JsonSerializer.Serialize(refFull), System.Text.Json.JsonSerializer.Serialize(refProj));
+
+        // Verdict calibration: attribution + legacy derivation inputs + proxy timestamps
+        Assert.Equal(full.LastEventAt, proj.LastEventAt);
+        Assert.Equal(full.VerdictPath, proj.VerdictPath);
+        Assert.Equal(full.PriorStatus, proj.PriorStatus);
+        Assert.Equal(full.PriorVerdictPath, proj.PriorVerdictPath);
+        Assert.Equal(full.FailureReason, proj.FailureReason);
+        Assert.Equal(full.ReconcileReason, proj.ReconcileReason);
+        Assert.Equal(full.FailureSource, proj.FailureSource);
+        Assert.Equal(full.EspSoftFailure, proj.EspSoftFailure);
+        Assert.Equal(VerdictPathDerivation.Derive(full), VerdictPathDerivation.Derive(proj));
+        // Unstamped variant exercises the derivation on the projected inputs.
+        var unstampedFull = FullRow(); unstampedFull.Remove("VerdictPath"); unstampedFull.Remove("AdminMarkedAction"); unstampedFull.Remove("FailureSource");
+        Assert.Equal(VerdictPathDerivation.Derive(Sut.MapToSessionSummary(unstampedFull)),
+                     VerdictPathDerivation.Derive(Sut.MapToSessionSummary(Project(unstampedFull))));
+    }
+
+    [Fact]
+    public void Aggregation_cores_produce_identical_rows_from_the_projected_scan()
+    {
+        var now = new DateTime(2026, 8, 23, 12, 0, 0, DateTimeKind.Utc);
+        var fullList = new List<SessionSummary> { Sut.MapToSessionSummary(FullRow()) };
+        var projList = new List<SessionSummary> { Sut.MapToSessionSummary(Project(FullRow())) };
+        var histories = new Dictionary<string, List<DeviceHistory>>();
+
+        var a = MaintenanceService.BuildVerdictCalibrationAggregates(fullList, histories, now);
+        var b = MaintenanceService.BuildVerdictCalibrationAggregates(projList, histories, now);
+        Assert.Equal(System.Text.Json.JsonSerializer.Serialize(a), System.Text.Json.JsonSerializer.Serialize(b));
+    }
+
+    [Fact]
+    public void SliceSweepWindow_keeps_sessions_from_the_window_start()
+    {
+        var window = new List<SessionSummary>
+        {
+            new() { SessionId = "old", StartedAt = Started.AddDays(-40) },
+            new() { SessionId = "edge", StartedAt = Started },
+            new() { SessionId = "new", StartedAt = Started.AddDays(1) },
+        };
+        var sliced = MaintenanceService.SliceSweepWindow(window, Started);
+        Assert.Equal(new[] { "edge", "new" }, sliced.Select(s => s.SessionId));
+    }
+
+    [Fact]
+    public void Shared_window_covers_every_sweeps_horizon()
+    {
+        // 7d window + 28d baseline anchored on yesterday = 35 days back ≥ the 30-day rolling sweeps.
+        Assert.Equal(35, MaintenanceService.SweepWindowDays);
+        Assert.True(MaintenanceService.SweepWindowDays >= 30);
+    }
+}

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

@@ -112,9 +112,11 @@ public class OffboardingSessionEnumeratorTests
         public Task<List<SessionSummary>> GetSessionsOlderThanAsync(string tenantId, DateTime cutoffDate, int maxResults = int.MaxValue, bool excludeInFlightDeletions = false) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetSessionsByDateRangeAsync(DateTime startDate, DateTime endDate, string? tenantId = null) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetUsageWindowSessionsAsync(DateTime startDate, DateTime endDate, string? tenantId = null) => throw new NotImplementedException();
+        public Task<List<SessionSummary>> GetMaintenanceWindowSessionsAsync(DateTime startDate, DateTime endDate) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetGeoWindowSessionsAsync(DateTime startDate, DateTime endDate, string? tenantId = null) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetStalledSessionsAsync(string tenantId, DateTime cutoffTime) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetLegacyTimeoutFailedSessionsAsync(string tenantId, int maxResults) => throw new NotImplementedException();
+        public Task<List<SessionSummary>> GetSelfDeployingSilentSessionsAsync(string tenantId, int maxResults) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetSessionsLeanAsync(string tenantId) => throw new NotImplementedException();
         public Task<List<SessionSummary>> GetAgentSilentSessionsAsync(string tenantId, DateTime silenceCutoff, DateTime hardCutoff) => throw new NotImplementedException();
         public Task<List<string>> GetAllTenantIdsAsync() => throw new NotImplementedException();

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

@@ -0,0 +1,76 @@
+using System.Collections.Concurrent;
+using AutopilotMonitor.Functions.Services;
+using Microsoft.Extensions.Caching.Memory;
+using Microsoft.Extensions.Logging.Abstractions;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Both sliding-window limiters keep the request history and its lock in one cache entry.
+/// Previously the entry was re-Set without its eviction callback, which (a) leaked per-key lock
+/// objects and (b) let two threads mutate the same List under different locks. These tests hammer
+/// one key from many threads: the count of allowed requests must be exactly the limit and nothing
+/// may throw (a List mutated concurrently throws InvalidOperationException or corrupts Count).
+/// </summary>
+public class RateLimitServiceConcurrencyTests
+{
+    private const int Threads = 16;
+    private const int RequestsPerThread = 500;
+
+    [Fact]
+    public void CheckRateLimit_ManyThreadsSameKey_AllowsExactlyTheLimit()
+    {
+        var sut = new RateLimitService(new MemoryCache(new MemoryCacheOptions()), NullLogger<RateLimitService>.Instance);
+        const int limit = 1000;
+        var allowed = 0;
+        var errors = new ConcurrentBag<Exception>();
+
+        RunInParallel(() =>
+        {
+            try
+            {
+                if (sut.CheckRateLimit("thumb-shared", limit).IsAllowed)
+                    Interlocked.Increment(ref allowed);
+            }
+            catch (Exception ex) { errors.Add(ex); }
+        });
+
+        Assert.Empty(errors);
+        Assert.Equal(limit, allowed);
+        Assert.False(sut.CheckRateLimit("thumb-shared", limit).IsAllowed);
+    }
+
+    [Fact]
+    public void DistressCheck_ManyThreadsSameIp_AllowsExactlyTheIpLimit()
+    {
+        var sut = new DistressRateLimitService(new MemoryCache(new MemoryCacheOptions()), NullLogger<DistressRateLimitService>.Instance);
+        var allowed = 0;
+        var errors = new ConcurrentBag<Exception>();
+
+        RunInParallel(() =>
+        {
+            try
+            {
+                if (sut.Check("10.0.0.9", "a1b2c3d4-e5f6-7890-abcd-ef1234567890").IsAllowed)
+                    Interlocked.Increment(ref allowed);
+            }
+            catch (Exception ex) { errors.Add(ex); }
+        });
+
+        Assert.Empty(errors);
+        Assert.Equal(5, allowed); // MaxPerIp
+    }
+
+    private static void RunInParallel(Action body)
+    {
+        using var gate = new ManualResetEventSlim(false);
+        var tasks = Enumerable.Range(0, Threads).Select(_ => Task.Run(() =>
+        {
+            gate.Wait();
+            for (var i = 0; i < RequestsPerThread; i++) body();
+        })).ToArray();
+        gate.Set();
+        Task.WaitAll(tasks);
+    }
+}

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

@@ -86,6 +86,9 @@ public class SessionIndexFieldManifestTests
         ["ValidatedBy"] = "AutopilotV1",
         ["FailureSnapshotJson"] = "{\"x\":1}",
         ["AdminMarkedAction"] = "failed",
+        ["VerdictPath"] = "sweep:r5_incomplete",
+        ["PriorStatus"] = "Incomplete",
+        ["PriorVerdictPath"] = "maxlife:r6",
         ["DurationSeconds"] = 360,
         ["DiagnosticsBlobName"] = "diag-0001.zip",
         ["DiagnosticsBlobDestination"] = "Hosted",

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

@@ -74,7 +74,7 @@ public class SessionStartedAtBackwardShiftGuardTests
         var harness = new Harness(SessionRow(), earliestStoredEvent: poisoned);
 
         var ok = await harness.Sut.UpdateSessionStatusAsync(
-            TenantId, SessionId, SessionStatus.Pending,
+            TenantId, SessionId, SessionStatus.Pending, VerdictPaths.AgentWhiteGlovePending,
             earliestEventTimestamp: poisoned, latestEventTimestamp: lastEvent);
 
         Assert.True(ok);
@@ -93,7 +93,7 @@ public class SessionStartedAtBackwardShiftGuardTests
         var harness = new Harness(SessionRow(), earliestStoredEvent: earliest);
 
         var ok = await harness.Sut.UpdateSessionStatusAsync(
-            TenantId, SessionId, SessionStatus.Pending,
+            TenantId, SessionId, SessionStatus.Pending, VerdictPaths.AgentWhiteGlovePending,
             latestEventTimestamp: lastEvent);
 
         Assert.True(ok);
@@ -132,7 +132,7 @@ public class SessionStartedAtBackwardShiftGuardTests
         });
 
         var ok = await harness.Sut.UpdateSessionStatusAsync(
-            TenantId, SessionId, SessionStatus.Pending,
+            TenantId, SessionId, SessionStatus.Pending, VerdictPaths.AgentWhiteGlovePending,
             latestEventTimestamp: lastEvent);
 
         Assert.True(ok);

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

@@ -0,0 +1,80 @@
+using System;
+using System.Collections.Generic;
+using System.Threading;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared;
+using Azure;
+using Azure.Data.Tables;
+using Azure.Data.Tables.Models;
+using Microsoft.ApplicationInsights;
+using Microsoft.ApplicationInsights.Channel;
+using Microsoft.ApplicationInsights.DataContracts;
+using Microsoft.ApplicationInsights.Extensibility;
+using Microsoft.Extensions.Hosting;
+using Microsoft.Extensions.Logging.Abstractions;
+using Moq;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests
+{
+    public class StartupTelemetryServiceTests
+    {
+        private sealed class CapturingChannel : ITelemetryChannel
+        {
+            public List<ITelemetry> Items { get; } = new();
+            public bool? DeveloperMode { get; set; }
+            public string EndpointAddress { get; set; } = "";
+            public void Send(ITelemetry item) => Items.Add(item);
+            public void Flush() { }
+            public void Dispose() { }
+        }
+
+        private sealed class Lifetime : IHostApplicationLifetime
+        {
+            private readonly CancellationTokenSource _started = new();
+            public CancellationToken ApplicationStarted => _started.Token;
+            public CancellationToken ApplicationStopping => CancellationToken.None;
+            public CancellationToken ApplicationStopped => CancellationToken.None;
+            public void StopApplication() { }
+            public void FireStarted() => _started.Cancel();
+        }
+
+        [Fact]
+        public async Task ApplicationStarted_EmitsStartupAndTableInitMetrics()
+        {
+            var channel = new CapturingChannel();
+            var telemetry = new TelemetryClient(new TelemetryConfiguration { TelemetryChannel = channel, ConnectionString = "InstrumentationKey=00000000-0000-0000-0000-000000000000" });
+
+            // Table storage: unreadable sentinel → full pass → FullPassRan = true.
+            var service = new Mock<TableServiceClient>();
+            var admin = new Mock<TableClient>();
+            service.Setup(s => s.GetTableClient(It.IsAny<string>())).Returns(admin.Object);
+            admin.Setup(c => c.GetEntityIfExistsAsync<TableEntity>(It.IsAny<string>(), It.IsAny<string>(), It.IsAny<IEnumerable<string>>(), It.IsAny<CancellationToken>()))
+                .ThrowsAsync(new RequestFailedException(404, "TableNotFound"));
+            admin.Setup(c => c.UpsertEntityAsync(It.IsAny<TableEntity>(), It.IsAny<TableUpdateMode>(), It.IsAny<CancellationToken>()))
+                .ReturnsAsync(Mock.Of<Response>());
+            service.Setup(s => s.CreateTableIfNotExistsAsync(It.IsAny<string>(), It.IsAny<CancellationToken>()))
+                .ReturnsAsync(Response.FromValue(new TableItem("x"), Mock.Of<Response>()));
+            var storage = new TableStorageService(service.Object, NullLogger<TableStorageService>.Instance);
+            await storage.InitializeTablesAsync();
+
+            var lifetime = new Lifetime();
+            var sut = new StartupTelemetryService(lifetime, telemetry, storage, new BackendBuildInfo(), NullLogger<StartupTelemetryService>.Instance);
+            await sut.StartAsync(CancellationToken.None);
+            Assert.Empty(channel.Items);
+
+            lifetime.FireStarted();
+
+            var metrics = channel.Items.ConvertAll(i => (MetricTelemetry)i);
+            Assert.Equal(2, metrics.Count);
+            var startup = metrics.Find(m => m.Name == StartupTelemetryService.StartupMetricName)!;
+            var tableInit = metrics.Find(m => m.Name == StartupTelemetryService.TableInitMetricName)!;
+            Assert.True(startup.Sum > 0);
+            Assert.True(tableInit.Sum >= 0);
+            Assert.Equal("True", startup.Properties["tableInitFullPass"]);
+            Assert.False(string.IsNullOrEmpty(startup.Properties["version"]));
+            Assert.True(startup.Sum >= tableInit.Sum);
+        }
+    }
+}

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

@@ -48,11 +48,74 @@ public class StorageDependencyFilterProcessorTests
         Assert.Empty(next.Received);
     }
 
-    [Fact]
-    public void FailedStorageDependency_IsKept()
+    [Theory]
+    [InlineData("429")]
+    [InlineData("500")]
+    [InlineData("503")]
+    [InlineData("403")]
+    [InlineData("")] // unknown outcome (timeout/network) — keep
+    public void FailedStorageDependency_IsKept(string resultCode)
     {
         var (processor, next) = Build();
         var dep = Dep("myacct.queue.core.windows.net", success: false);
+        dep.ResultCode = resultCode;
+        processor.Process(dep);
+        Assert.Same(dep, Assert.Single(next.Received));
+    }
+
+    // Expected outcomes (point-read miss, ETag precondition, insert conflict) are normal control
+    // flow — live-verified as 100% of the remaining "failed" storage rows. Dropped.
+    [Theory]
+    [InlineData("404")]
+    [InlineData("412")]
+    [InlineData("409")]
+    public void ExpectedStorageOutcome_HttpShape_IsDropped(string resultCode)
+    {
+        var (processor, next) = Build();
+        var dep = Dep("myacct.table.core.windows.net", success: false);
+        dep.ResultCode = resultCode;
+        processor.Process(dep);
+        Assert.Empty(next.Received);
+    }
+
+    [Theory]
+    [InlineData("Azure.RequestFailedException: The specified resource does not exist.\nRequestId:x\nStatus: 404 (Not Found)\nErrorCode: ResourceNotFound")]
+    [InlineData("Azure.RequestFailedException: The update condition specified in the request was not satisfied.\nStatus: 412 (Precondition Failed)\nErrorCode: UpdateConditionNotSatisfied")]
+    [InlineData("Azure.RequestFailedException: The specified entity already exists.\nStatus: 409 (Conflict)\nErrorCode: EntityAlreadyExists")]
+    public void ExpectedStorageOutcome_InProcShape_IsDropped(string error)
+    {
+        // InProc spans have no ResultCode — only the exception text in Properties["Error"].
+        var (processor, next) = Build();
+        var dep = Dep(target: "TableClient.GetEntity", success: false, type: "InProc | Microsoft.Tables");
+        dep.Properties["Error"] = error;
+        processor.Process(dep);
+        Assert.Empty(next.Received);
+    }
+
+    [Fact]
+    public void ThrottledInProcStorageDependency_IsKept()
+    {
+        var (processor, next) = Build();
+        var dep = Dep(target: "TableClient.SubmitTransaction", success: false, type: "InProc | Microsoft.Tables");
+        dep.Properties["Error"] = "Azure.RequestFailedException: Too many requests.\nStatus: 429 (Too Many Requests)\nErrorCode: ServerBusy";
+        processor.Process(dep);
+        Assert.Same(dep, Assert.Single(next.Received));
+    }
+
+    [Fact]
+    public void SuccessfulSignalRRestCall_IsDropped()
+    {
+        var (processor, next) = Build();
+        processor.Process(Dep("autopilotmonitor-eu.service.signalr.net", success: true, data: "POST /api/hubs/autopilotmonitor/groups/t-x/connections/y", type: "HTTP"));
+        Assert.Empty(next.Received);
+    }
+
+    [Fact]
+    public void FailedSignalRRestCall_IsKept()
+    {
+        var (processor, next) = Build();
+        var dep = Dep("autopilotmonitor-eu.service.signalr.net", success: false, type: "HTTP");
+        dep.ResultCode = "500";
         processor.Process(dep);
         Assert.Same(dep, Assert.Single(next.Received));
     }

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

@@ -0,0 +1,180 @@
+using System;
+using System.Linq;
+using System.Threading;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared;
+using Azure;
+using Azure.Data.Tables;
+using Azure.Data.Tables.Models;
+using Microsoft.Extensions.Logging.Abstractions;
+using Moq;
+using Xunit;
+
+namespace AutopilotMonitor.Functions.Tests
+{
+    /// <summary>
+    /// Startup table initialization is gated by a schema sentinel (hash over TableNames.All)
+    /// stored in AdminConfiguration: a matching sentinel means zero CreateTableIfNotExists
+    /// calls on a cold start; anything else runs the full idempotent pass and rewrites it.
+    /// </summary>
+    public class TableSchemaSentinelTests
+    {
+        private sealed class Harness
+        {
+            public Mock<TableServiceClient> Service { get; } = new();
+            public Mock<TableClient> AdminConfig { get; } = new();
+            public TableEntity? UpsertedSentinel { get; private set; }
+            public int CreateCalls;
+
+            public Harness()
+            {
+                Service.Setup(s => s.GetTableClient(Constants.TableNames.AdminConfiguration)).Returns(AdminConfig.Object);
+                Service.Setup(s => s.CreateTableIfNotExistsAsync(It.IsAny<string>(), It.IsAny<CancellationToken>()))
+                    .Returns<string, CancellationToken>((_, _) =>
+                    {
+                        Interlocked.Increment(ref CreateCalls);
+                        return Task.FromResult(Response.FromValue(new TableItem("x"), Mock.Of<Response>()));
+                    });
+                AdminConfig.Setup(c => c.UpsertEntityAsync(It.IsAny<TableEntity>(), It.IsAny<TableUpdateMode>(), It.IsAny<CancellationToken>()))
+                    .Callback<ITableEntity, TableUpdateMode, CancellationToken>((e, _, _) => UpsertedSentinel = (TableEntity)e)
+                    .ReturnsAsync(Mock.Of<Response>());
+            }
+
+            public void SentinelReturns(string? hash)
+            {
+                var response = new Mock<NullableResponse<TableEntity>>();
+                response.SetupGet(r => r.HasValue).Returns(hash != null);
+                if (hash != null)
+                {
+                    response.SetupGet(r => r.Value).Returns(new TableEntity(
+                        TableStorageService.SchemaSentinelPartitionKey, TableStorageService.SchemaSentinelRowKey)
+                    {
+                        [TableStorageService.SchemaSentinelHashProperty] = hash
+                    });
+                }
+                AdminConfig.Setup(c => c.GetEntityIfExistsAsync<TableEntity>(
+                        TableStorageService.SchemaSentinelPartitionKey, TableStorageService.SchemaSentinelRowKey,
+                        It.IsAny<System.Collections.Generic.IEnumerable<string>>(), It.IsAny<CancellationToken>()))
+                    .ReturnsAsync(response.Object);
+            }
+
+            public void SentinelThrows()
+            {
+                AdminConfig.Setup(c => c.GetEntityIfExistsAsync<TableEntity>(
+                        It.IsAny<string>(), It.IsAny<string>(),
+                        It.IsAny<System.Collections.Generic.IEnumerable<string>>(), It.IsAny<CancellationToken>()))
+                    .ThrowsAsync(new RequestFailedException(404, "TableNotFound"));
+            }
+
+            public TableStorageService Build() => new(Service.Object, NullLogger<TableStorageService>.Instance);
+        }
+
+        private static string CurrentHash => TableStorageService.ComputeTableSchemaHash(Constants.TableNames.All);
+
+        [Fact]
+        public void Hash_IsOrderIndependent_AndChangesWhenTableAdded()
+        {
+            var a = TableStorageService.ComputeTableSchemaHash(new[] { "B", "A" });
+            var b = TableStorageService.ComputeTableSchemaHash(new[] { "A", "B" });
+            var c = TableStorageService.ComputeTableSchemaHash(new[] { "A", "B", "C" });
+
+            Assert.Equal(a, b);
+            Assert.NotEqual(a, c);
+            Assert.Equal(64, a.Length);
+        }
+
+        [Fact]
+        public async Task MatchingSentinel_SkipsTableCreation()
+        {
+            var h = new Harness();
+            h.SentinelReturns(CurrentHash);
+
+            var fullPass = await h.Build().InitializeTablesAsync();
+
+            Assert.False(fullPass);
+            Assert.Equal(0, h.CreateCalls);
+            Assert.Null(h.UpsertedSentinel);
+        }
+
+        [Fact]
+        public async Task StaleSentinel_RunsFullPass_AndRewritesSentinel()
+        {
+            var h = new Harness();
+            h.SentinelReturns("STALE");
+
+            var fullPass = await h.Build().InitializeTablesAsync();
+
+            Assert.True(fullPass);
+            Assert.Equal(Constants.TableNames.All.Length, h.CreateCalls);
+            Assert.NotNull(h.UpsertedSentinel);
+            Assert.Equal(CurrentHash, h.UpsertedSentinel!.GetString(TableStorageService.SchemaSentinelHashProperty));
+        }
+
+        [Fact]
+        public async Task MissingSentinel_RunsFullPass()
+        {
+            var h = new Harness();
+            h.SentinelReturns(null);
+
+            Assert.True(await h.Build().InitializeTablesAsync());
+            Assert.Equal(Constants.TableNames.All.Length, h.CreateCalls);
+        }
+
+        [Fact]
+        public async Task UnreadableSentinel_FreshStorage_RunsFullPass()
+        {
+            var h = new Harness();
+            h.SentinelThrows();
+
+            Assert.True(await h.Build().InitializeTablesAsync());
+            Assert.Equal(Constants.TableNames.All.Length, h.CreateCalls);
+            Assert.NotNull(h.UpsertedSentinel);
+        }
+
+        [Fact]
+        public async Task FailedTableCreate_DoesNotWriteSentinel_AndRetriesNextCall()
+        {
+            var h = new Harness();
+            h.SentinelReturns(null);
+            var failing = Constants.TableNames.All.First();
+            h.Service.Setup(s => s.CreateTableIfNotExistsAsync(failing, It.IsAny<CancellationToken>()))
+                .ThrowsAsync(new RequestFailedException(500, "boom"));
+            var svc = h.Build();
+
+            Assert.True(await svc.InitializeTablesAsync());
+            Assert.Null(h.UpsertedSentinel);
+
+            // Not marked initialized → a second call runs the pass again instead of short-circuiting.
+            var before = h.CreateCalls;
+            Assert.True(await svc.InitializeTablesAsync());
+            Assert.True(h.CreateCalls > before);
+        }
+
+        [Fact]
+        public async Task SecondCallAfterSuccess_IsNoOp()
+        {
+            var h = new Harness();
+            h.SentinelReturns(null);
+            var svc = h.Build();
+
+            await svc.InitializeTablesAsync();
+            var calls = h.CreateCalls;
+            Assert.False(await svc.InitializeTablesAsync());
+            Assert.Equal(calls, h.CreateCalls);
+        }
+
+        [Fact]
+        public async Task BackfillClaim_WinnerTrue_LoserFalse()
+        {
+            var h = new Harness();
+            h.AdminConfig.SetupSequence(c => c.AddEntityAsync(It.IsAny<TableEntity>(), It.IsAny<CancellationToken>()))
+                .ReturnsAsync(Mock.Of<Response>())
+                .ThrowsAsync(new RequestFailedException(409, "EntityAlreadyExists"));
+            var svc = h.Build();
+
+            Assert.True(await svc.TryClaimSessionIndexBackfillAsync());
+            Assert.False(await svc.TryClaimSessionIndexBackfillAsync());
+        }
+    }
+}

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

@@ -0,0 +1,206 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text.Json;
+using AutopilotMonitor.Functions.DataAccess.TableStorage;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Verdict-calibration drift radar (docs/backend/verdict-calibration.md): the three finding
+/// kinds, the small-n / lift / Wilson gates, re-arm, and the tracker round-trip.
+/// </summary>
+public class VerdictCalibrationRadarTests
+{
+    private static readonly DateTime Target = new(2026, 8, 22, 0, 0, 0, DateTimeKind.Utc);
+
+    /// <summary>35 daily rows ending on Target; per-day buckets chosen by window/baseline membership.</summary>
+    private static List<VerdictCalibrationDailyAggregate> Horizon(
+        Func<bool, (string Path, string Status, int Count)[]> bucketsFor, int sessionsPerDay = 20)
+    {
+        var rows = new List<VerdictCalibrationDailyAggregate>();
+        for (var i = 34; i >= 0; i--)
+        {
+            var inWindow = i < 7;
+            var date = Target.AddDays(-i);
+            rows.Add(new VerdictCalibrationDailyAggregate
+            {
+                TenantId = "t", Date = date.ToString("yyyy-MM-dd"), SessionCount = sessionsPerDay, TerminalSessionCount = sessionsPerDay,
+                Buckets = bucketsFor(inWindow).Select(b => new VerdictCalibrationBucket { VerdictPath = b.Path, Status = b.Status, Count = b.Count }).ToList(),
+            });
+        }
+        return rows;
+    }
+
+    [Fact]
+    public void Share_regression_fires_on_a_doubled_separated_share()
+    {
+        // baseline: 1/20 per day (5%), window: 4/20 per day (20%) → lift 4, clearly separated
+        var rows = Horizon(w => new[] { ("sweep:r5_incomplete", "Incomplete", w ? 4 : 1), ("agent:complete", "Succeeded", w ? 16 : 19) });
+        var findings = VerdictCalibrationRadar.Evaluate(rows, Target);
+
+        var f = Assert.Single(findings, x => x.Kind == VerdictCalibrationAlertKinds.ShareRegression);
+        Assert.Equal("sweep:r5_incomplete", f.VerdictPath);
+        Assert.Equal("Incomplete", f.Status);
+        Assert.Equal(28, f.WindowHitCount);
+        Assert.Equal(140, f.WindowSessionCount);
+        Assert.Equal(28, f.BaselineHitCount);
+        Assert.Equal(560, f.BaselineSessionCount);
+        Assert.Equal(20.0, f.WindowRatePct);
+        Assert.Equal(5.0, f.BaselineRatePct);
+        Assert.Equal(4.0, f.Lift);
+        Assert.Equal("2026-08-16", f.WindowStartDate);
+        Assert.Equal("2026-08-22", f.WindowEndDate);
+        // sweep:* is also the silence group → silence share doubled as well
+        Assert.Contains(findings, x => x.Kind == VerdictCalibrationAlertKinds.SilenceShareRegression && x.VerdictPath == VerdictCalibrationRadar.SilenceGroupPath);
+    }
+
+    [Fact]
+    public void Share_regression_is_suppressed_below_lift_two_or_without_separation()
+    {
+        // 5% → 8%: lift 1.6 — under the gate
+        var mild = Horizon(w => new[] { ("agent:failed", "Failed", w ? 2 : 1), ("agent:complete", "Succeeded", w ? 18 : 19) }, sessionsPerDay: 20);
+        Assert.DoesNotContain(VerdictCalibrationRadar.Evaluate(mild, Target), f => f.Kind == VerdictCalibrationAlertKinds.ShareRegression);
+
+        // tiny n: 1/2 per day in window vs 0/2 baseline — lift infinite but Wilson intervals overlap / hits < 5
+        var tiny = Horizon(w => new[] { ("agent:failed", "Failed", w ? 1 : 0), ("agent:complete", "Succeeded", w ? 1 : 2) }, sessionsPerDay: 2);
+        Assert.Empty(VerdictCalibrationRadar.Evaluate(tiny, Target));
+    }
+
+    [Fact]
+    public void Agent_and_registration_paths_never_share_alert_but_still_feed_the_silence_group()
+    {
+        // Workflow mix, not classifier drift: a WhiteGlove rollout week doubles the pending share.
+        var wg = Horizon(w => new[] { ("agent:whiteglove_pending", "Pending", w ? 6 : 1), ("agent:complete", "Succeeded", w ? 14 : 19) });
+        Assert.DoesNotContain(VerdictCalibrationRadar.Evaluate(wg, Target), f => f.Kind == VerdictCalibrationAlertKinds.ShareRegression);
+        var reg = Horizon(w => new[] { ("register:superseded", "Incomplete", w ? 6 : 1), ("agent:complete", "Succeeded", w ? 14 : 19) });
+        Assert.DoesNotContain(VerdictCalibrationRadar.Evaluate(reg, Target), f => f.Kind == VerdictCalibrationAlertKinds.ShareRegression);
+        // Backend-decided paths stay eligible.
+        Assert.True(VerdictCalibrationRadar.IsShareRegressionEligible("sweep:r6"));
+        Assert.True(VerdictCalibrationRadar.IsShareRegressionEligible("rule:ANALYZE-ESP-001"));
+        Assert.True(VerdictCalibrationRadar.IsShareRegressionEligible("manual:failed"));
+        Assert.False(VerdictCalibrationRadar.IsShareRegressionEligible("agent:complete_soft"));
+
+        // An episode opened by the earlier, broader gate re-arms on the next pass regardless of numbers.
+        var stale = new VerdictCalibrationAlert { Kind = VerdictCalibrationAlertKinds.ShareRegression, VerdictPath = "agent:whiteglove_pending", Status = "Pending" };
+        Assert.True(VerdictCalibrationRadar.ShouldReArm(stale, wg, Target));
+    }
+
+    [Fact]
+    public void Derived_legacy_paths_never_alert()
+    {
+        var rows = Horizon(w => new[] { ("legacy:r6", "Incomplete", w ? 10 : 1), ("agent:complete", "Succeeded", w ? 10 : 19) });
+        Assert.DoesNotContain(VerdictCalibrationRadar.Evaluate(rows, Target), f => f.Kind == VerdictCalibrationAlertKinds.ShareRegression);
+    }
+
+    [Fact]
+    public void Evidence_gap_fires_on_r6_share_and_ignores_non_rule_sweep_paths()
+    {
+        // classifier verdicts in window: r6 5/day + r5 15/day = 140; r6 share 25% → fires. sweep:stalled is not a rule verdict.
+        var rows = Horizon(w => new[]
+        {
+            ("sweep:r6", "Incomplete", 5), ("sweep:r5_incomplete", "Incomplete", 15), ("sweep:stalled", "Stalled", 30),
+        }, sessionsPerDay: 50);
+        var f = Assert.Single(VerdictCalibrationRadar.Evaluate(rows, Target), x => x.Kind == VerdictCalibrationAlertKinds.EvidenceGap);
+        Assert.Equal(VerdictCalibrationRadar.EvidenceGapPath, f.VerdictPath);
+        Assert.Equal(35, f.WindowHitCount);
+        Assert.Equal(140, f.WindowSessionCount);
+        Assert.Equal(25.0, f.WindowRatePct);
+        Assert.Null(f.Lift);
+    }
+
+    [Fact]
+    public void Evidence_gap_needs_twenty_classifier_verdicts()
+    {
+        var rows = Horizon(w => new[] { ("sweep:r6", "Incomplete", w ? 2 : 0), ("agent:complete", "Succeeded", 18) });
+        Assert.DoesNotContain(VerdictCalibrationRadar.Evaluate(rows, Target), f => f.Kind == VerdictCalibrationAlertKinds.EvidenceGap);
+    }
+
+    [Fact]
+    public void ReArm_when_share_falls_under_one_point_five_or_path_stops()
+    {
+        var alert = new VerdictCalibrationAlert { Kind = VerdictCalibrationAlertKinds.ShareRegression, VerdictPath = "sweep:r6", Status = "Incomplete" };
+
+        var stillHigh = Horizon(w => new[] { ("sweep:r6", "Incomplete", w ? 4 : 1), ("agent:complete", "Succeeded", w ? 16 : 19) });
+        Assert.False(VerdictCalibrationRadar.ShouldReArm(alert, stillHigh, Target));
+
+        var elevatedButUnderGate = Horizon(w => new[] { ("sweep:r6", "Incomplete", w ? 2 : 1), ("agent:complete", "Succeeded", w ? 18 : 19) }, sessionsPerDay: 20);
+        // 10% vs 5% = 2.0× — still at/above 1.5× → keep
+        Assert.False(VerdictCalibrationRadar.ShouldReArm(alert, elevatedButUnderGate, Target));
+
+        var back = Horizon(w => new[] { ("sweep:r6", "Incomplete", 1), ("agent:complete", "Succeeded", 19) });
+        Assert.True(VerdictCalibrationRadar.ShouldReArm(alert, back, Target));
+
+        var gone = Horizon(w => new[] { ("sweep:r6", "Incomplete", w ? 0 : 1), ("agent:complete", "Succeeded", w ? 20 : 19) });
+        Assert.True(VerdictCalibrationRadar.ShouldReArm(alert, gone, Target));
+    }
+
+    [Fact]
+    public void Evidence_gap_rearms_under_fifteen_percent()
+    {
+        var alert = new VerdictCalibrationAlert { Kind = VerdictCalibrationAlertKinds.EvidenceGap, VerdictPath = "r6", Status = "*" };
+        var still = Horizon(w => new[] { ("sweep:r6", "Incomplete", 4), ("sweep:r5_incomplete", "Incomplete", 16) });
+        Assert.False(VerdictCalibrationRadar.ShouldReArm(alert, still, Target)); // 20%
+        var between = Horizon(w => new[] { ("sweep:r6", "Incomplete", 17), ("sweep:r5_incomplete", "Incomplete", 83) }, sessionsPerDay: 100);
+        Assert.False(VerdictCalibrationRadar.ShouldReArm(alert, between, Target)); // 17%
+        var low = Horizon(w => new[] { ("sweep:r6", "Incomplete", 1), ("sweep:r5_incomplete", "Incomplete", 19) });
+        Assert.True(VerdictCalibrationRadar.ShouldReArm(alert, low, Target)); // 5%
+    }
+
+    [Fact]
+    public void CurrentSums_follow_the_alert_kind()
+    {
+        var rows = Horizon(w => new[] { ("sweep:r6", "Incomplete", 2), ("maxlife:r3", "Incomplete", 1), ("agent:complete", "Succeeded", 17) });
+        var share = VerdictCalibrationRadar.CurrentSums(new VerdictCalibrationAlert { Kind = VerdictCalibrationAlertKinds.ShareRegression, VerdictPath = "sweep:r6", Status = "Incomplete" }, rows, Target);
+        Assert.Equal((14, 140, 56, 560), share);
+        var silence = VerdictCalibrationRadar.CurrentSums(new VerdictCalibrationAlert { Kind = VerdictCalibrationAlertKinds.SilenceShareRegression }, rows, Target);
+        Assert.Equal((21, 140, 84, 560), silence);
+        var gap = VerdictCalibrationRadar.CurrentSums(new VerdictCalibrationAlert { Kind = VerdictCalibrationAlertKinds.EvidenceGap }, rows, Target);
+        Assert.Equal((14, 21, 56, 84), gap);
+    }
+
+    [Fact]
+    public void Rows_outside_the_horizon_are_ignored()
+    {
+        var rows = Horizon(w => new[] { ("agent:complete", "Succeeded", 20) });
+        rows.Add(new VerdictCalibrationDailyAggregate { TenantId = "t", Date = "2026-08-23", SessionCount = 1000, Buckets = { new VerdictCalibrationBucket { VerdictPath = "sweep:r6", Status = "Incomplete", Count = 1000 } } });
+        rows.Add(new VerdictCalibrationDailyAggregate { TenantId = "t", Date = "2026-07-01", SessionCount = 1000, Buckets = { new VerdictCalibrationBucket { VerdictPath = "sweep:r6", Status = "Incomplete", Count = 1000 } } });
+        Assert.Empty(VerdictCalibrationRadar.Evaluate(rows, Target));
+    }
+
+    // ---- tracker round-trip ----
+
+    [Fact]
+    public void Alert_round_trips_through_the_tracker_entity()
+    {
+        var alert = new VerdictCalibrationAlert
+        {
+            TenantId = "11111111-1111-1111-1111-111111111111", Kind = VerdictCalibrationAlertKinds.ShareRegression,
+            VerdictPath = "rule:ANALYZE-ESP-001", Status = "Failed",
+            WindowHitCount = 7, WindowSessionCount = 40, BaselineHitCount = 3, BaselineSessionCount = 160,
+            WindowRatePct = 17.5, BaselineRatePct = 1.9, Lift = 9.3, WindowStartDate = "2026-08-16", WindowEndDate = "2026-08-22",
+            Dimension = new RuleRegressionDimension { Dimension = "osBuild", Value = "26200.1234", HitCount = 6, HitSharePct = 85.7, AllSharePct = 20.0, Lift = 4.3 },
+            FirstNotifiedAt = new DateTime(2026, 8, 20, 6, 0, 0, DateTimeKind.Utc), LastEvaluatedAt = new DateTime(2026, 8, 23, 6, 0, 0, DateTimeKind.Utc),
+        };
+        var entity = TableHardwareRejectionNotificationTracker.BuildVerdictCalibrationEntity(alert.TenantId, alert);
+        Assert.Equal("verdictcalibration|share_regression|rule:analyze-esp-001|failed", entity.RowKey);
+        Assert.Equal(alert.TenantId, entity.PartitionKey);
+
+        var back = TableHardwareRejectionNotificationTracker.MapToVerdictCalibrationAlert(entity);
+        Assert.Equal(JsonSerializer.Serialize(alert), JsonSerializer.Serialize(back));
+
+        // Tri-states: no lift / no dimension → absent columns → null on read, never 0 / a guessed dimension.
+        var bare = new VerdictCalibrationAlert
+        {
+            TenantId = alert.TenantId, Kind = VerdictCalibrationAlertKinds.EvidenceGap, VerdictPath = "r6", Status = "*",
+            FirstNotifiedAt = alert.FirstNotifiedAt, LastEvaluatedAt = alert.LastEvaluatedAt,
+        };
+        var bareBack = TableHardwareRejectionNotificationTracker.MapToVerdictCalibrationAlert(
+            TableHardwareRejectionNotificationTracker.BuildVerdictCalibrationEntity(bare.TenantId, bare));
+        Assert.Null(bareBack.Lift);
+        Assert.Null(bareBack.Dimension);
+        Assert.Equal("verdictcalibration|evidence_gap|r6|*", TableHardwareRejectionNotificationTracker.BuildVerdictCalibrationRowKey(bare.Kind, bare.VerdictPath, bare.Status));
+    }
+}

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

@@ -0,0 +1,301 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using System.Text.Json;
+using AutopilotMonitor.Functions.Functions.Metrics;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Verdict calibration aggregate (docs/backend/verdict-calibration.md): the pure bucketing core,
+/// the persistence round-trip and the matrix arithmetic of the read endpoint.
+/// </summary>
+public class VerdictCalibrationTests
+{
+    private const string TenantA = "11111111-1111-1111-1111-111111111111";
+    private const string TenantB = "22222222-2222-2222-2222-222222222222";
+    private static readonly DateTime Now = new(2026, 8, 23, 12, 0, 0, DateTimeKind.Utc);
+
+    private static SessionSummary Session(string tenant, string id, SessionStatus status, DateTime started,
+        string? path = null, string? priorPath = null, string? priorStatus = null, string? adminMarked = null,
+        DateTime? completed = null, string serial = "SN-1", string? failureReason = null, string? reconcileReason = null) => new()
+    {
+        TenantId = tenant, SessionId = id, SerialNumber = serial, Status = status, StartedAt = started,
+        CompletedAt = completed, VerdictPath = path, PriorVerdictPath = priorPath, PriorStatus = priorStatus,
+        AdminMarkedAction = adminMarked, FailureReason = failureReason ?? string.Empty,
+        ReconcileReason = reconcileReason ?? string.Empty, FailureSource = string.Empty,
+    };
+
+    private static DeviceHistory History(string tenant, string serial, params (string Id, DateTime Started, DateTime? Completed, string Status)[] refs) => new()
+    {
+        TenantId = tenant, SerialKey = serial.ToLowerInvariant(), SerialNumber = serial,
+        Chain = refs.Select(r => new DeviceSessionRef { SessionId = r.Id, StartedAt = r.Started, CompletedAt = r.Completed, Status = r.Status }).ToList(),
+    };
+
+    private static VerdictCalibrationBucket Bucket(VerdictCalibrationDailyAggregate row, string path, string status)
+        => Assert.Single(row.Buckets, b => b.VerdictPath == path && b.Status == status);
+
+    // ---- BuildVerdictCalibrationAggregates ----
+
+    [Fact]
+    public void Buckets_by_tenant_date_path_and_status_with_global_mirror()
+    {
+        var d1 = new DateTime(2026, 8, 10, 9, 0, 0, DateTimeKind.Utc);
+        var sessions = new List<SessionSummary>
+        {
+            Session(TenantA, "a1", SessionStatus.Succeeded, d1, path: "agent:complete", completed: d1.AddHours(1)),
+            Session(TenantA, "a2", SessionStatus.Succeeded, d1.AddHours(2), path: "agent:complete", completed: d1.AddHours(3)),
+            Session(TenantA, "a3", SessionStatus.Incomplete, d1, path: "sweep:r6", completed: d1.AddHours(6)),
+            Session(TenantA, "a4", SessionStatus.Stalled, d1.AddDays(1), path: "sweep:stalled"),
+            Session(TenantB, "b1", SessionStatus.Failed, d1, path: "agent:failed", completed: d1.AddHours(1)),
+        };
+        var rows = MaintenanceService.BuildVerdictCalibrationAggregates(sessions, new Dictionary<string, List<DeviceHistory>>(), Now);
+
+        var a10 = Assert.Single(rows, r => r.TenantId == TenantA && r.Date == "2026-08-10");
+        Assert.Equal(3, a10.SessionCount);
+        Assert.Equal(3, a10.TerminalSessionCount);
+        Assert.Equal(2, Bucket(a10, "agent:complete", "Succeeded").Count);
+        Assert.Equal(1, Bucket(a10, "sweep:r6", "Incomplete").Count);
+
+        var a11 = Assert.Single(rows, r => r.TenantId == TenantA && r.Date == "2026-08-11");
+        Assert.Equal(1, a11.SessionCount);
+        Assert.Equal(0, a11.TerminalSessionCount); // Stalled is not terminal
+        Assert.Equal(1, Bucket(a11, "sweep:stalled", "Stalled").Count);
+
+        var g10 = Assert.Single(rows, r => r.TenantId == "global" && r.Date == "2026-08-10");
+        Assert.Equal(4, g10.SessionCount);
+        Assert.Equal(1, Bucket(g10, "agent:failed", "Failed").Count);
+        Assert.Equal(VerdictCalibrationDailyAggregate.CurrentVersion, g10.Version);
+        Assert.Equal(Now, g10.ComputedAt);
+
+        // Deterministic ordering: rows by tenant then date, buckets by path then status.
+        Assert.Equal(rows.Select(r => (r.TenantId, r.Date)).OrderBy(k => k.TenantId, StringComparer.Ordinal).ThenBy(k => k.Date, StringComparer.Ordinal), rows.Select(r => (r.TenantId, r.Date)));
+        Assert.Equal(a10.Buckets.Select(b => b.VerdictPath).OrderBy(p => p, StringComparer.Ordinal), a10.Buckets.Select(b => b.VerdictPath));
+    }
+
+    [Fact]
+    public void Unstamped_rows_are_derived_and_counted_as_derived()
+    {
+        var d = new DateTime(2026, 8, 10, 9, 0, 0, DateTimeKind.Utc);
+        var sessions = new List<SessionSummary>
+        {
+            Session(TenantA, "a1", SessionStatus.Succeeded, d, path: null, completed: d.AddHours(1)), // legacy agent completion
+            Session(TenantA, "a2", SessionStatus.Incomplete, d, path: null, completed: d.AddHours(5),
+                failureReason: "No Device Setup completion or explicit failure signal observed before timeout"),
+            Session(TenantA, "a3", SessionStatus.Succeeded, d, path: "agent:complete", completed: d.AddHours(1)),
+        };
+        var row = MaintenanceService.BuildVerdictCalibrationAggregates(sessions, new Dictionary<string, List<DeviceHistory>>(), Now)
+            .Single(r => r.TenantId == TenantA);
+
+        var complete = Bucket(row, "agent:complete", "Succeeded");
+        Assert.Equal(2, complete.Count);
+        Assert.Equal(1, complete.DerivedCount);
+        var r6 = Bucket(row, "legacy:r6", "Incomplete");
+        Assert.Equal(1, r6.Count);
+        Assert.Equal(1, r6.DerivedCount);
+    }
+
+    [Fact]
+    public void Overrides_are_attributed_to_the_prior_path()
+    {
+        var d = new DateTime(2026, 8, 10, 9, 0, 0, DateTimeKind.Utc);
+        var sessions = new List<SessionSummary>
+        {
+            // sweep said Incomplete, agent later completed → late completion against sweep:r5_incomplete
+            Session(TenantA, "a1", SessionStatus.Succeeded, d, path: "agent:complete", priorPath: "sweep:r5_incomplete", priorStatus: "Incomplete", completed: d.AddHours(1)),
+            // admin flipped an agent success to Failed
+            Session(TenantA, "a2", SessionStatus.Failed, d, path: "manual:failed", priorPath: "agent:complete", priorStatus: "Succeeded", adminMarked: "Failed", completed: d.AddHours(1)),
+            // retro reclassification Failed → Incomplete
+            Session(TenantA, "a3", SessionStatus.Incomplete, d, path: "retro:r6", priorPath: "agent:failed", priorStatus: "Failed", completed: d.AddHours(1)),
+            // admin mark to Succeeded counts as admin, not late completion
+            Session(TenantA, "a4", SessionStatus.Succeeded, d, path: "manual:succeeded", priorPath: "sweep:r6", priorStatus: "Incomplete", adminMarked: "Succeeded", completed: d.AddHours(1)),
+        };
+        var row = MaintenanceService.BuildVerdictCalibrationAggregates(sessions, new Dictionary<string, List<DeviceHistory>>(), Now)
+            .Single(r => r.TenantId == TenantA);
+
+        var priorSweep = Bucket(row, "sweep:r5_incomplete", "Incomplete");
+        Assert.Equal(0, priorSweep.Count); // nobody is on it any more
+        Assert.Equal(1, priorSweep.OverriddenByLateCompletion);
+
+        var priorAgent = Bucket(row, "agent:complete", "Succeeded");
+        Assert.Equal(1, priorAgent.Count);  // a1 is on it now
+        Assert.Equal(1, priorAgent.OverriddenByAdmin); // a2 used to be
+
+        Assert.Equal(1, Bucket(row, "agent:failed", "Failed").OverriddenOther);
+        Assert.Equal(1, Bucket(row, "sweep:r6", "Incomplete").OverriddenByAdmin);
+        Assert.Equal(0, Bucket(row, "manual:succeeded", "Succeeded").OverriddenByAdmin);
+    }
+
+    [Fact]
+    public void ReEnrollment_proxy_counts_only_eligible_terminal_sessions()
+    {
+        var old = Now.AddDays(-20);
+        var young = Now.AddDays(-2);
+        var sessions = new List<SessionSummary>
+        {
+            // old, re-enrolled 3 days later → eligible + re-enrolled
+            Session(TenantA, "a1", SessionStatus.Incomplete, old, path: "sweep:r6", completed: old.AddHours(5), serial: "SN-A"),
+            // old, next attempt 10 days later → eligible, not re-enrolled
+            Session(TenantA, "a2", SessionStatus.Failed, old, path: "agent:failed", completed: old.AddHours(1), serial: "SN-B"),
+            // old, no next attempt → eligible, not re-enrolled
+            Session(TenantA, "a3", SessionStatus.Succeeded, old, path: "agent:complete", completed: old.AddHours(1), serial: "SN-C"),
+            // too young to judge → not eligible even though the chain shows a follow-up
+            Session(TenantA, "a4", SessionStatus.Failed, young, path: "agent:failed", completed: young.AddHours(1), serial: "SN-D"),
+            // non-terminal → never eligible
+            Session(TenantA, "a5", SessionStatus.AwaitingUser, old, path: "sweep:r5_awaiting", serial: "SN-E"),
+        };
+        var histories = new Dictionary<string, List<DeviceHistory>>
+        {
+            [TenantA] = new()
+            {
+                History(TenantA, "SN-A", ("a1", old, old.AddHours(5), "Incomplete"), ("x1", old.AddDays(3), old.AddDays(3).AddHours(1), "Succeeded")),
+                History(TenantA, "SN-B", ("a2", old, old.AddHours(1), "Failed"), ("x2", old.AddDays(10), null, "Succeeded")),
+                History(TenantA, "SN-C", ("a3", old, old.AddHours(1), "Succeeded")),
+                History(TenantA, "SN-D", ("a4", young, young.AddHours(1), "Failed"), ("x4", young.AddHours(2), null, "Succeeded")),
+            },
+        };
+        var rows = MaintenanceService.BuildVerdictCalibrationAggregates(sessions, histories, Now);
+
+        var oldRow = rows.Single(r => r.TenantId == TenantA && r.Date == old.ToString("yyyy-MM-dd"));
+        var r6 = Bucket(oldRow, "sweep:r6", "Incomplete");
+        Assert.Equal((1, 1), (r6.Eligible7d, r6.ReEnrolled7d));
+        var failed = Bucket(oldRow, "agent:failed", "Failed");
+        Assert.Equal((1, 0), (failed.Eligible7d, failed.ReEnrolled7d));
+        var complete = Bucket(oldRow, "agent:complete", "Succeeded");
+        Assert.Equal((1, 0), (complete.Eligible7d, complete.ReEnrolled7d));
+        var awaiting = Bucket(oldRow, "sweep:r5_awaiting", "AwaitingUser");
+        Assert.Equal((0, 0), (awaiting.Eligible7d, awaiting.ReEnrolled7d));
+
+        var youngRow = rows.Single(r => r.TenantId == TenantA && r.Date == young.ToString("yyyy-MM-dd"));
+        var youngFailed = Bucket(youngRow, "agent:failed", "Failed");
+        Assert.Equal((0, 0), (youngFailed.Eligible7d, youngFailed.ReEnrolled7d));
+    }
+
+    [Fact]
+    public void Sessions_inside_a_deletion_cascade_are_skipped()
+    {
+        var d = new DateTime(2026, 8, 10, 9, 0, 0, DateTimeKind.Utc);
+        var deleting = Session(TenantA, "a1", SessionStatus.Succeeded, d, path: "agent:complete");
+        deleting.DeletionState = "Pending";
+        var rows = MaintenanceService.BuildVerdictCalibrationAggregates(new[] { deleting }, new Dictionary<string, List<DeviceHistory>>(), Now);
+        Assert.Empty(rows);
+    }
+
+    // ---- Persistence round-trip ----
+
+    [Fact]
+    public void Aggregate_round_trips_through_the_table_entity()
+    {
+        var aggregate = new VerdictCalibrationDailyAggregate
+        {
+            TenantId = TenantA, Date = "2026-08-10", Version = 1, SessionCount = 5, TerminalSessionCount = 4, ComputedAt = Now,
+            Buckets =
+            {
+                new VerdictCalibrationBucket { VerdictPath = "agent:complete", Status = "Succeeded", Count = 3, DerivedCount = 1, Eligible7d = 3, ReEnrolled7d = 1, OverriddenByAdmin = 1, OverriddenByLateCompletion = 0, OverriddenOther = 2 },
+                new VerdictCalibrationBucket { VerdictPath = "sweep:r6", Status = "Incomplete", Count = 1 },
+            },
+        };
+        var entity = TableStorageService.BuildVerdictCalibrationAggregateEntity(aggregate);
+        Assert.Equal(TenantA, entity.PartitionKey);
+        Assert.Equal("2026-08-10", entity.RowKey);
+
+        var back = TableStorageService.MapToVerdictCalibrationAggregate(entity);
+        Assert.Equal(JsonSerializer.Serialize(aggregate), JsonSerializer.Serialize(back));
+    }
+
+    // ---- Response builder ----
+
+    private static VerdictCalibrationDailyAggregate Day(string tenant, DateTime date, int sessions, params (string Path, string Status, int Count)[] buckets) => new()
+    {
+        TenantId = tenant, Date = date.ToString("yyyy-MM-dd"), SessionCount = sessions, TerminalSessionCount = sessions, Version = 1, ComputedAt = Now,
+        Buckets = buckets.Select(b => new VerdictCalibrationBucket { VerdictPath = b.Path, Status = b.Status, Count = b.Count, Eligible7d = b.Count, ReEnrolled7d = b.Count / 5 }).ToList(),
+    };
+
+    // Same naming policy as the worker host (Program.cs: camelCase) so the assertions pin the wire shape.
+    private static readonly JsonSerializerOptions Wire = new() { PropertyNamingPolicy = JsonNamingPolicy.CamelCase };
+
+    private static JsonElement BuildJson(IReadOnlyList<VerdictCalibrationDailyAggregate> daily, DateTime today, int days)
+        => JsonSerializer.SerializeToElement(VerdictCalibrationResponse.Build(daily, "global", today, days), Wire);
+
+    [Fact]
+    public void Response_sums_the_window_and_computes_share_rate_and_trend()
+    {
+        var today = new DateTime(2026, 8, 23, 0, 0, 0, DateTimeKind.Utc);
+        var daily = new List<VerdictCalibrationDailyAggregate>();
+        // 35 days: baseline days carry 10 sessions (1 sweep:r6), the last 7 days carry 10 sessions (3 sweep:r6) → share 10% → 30%, lift 3.0
+        for (var i = 34; i >= 0; i--)
+        {
+            var date = today.AddDays(-i);
+            var r6 = i < 7 ? 3 : 1;
+            daily.Add(Day("global", date, 10, ("agent:complete", "Succeeded", 10 - r6), ("sweep:r6", "Incomplete", r6)));
+        }
+        var json = BuildJson(daily, today, days: 30);
+
+        Assert.True(json.GetProperty("success").GetBoolean());
+        Assert.Equal(30, json.GetProperty("windowDays").GetInt32());
+        Assert.Equal(300, json.GetProperty("totals").GetProperty("sessions").GetInt32());
+        Assert.Equal(30, json.GetProperty("totals").GetProperty("days").GetInt32());
+
+        var paths = json.GetProperty("paths").EnumerateArray().ToList();
+        var r6Row = paths.Single(p => p.GetProperty("verdictPath").GetString() == "sweep:r6");
+        Assert.Equal(7 * 3 + 23 * 1, r6Row.GetProperty("count").GetInt32());
+        Assert.Equal(Math.Round(100.0 * 44 / 300, 1), r6Row.GetProperty("sharePct").GetDouble());
+        Assert.Equal(30.0, r6Row.GetProperty("window7").GetProperty("sharePct").GetDouble());
+        Assert.Equal(10.0, r6Row.GetProperty("baseline28").GetProperty("sharePct").GetDouble());
+        Assert.Equal(3.0, r6Row.GetProperty("lift").GetDouble());
+        // Eligible 44 ≥ 20 → rate stated; ReEnrolled7d = count/5 per day → 7*0 + 23*0 = 0 → 0%
+        Assert.Equal(0.0, r6Row.GetProperty("reEnrollRatePct").GetDouble());
+
+        // Sorted by count desc → agent:complete first.
+        Assert.Equal("agent:complete", paths[0].GetProperty("verdictPath").GetString());
+    }
+
+    [Fact]
+    public void Response_withholds_rate_below_twenty_eligible_and_lift_without_baseline()
+    {
+        var today = new DateTime(2026, 8, 23, 0, 0, 0, DateTimeKind.Utc);
+        var daily = new List<VerdictCalibrationDailyAggregate>
+        {
+            Day("global", today, 5, ("maxlife:r3", "Incomplete", 5)),
+        };
+        var json = BuildJson(daily, today, days: 7);
+        var row = json.GetProperty("paths").EnumerateArray().Single();
+        Assert.Equal(JsonValueKind.Null, row.GetProperty("reEnrollRatePct").ValueKind);
+        Assert.Equal(JsonValueKind.Null, row.GetProperty("lift").ValueKind); // no baseline rows → never invented
+        Assert.Equal(100.0, row.GetProperty("sharePct").GetDouble());
+    }
+
+    [Fact]
+    public void Response_keeps_a_zero_count_bucket_that_carries_overrides()
+    {
+        var today = new DateTime(2026, 8, 23, 0, 0, 0, DateTimeKind.Utc);
+        var day = Day("global", today, 1, ("agent:complete", "Succeeded", 1));
+        day.Buckets.Add(new VerdictCalibrationBucket { VerdictPath = "sweep:r5_incomplete", Status = "Incomplete", Count = 0, OverriddenByLateCompletion = 1 });
+        var json = BuildJson(new[] { day }, today, days: 7);
+        var rows = json.GetProperty("paths").EnumerateArray().ToList();
+        var prior = rows.Single(p => p.GetProperty("verdictPath").GetString() == "sweep:r5_incomplete");
+        Assert.Equal(0, prior.GetProperty("count").GetInt32());
+        Assert.Equal(1, prior.GetProperty("overriddenByLateCompletion").GetInt32());
+    }
+
+    [Fact]
+    public void Response_is_empty_but_well_formed_before_the_first_sweep()
+    {
+        var json = BuildJson(Array.Empty<VerdictCalibrationDailyAggregate>(), new DateTime(2026, 8, 23), 30);
+        Assert.True(json.GetProperty("success").GetBoolean());
+        Assert.Empty(json.GetProperty("paths").EnumerateArray());
+        Assert.Equal(0, json.GetProperty("totals").GetProperty("sessions").GetInt32());
+        Assert.Equal(JsonValueKind.Null, json.GetProperty("computedAt").ValueKind);
+    }
+
+    [Theory]
+    [InlineData(null, 30)]
+    [InlineData("abc", 30)]
+    [InlineData("0", 1)]
+    [InlineData("500", 180)]
+    [InlineData("14", 14)]
+    public void ClampDays_bounds_the_window(string? raw, int expected)
+        => Assert.Equal(expected, VerdictCalibrationResponse.ClampDays(raw));
+}

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

@@ -0,0 +1,201 @@
+using System;
+using System.Collections.Generic;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Tests;
+
+/// <summary>
+/// Verdict calibration instrumentation (docs/backend/verdict-calibration.md): every status write
+/// declares its origin, overriding a prior verdict preserves it, the classifier names the rule
+/// behind each verdict, and pre-instrumentation rows derive an honest path from their literals.
+/// </summary>
+public class VerdictPathTests
+{
+    private static readonly DateTime Start = new(2026, 8, 20, 8, 0, 0, DateTimeKind.Utc);
+
+    // ---- ComputePriorVerdict: the correction stream ----
+
+    [Theory]
+    [InlineData("Succeeded", "agent:complete", "Failed", true)]        // admin mark over an agent success
+    [InlineData("Incomplete", "sweep:r5_incomplete", "Succeeded", true)] // late agent completion upgrades a sweep verdict
+    [InlineData("AwaitingUser", "sweep:r5_awaiting", "Incomplete", true)] // grace expired
+    [InlineData("Failed", "agent:failed", "Incomplete", true)]        // retro reclassification
+    [InlineData("Stalled", "sweep:stalled", "InProgress", false)]      // stall heal is not a correction
+    [InlineData("Stalled", "sweep:stalled", "Succeeded", false)]
+    [InlineData("InProgress", "register:new", "Succeeded", false)]
+    [InlineData("Pending", "agent:whiteglove_pending", "InProgress", false)]
+    [InlineData("AwaitingUser", "sweep:wg_awaiting", "AwaitingUser", false)] // same-status refresh keeps the original prior
+    public void Prior_is_preserved_only_when_a_verdict_is_overridden(string existing, string existingPath, string incoming, bool expectPrior)
+    {
+        var prior = TableStorageService.ComputePriorVerdict(existing, existingPath, Enum.Parse<SessionStatus>(incoming));
+        if (expectPrior)
+        {
+            Assert.NotNull(prior);
+            Assert.Equal(existing, prior!.Value.PriorStatus);
+            Assert.Equal(existingPath, prior.Value.PriorVerdictPath);
+        }
+        else
+        {
+            Assert.Null(prior);
+        }
+    }
+
+    [Fact]
+    public void Prior_needs_a_stamped_path_on_the_existing_row()
+    {
+        // Pre-instrumentation rows carry no path — there is nothing trustworthy to preserve.
+        Assert.Null(TableStorageService.ComputePriorVerdict("Incomplete", null, SessionStatus.Succeeded));
+        Assert.Null(TableStorageService.ComputePriorVerdict("Incomplete", "", SessionStatus.Succeeded));
+        Assert.Null(TableStorageService.ComputePriorVerdict(null, "sweep:r6", SessionStatus.Succeeded));
+    }
+
+    // ---- VerdictPaths vocabulary ----
+
+    [Fact]
+    public void Compose_and_Origin_round_trip()
+    {
+        var path = VerdictPaths.Compose(VerdictPaths.OriginSweep, ClassifierRules.R5DeviceSetupIncomplete);
+        Assert.Equal("sweep:r5_incomplete", path);
+        Assert.Equal("sweep", VerdictPaths.Origin(path));
+        Assert.True(VerdictPaths.IsClassifierPath(path));
+        Assert.True(VerdictPaths.IsClassifierPath("maxlife:r1"));
+        Assert.True(VerdictPaths.IsClassifierPath("late:r4"));
+        Assert.True(VerdictPaths.IsClassifierPath("retro:r6"));
+        Assert.False(VerdictPaths.IsClassifierPath(VerdictPaths.AgentComplete));
+        Assert.False(VerdictPaths.IsClassifierPath(VerdictPaths.SweepStalled));
+        Assert.False(VerdictPaths.IsClassifierPath(VerdictPaths.RetroSuperseded));
+        Assert.False(VerdictPaths.IsClassifierPath("legacy:r6"));
+        Assert.False(VerdictPaths.IsClassifierPath(VerdictPaths.ManualFailed));
+        Assert.Throws<ArgumentException>(() => VerdictPaths.Compose("", "r1"));
+        Assert.Throws<ArgumentException>(() => VerdictPaths.Compose("sweep", " "));
+    }
+
+    // ---- Classifier: every return names its rule ----
+
+    private static EnrollmentEvent Evt(string type, string? message = null) => new()
+    {
+        EventType = type, Timestamp = Start, Source = "test", Message = message ?? type,
+    };
+
+    private static EnrollmentEvent Esp(string message) => Evt("esp_provisioning_status", message);
+    private const string DeviceSetup44 = "ESP provisioning status: DeviceSetup — 4 of 4 subcategories completed";
+    private const string AccountSetup55 = "ESP provisioning status: AccountSetup — 5 of 5 subcategories completed";
+    private const string AccountSetup15 = "ESP provisioning status: AccountSetup — 1 of 5 subcategories completed";
+
+    private static string RuleOf(IReadOnlyList<EnrollmentEvent> events, double hours = 6, int grace = 72,
+        bool preProvisioned = false, DateTime? resumedAt = null, bool selfDeploying = false)
+    {
+        var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
+        var (_, _, rule) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+            rollup, Start, Start.AddHours(hours), grace, isPreProvisioned: preProvisioned,
+            resumedAt: resumedAt, isSelfDeployingProfile: selfDeploying);
+        return rule;
+    }
+
+    [Fact]
+    public void Classifier_returns_the_rule_id_behind_each_verdict()
+    {
+        Assert.Equal(ClassifierRules.R1ExplicitFailure, RuleOf(new[] { Evt("enrollment_failed") }));
+        Assert.Equal(ClassifierRules.R1bWhiteGloveAwaiting, RuleOf(
+            new[] { Esp(DeviceSetup44), Evt("whiteglove_complete") }, hours: 6, grace: 51, preProvisioned: true, resumedAt: Start.AddHours(1)));
+        Assert.Equal(ClassifierRules.R1bWhiteGloveSucceeded, RuleOf(
+            new[] { Esp(DeviceSetup44), Evt("whiteglove_complete") }, hours: 60, grace: 51, preProvisioned: true, resumedAt: Start.AddHours(1)));
+        Assert.Equal(ClassifierRules.R1cSelfDeploying, RuleOf(new[] { Esp(DeviceSetup44) }, selfDeploying: true));
+        Assert.Equal(ClassifierRules.R2AccountSetupComplete, RuleOf(new[] { Esp(DeviceSetup44), Esp(AccountSetup55) }));
+        Assert.Equal(ClassifierRules.R3EmergencyBreak, RuleOf(new[] { Esp(DeviceSetup44), Evt("agent_emergency_break") }));
+        Assert.Equal(ClassifierRules.R4DesktopHello, RuleOf(new[] { Esp(DeviceSetup44), Evt("desktop_arrived"), Evt("hello_provisioning_completed") }));
+        Assert.Equal(ClassifierRules.R5DeviceSetupAwaiting, RuleOf(new[] { Esp(DeviceSetup44), Esp(AccountSetup15) }, hours: 6, grace: 72));
+        Assert.Equal(ClassifierRules.R5DeviceSetupIncomplete, RuleOf(new[] { Esp(DeviceSetup44), Esp(AccountSetup15) }, hours: 80, grace: 72));
+        Assert.Equal(ClassifierRules.R6Fallthrough, RuleOf(Array.Empty<EnrollmentEvent>()));
+    }
+
+    // ---- Derivation for pre-instrumentation rows ----
+
+    private static SessionSummary Row(SessionStatus status, string? failureReason = null, string? reconcileReason = null,
+        string? failureSource = null, string? adminMarked = null, bool soft = false, string? verdictPath = null) => new()
+    {
+        SessionId = "s", TenantId = "t", Status = status,
+        FailureReason = failureReason ?? string.Empty, ReconcileReason = reconcileReason ?? string.Empty,
+        FailureSource = failureSource ?? string.Empty, AdminMarkedAction = adminMarked, EspSoftFailure = soft,
+        VerdictPath = verdictPath,
+    };
+
+    private static string Classified(IReadOnlyList<EnrollmentEvent> events, double hours = 80, int grace = 72, bool selfDeploying = false)
+    {
+        var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
+        var (_, reason, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+            rollup, Start, Start.AddHours(hours), grace, isSelfDeployingProfile: selfDeploying);
+        return reason;
+    }
+
+    [Fact]
+    public void Stamped_rows_are_returned_verbatim()
+    {
+        var (path, derived) = VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete, failureReason: "anything", verdictPath: "sweep:r6"));
+        Assert.Equal("sweep:r6", path);
+        Assert.False(derived);
+    }
+
+    [Fact]
+    public void Derivation_uses_the_real_classifier_literals()
+    {
+        // Reasons come straight from the classifier so a wording change there surfaces here.
+        Assert.Equal(("legacy:r6", true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete,
+            failureReason: Classified(Array.Empty<EnrollmentEvent>()))));
+        Assert.Equal(("legacy:r5_incomplete", true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete,
+            failureReason: Classified(new[] { Esp(DeviceSetup44), Esp(AccountSetup15) }))));
+        Assert.Equal(("legacy:r5_awaiting", true), VerdictPathDerivation.Derive(Row(SessionStatus.AwaitingUser,
+            failureReason: Classified(new[] { Esp(DeviceSetup44), Esp(AccountSetup15) }, hours: 6))));
+        Assert.Equal(("legacy:r3", true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete,
+            failureReason: Classified(new[] { Esp(DeviceSetup44), Evt("agent_emergency_break") }))));
+        Assert.Equal(("legacy:r2", true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded,
+            reconcileReason: Classified(new[] { Esp(DeviceSetup44), Esp(AccountSetup55) }))));
+        Assert.Equal(("legacy:r4", true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded,
+            reconcileReason: Classified(new[] { Esp(DeviceSetup44), Evt("desktop_arrived"), Evt("hello_provisioning_completed") }))));
+        Assert.Equal(("legacy:r1c", true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded,
+            reconcileReason: Classified(new[] { Esp(DeviceSetup44) }, selfDeploying: true))));
+        Assert.Equal(("legacy:r1", true), VerdictPathDerivation.Derive(Row(SessionStatus.Failed,
+            failureReason: Classified(new[] { Evt("enrollment_failed") }))));
+    }
+
+    [Fact]
+    public void Derivation_recovers_the_origin_from_writer_suffixes()
+    {
+        var r6 = Classified(Array.Empty<EnrollmentEvent>());
+        Assert.Equal(("maxlife:r6", true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete,
+            failureReason: r6 + " Verdict triggered by the agent's max-lifetime watchdog shutdown.")));
+        Assert.Equal(("retro:r6", true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete,
+            failureReason: r6 + " Retro-reclassified from the legacy blanket timeout verdict.")));
+    }
+
+    [Fact]
+    public void Derivation_attributes_unambiguous_non_classifier_paths()
+    {
+        Assert.Equal((VerdictPaths.ManualFailed, true), VerdictPathDerivation.Derive(Row(SessionStatus.Failed,
+            failureReason: "Manually marked as failed by administrator", failureSource: "manual", adminMarked: "Failed")));
+        Assert.Equal((VerdictPaths.ManualSucceeded, true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded, adminMarked: "Succeeded")));
+        Assert.Equal(("rule:ANALYZE-ESP-001", true), VerdictPathDerivation.Derive(Row(SessionStatus.Failed,
+            failureReason: "Rule: ESP timeout", failureSource: "rule:ANALYZE-ESP-001")));
+        Assert.Equal((VerdictPaths.AgentComplete, true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded)));
+        Assert.Equal((VerdictPaths.AgentComplete, true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded,
+            reconcileReason: "Late completion report received — upgraded prior 'Incomplete' verdict")));
+        Assert.Equal((VerdictPaths.AgentCompleteSoft, true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded, soft: true)));
+        Assert.Equal((VerdictPaths.AgentFailed, true), VerdictPathDerivation.Derive(Row(SessionStatus.Failed, failureReason: "ESP timeout (0x800705B4)")));
+        Assert.Equal((VerdictPaths.AgentEspFailureFallback, true), VerdictPathDerivation.Derive(Row(SessionStatus.Failed, failureReason: "ESP failure (backend fallback)")));
+        Assert.Equal((VerdictPaths.SweepStalled, true), VerdictPathDerivation.Derive(Row(SessionStatus.Stalled, failureReason: "Agent silent for 130min (detected by maintenance sweep)")));
+        Assert.Equal((VerdictPaths.AgentStallProbe, true), VerdictPathDerivation.Derive(Row(SessionStatus.Stalled, failureReason: "Agent reported stall after 60min without progress (stall_probe)")));
+        Assert.Equal(("legacy:superseded", true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete, failureReason: "Superseded by session abc: re-registered")));
+        Assert.Equal((VerdictPaths.AgentWhiteGlovePending, true), VerdictPathDerivation.Derive(Row(SessionStatus.Pending)));
+    }
+
+    [Fact]
+    public void Derivation_never_invents_a_path()
+    {
+        Assert.Equal((VerdictPaths.LegacyUnknown, true), VerdictPathDerivation.Derive(Row(SessionStatus.Failed, failureReason: "Session timed out after 5 hours (started at 2026-01-01 UTC)")));
+        Assert.Equal((VerdictPaths.LegacyUnknown, true), VerdictPathDerivation.Derive(Row(SessionStatus.Incomplete)));
+        Assert.Equal((VerdictPaths.LegacyUnknown, true), VerdictPathDerivation.Derive(Row(SessionStatus.InProgress)));
+        Assert.Equal((VerdictPaths.LegacyUnknown, true), VerdictPathDerivation.Derive(Row(SessionStatus.Succeeded, reconcileReason: "Some future wording")));
+    }
+}

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

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Collections.Generic;
 using System.Linq;
 using System.Text.Json;
@@ -652,6 +652,7 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
                 { "ValidateCorporateIdentifier", config.ValidateCorporateIdentifier },
                 { "ValidateDeviceAssociation", config.ValidateDeviceAssociation },
                 { "ValidateCloudPcDevice", config.ValidateCloudPcDevice },
+                { "ValidateIntuneDeviceBinding", config.ValidateIntuneDeviceBinding },
                 { "AllowInsecureAgentRequests", config.AllowInsecureAgentRequests },
                 { "DataRetentionDays", config.DataRetentionDays },
                 { "SessionTimeoutHours", config.SessionTimeoutHours },
@@ -757,6 +758,7 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
                 ValidateCorporateIdentifier = entity.GetBoolean("ValidateCorporateIdentifier") ?? false,
                 ValidateDeviceAssociation = entity.GetBoolean("ValidateDeviceAssociation") ?? false,
                 ValidateCloudPcDevice = entity.GetBoolean("ValidateCloudPcDevice") ?? false,
+                ValidateIntuneDeviceBinding = entity.GetBoolean("ValidateIntuneDeviceBinding") ?? false,
                 AllowInsecureAgentRequests = entity.GetBoolean("AllowInsecureAgentRequests") ?? false,
                 DataRetentionDays = entity.GetInt32("DataRetentionDays") ?? 90,
                 SessionTimeoutHours = entity.GetInt32("SessionTimeoutHours") ?? 5,

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

@@ -422,6 +422,152 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             };
         }
 
+        // ===== VERDICT CALIBRATION DRIFT (docs/backend/verdict-calibration.md) =====
+
+        internal const string VerdictCalibrationRowKeyPrefix = "verdictcalibration|";
+
+        internal static string BuildVerdictCalibrationRowKey(string kind, string verdictPath, string status)
+        {
+            // Paths are a closed vocabulary plus rule ids (ANALYZE-...), statuses are enum names and
+            // the group marker "*" - none contain table-key-hostile characters, but sanitize anyway.
+            var k = SanitizeTableKey((kind ?? string.Empty).Trim().ToLowerInvariant());
+            var p = SanitizeTableKey((verdictPath ?? string.Empty).Trim().ToLowerInvariant());
+            var s = SanitizeTableKey((status ?? string.Empty).Trim().ToLowerInvariant());
+            return $"{VerdictCalibrationRowKeyPrefix}{k}|{p}|{s}";
+        }
+
+        public async Task<bool> TryRegisterVerdictCalibrationAlertAsync(string tenantId, VerdictCalibrationAlert alert)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(alert?.Kind) || string.IsNullOrWhiteSpace(alert?.VerdictPath))
+                return false;
+            try
+            {
+                await _table.AddEntityAsync(BuildVerdictCalibrationEntity(tenantId, alert!));
+                _logger.LogInformation(
+                    "VerdictCalibration tracker registered: tenant={TenantId} kind={Kind} path={Path}", tenantId, alert!.Kind, alert.VerdictPath);
+                return true;
+            }
+            catch (RequestFailedException ex) when (ex.Status == 409)
+            {
+                return false; // episode already active - one ops event per episode
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "VerdictCalibration tracker failed: tenant={TenantId} kind={Kind} path={Path}", tenantId, alert!.Kind, alert.VerdictPath);
+                return false; // fail closed - never double-fire
+            }
+        }
+
+        public async Task RefreshVerdictCalibrationAlertAsync(string tenantId, VerdictCalibrationAlert alert)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(alert?.Kind) || string.IsNullOrWhiteSpace(alert?.VerdictPath))
+                return;
+            try
+            {
+                await _table.UpsertEntityAsync(BuildVerdictCalibrationEntity(tenantId, alert!), TableUpdateMode.Replace);
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "VerdictCalibration tracker refresh failed: tenant={TenantId} kind={Kind} path={Path}", tenantId, alert!.Kind, alert.VerdictPath);
+            }
+        }
+
+        public async Task DeleteVerdictCalibrationAlertAsync(string tenantId, string kind, string verdictPath, string status)
+        {
+            if (string.IsNullOrWhiteSpace(tenantId) || string.IsNullOrWhiteSpace(kind) || string.IsNullOrWhiteSpace(verdictPath))
+                return;
+            try
+            {
+                await _table.DeleteEntityAsync(tenantId.ToLowerInvariant(), BuildVerdictCalibrationRowKey(kind, verdictPath, status));
+            }
+            catch (RequestFailedException ex) when (ex.Status == 404)
+            {
+                // already gone - idempotent
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "VerdictCalibration tracker delete failed: tenant={TenantId} kind={Kind} path={Path}", tenantId, kind, verdictPath);
+            }
+        }
+
+        public async Task<List<VerdictCalibrationAlert>> GetVerdictCalibrationAlertsAsync(string tenantId)
+        {
+            var results = new List<VerdictCalibrationAlert>();
+            if (string.IsNullOrWhiteSpace(tenantId)) return results;
+            try
+            {
+                var partitionKey = tenantId.ToLowerInvariant();
+                // '}' sorts directly after '|' - prefix range over "verdictcalibration|...".
+                var filter = $"PartitionKey eq '{partitionKey}' and RowKey ge '{VerdictCalibrationRowKeyPrefix}' and RowKey lt 'verdictcalibration}}'";
+                await foreach (var entity in _table.QueryAsync<TableEntity>(filter: filter))
+                {
+                    results.Add(MapToVerdictCalibrationAlert(entity));
+                }
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "VerdictCalibration tracker list failed for tenant {TenantId}", tenantId);
+            }
+            return results;
+        }
+
+        // internal static: pinned by round-trip unit tests (table-storage-serialization rule).
+        internal static TableEntity BuildVerdictCalibrationEntity(string tenantId, VerdictCalibrationAlert alert)
+        {
+            var entity = new TableEntity(tenantId.ToLowerInvariant(), BuildVerdictCalibrationRowKey(alert.Kind, alert.VerdictPath, alert.Status))
+            {
+                ["TenantId"] = tenantId,
+                ["Kind"] = alert.Kind,
+                ["VerdictPath"] = alert.VerdictPath,
+                ["Status"] = alert.Status ?? string.Empty,
+                ["WindowHitCount"] = alert.WindowHitCount,
+                ["WindowSessionCount"] = alert.WindowSessionCount,
+                ["BaselineHitCount"] = alert.BaselineHitCount,
+                ["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,
+            };
+            if (alert.Lift.HasValue)
+                entity["Lift"] = alert.Lift.Value;
+            if (alert.Dimension != null)
+                entity["DimensionJson"] = JsonSerializer.Serialize(alert.Dimension);
+            return entity;
+        }
+
+        internal static VerdictCalibrationAlert MapToVerdictCalibrationAlert(TableEntity entity)
+        {
+            RuleRegressionDimension? dimension = null;
+            var dimensionJson = entity.GetString("DimensionJson");
+            if (!string.IsNullOrEmpty(dimensionJson))
+            {
+                try { dimension = JsonSerializer.Deserialize<RuleRegressionDimension>(dimensionJson!); }
+                catch (JsonException) { dimension = null; }
+            }
+            return new VerdictCalibrationAlert
+            {
+                TenantId = entity.GetString("TenantId") ?? entity.PartitionKey,
+                Kind = entity.GetString("Kind") ?? string.Empty,
+                VerdictPath = entity.GetString("VerdictPath") ?? string.Empty,
+                Status = entity.GetString("Status") ?? string.Empty,
+                WindowHitCount = entity.GetInt32("WindowHitCount") ?? 0,
+                WindowSessionCount = entity.GetInt32("WindowSessionCount") ?? 0,
+                BaselineHitCount = entity.GetInt32("BaselineHitCount") ?? 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,
+            };
+        }
+
         internal const string AppVersionRegressionRowKeyPrefix = "appversionregression|";
 
         /// <summary>

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

@@ -60,6 +60,9 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
         public Task<List<SessionSummary>> GetUsageWindowSessionsAsync(DateTime startDate, DateTime endDate, string? tenantId = null)
             => _storage.GetUsageWindowSessionsAsync(startDate, endDate, tenantId);
 
+        public Task<List<SessionSummary>> GetMaintenanceWindowSessionsAsync(DateTime startDate, DateTime endDate)
+            => _storage.GetMaintenanceWindowSessionsAsync(startDate, endDate);
+
         public Task<List<SessionSummary>> GetGeoWindowSessionsAsync(DateTime startDate, DateTime endDate, string? tenantId = null)
             => _storage.GetGeoWindowSessionsAsync(startDate, endDate, tenantId);
 
@@ -72,6 +75,9 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
         public Task<List<SessionSummary>> GetLegacyTimeoutFailedSessionsAsync(string tenantId, int maxResults)
             => _storage.GetLegacyTimeoutFailedSessionsAsync(tenantId, maxResults);
 
+        public Task<List<SessionSummary>> GetSelfDeployingSilentSessionsAsync(string tenantId, int maxResults)
+            => _storage.GetSelfDeployingSilentSessionsAsync(tenantId, maxResults);
+
         public Task<List<SessionSummary>> GetSessionsLeanAsync(string tenantId)
             => _storage.GetSessionsLeanAsync(tenantId);
 

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

@@ -159,5 +159,17 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
 
         public Task<int> DeleteDeviceJourneyAggregatesOlderThanAsync(DateTime cutoffDate)
             => _storage.DeleteDeviceJourneyAggregatesOlderThanAsync(cutoffDate);
+
+        public Task<bool> SaveVerdictCalibrationAggregateAsync(VerdictCalibrationDailyAggregate aggregate)
+            => _storage.SaveVerdictCalibrationAggregateAsync(aggregate);
+
+        public Task<List<VerdictCalibrationDailyAggregate>> GetVerdictCalibrationAggregatesAsync(string tenantId, DateTime startDate, DateTime endDate)
+            => _storage.GetVerdictCalibrationAggregatesAsync(tenantId, startDate, endDate);
+
+        public Task DeleteVerdictCalibrationAggregateAsync(string tenantId, string dateKey)
+            => _storage.DeleteVerdictCalibrationAggregateAsync(tenantId, dateKey);
+
+        public Task<int> DeleteVerdictCalibrationAggregatesOlderThanAsync(DateTime cutoffDate)
+            => _storage.DeleteVerdictCalibrationAggregatesOlderThanAsync(cutoffDate);
     }
 }

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

@@ -58,7 +58,7 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             => _storage.GetAllSessionStatsAsync(tenantIdFilter, days, allowedTenantIds);
 
         public async Task<bool> UpdateSessionStatusAsync(
-            string tenantId, string sessionId, SessionStatus status,
+            string tenantId, string sessionId, SessionStatus status, string verdictPath,
             EnrollmentPhase? currentPhase = null, string? failureReason = null,
             DateTime? completedAt = null, DateTime? earliestEventTimestamp = null,
             DateTime? latestEventTimestamp = null, bool? isPreProvisioned = null,
@@ -68,7 +68,7 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             string? failureSnapshotJson = null, bool allowTerminalReclassification = false,
             bool espSoftFailure = false, string? completionSource = null)
         {
-            var transitioned = await _storage.UpdateSessionStatusAsync(tenantId, sessionId, status,
+            var transitioned = await _storage.UpdateSessionStatusAsync(tenantId, sessionId, status, verdictPath,
                 currentPhase, failureReason, completedAt, earliestEventTimestamp,
                 latestEventTimestamp, isPreProvisioned, isUserDriven, resumedAt,
                 stalledAt, clearStalledAt, clearFailureReason, failureSource, adminMarkedAction,
@@ -142,8 +142,8 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
             => _storage.UpdateSessionDiagnosticsBlobAsync(tenantId, sessionId, blobName, destination);
 
         public Task SetSessionPreProvisionedAsync(string tenantId, string sessionId, bool isPreProvisioned,
-            SessionStatus? status = null, bool? isUserDriven = null)
-            => _storage.SetSessionPreProvisionedAsync(tenantId, sessionId, isPreProvisioned, status, isUserDriven);
+            SessionStatus? status = null, bool? isUserDriven = null, string? verdictPath = null)
+            => _storage.SetSessionPreProvisionedAsync(tenantId, sessionId, isPreProvisioned, status, isUserDriven, verdictPath);
 
         public Task UpdateSessionGeoAsync(string tenantId, string sessionId,
             string? country, string? region, string? city, string? loc)

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

@@ -18,5 +18,8 @@ namespace AutopilotMonitor.Functions.DataAccess.TableStorage
 
         public Task InitializeAsync()
             => _storage.InitializeTablesAsync();
+
+        public Task EnsureAllAsync()
+            => _storage.EnsureAllTablesAsync();
     }
 }

src/Backend/AutopilotMonitor.Functions/Functions/Admin/ReclassifyLegacySessionsFunction.cs

@@ -35,6 +35,8 @@ namespace AutopilotMonitor.Functions.Functions.Admin
         /// Query parameters:
         /// - mode: "legacy_timeouts" (default) — re-classify pre-classifier "Session timed out" Failed rows;
         ///         "pending_orphans" — resolve Pending rows superseded by a later session of the same device.
+        ///         "self_deploying_silent" — heal self-deploying (kiosk) rows parked Incomplete/AwaitingUser/Stalled
+        ///         after Device ESP provisioned to Succeeded (classifier rule 1c, 2026-08-23).
         /// - dryRun: "false" to actually write; anything else (or absent) reports only.
         /// - tenantId: optional single-tenant scope (GUID); absent = all tenants.
         /// - maxSessions: per-invocation examination cap (default 200, max 2000). Re-run to continue.
@@ -82,6 +84,8 @@ namespace AutopilotMonitor.Functions.Functions.Admin
                         await _service.ReclassifyLegacyTimeoutsAsync(tenantIdScope, dryRun, maxSessions, userEmail),
                     LegacyReclassificationService.ModePendingOrphans =>
                         await _service.ResolvePendingOrphansAsync(tenantIdScope, dryRun, maxSessions, userEmail),
+                    LegacyReclassificationService.ModeSelfDeployingSilent =>
+                        await _service.ReconcileSelfDeployingSilentAsync(tenantIdScope, dryRun, maxSessions, userEmail),
                     _ => null,
                 };
 
@@ -90,7 +94,7 @@ namespace AutopilotMonitor.Functions.Functions.Admin
                     var bad = req.CreateResponse(HttpStatusCode.BadRequest);
                     await bad.WriteAsJsonAsync(new
                     {
-                        error = $"Unknown mode '{mode}'. Valid: {LegacyReclassificationService.ModeLegacyTimeouts}, {LegacyReclassificationService.ModePendingOrphans}"
+                        error = $"Unknown mode '{mode}'. Valid: {LegacyReclassificationService.ModeLegacyTimeouts}, {LegacyReclassificationService.ModePendingOrphans}, {LegacyReclassificationService.ModeSelfDeployingSilent}"
                     });
                     return bad;
                 }

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

@@ -1,4 +1,4 @@
-using System.Linq;
+using System.Linq;
 using System.Net;
 using AutopilotMonitor.Functions.Security;
 using AutopilotMonitor.Functions.Services;
@@ -26,6 +26,7 @@ namespace AutopilotMonitor.Functions.Functions.Config
         private readonly CorporateIdentifierValidator _corporateIdentifierValidator;
         private readonly DeviceAssociationValidator _deviceAssociationValidator;
         private readonly CloudPcDeviceValidator _cloudPcDeviceValidator;
+        private readonly IntuneDeviceBindingValidator _intuneDeviceBindingValidator;
         private readonly BootstrapSessionService _bootstrapSessionService;
         private readonly KillSwitchEvaluator _killSwitchEvaluator;
 
@@ -40,6 +41,7 @@ namespace AutopilotMonitor.Functions.Functions.Config
             CorporateIdentifierValidator corporateIdentifierValidator,
             DeviceAssociationValidator deviceAssociationValidator,
             CloudPcDeviceValidator cloudPcDeviceValidator,
+            IntuneDeviceBindingValidator intuneDeviceBindingValidator,
             BootstrapSessionService bootstrapSessionService,
             KillSwitchEvaluator killSwitchEvaluator)
         {
@@ -53,6 +55,7 @@ namespace AutopilotMonitor.Functions.Functions.Config
             _corporateIdentifierValidator = corporateIdentifierValidator;
             _deviceAssociationValidator = deviceAssociationValidator;
             _cloudPcDeviceValidator = cloudPcDeviceValidator;
+            _intuneDeviceBindingValidator = intuneDeviceBindingValidator;
             _bootstrapSessionService = bootstrapSessionService;
             _killSwitchEvaluator = killSwitchEvaluator;
         }
@@ -88,7 +91,8 @@ namespace AutopilotMonitor.Functions.Functions.Config
                     _logger,
                     bootstrapSessionService: _bootstrapSessionService,
                     deviceAssociationValidator: _deviceAssociationValidator,
-                    cloudPcDeviceValidator: _cloudPcDeviceValidator
+                    cloudPcDeviceValidator: _cloudPcDeviceValidator,
+                    intuneDeviceBindingValidator: _intuneDeviceBindingValidator
                 );
 
                 if (errorResponse != null)

src/Backend/AutopilotMonitor.Functions/Functions/Diagnostics/GetDiagnosticsUploadUrlFunction.cs

@@ -1,4 +1,4 @@
-using System.Net;
+using System.Net;
 using System.Web;
 using AutopilotMonitor.Functions.Security;
 using AutopilotMonitor.Functions.Services;
@@ -53,6 +53,7 @@ namespace AutopilotMonitor.Functions.Functions.Diagnostics
         private readonly CorporateIdentifierValidator _corporateIdentifierValidator;
         private readonly DeviceAssociationValidator _deviceAssociationValidator;
         private readonly CloudPcDeviceValidator _cloudPcDeviceValidator;
+        private readonly IntuneDeviceBindingValidator _intuneDeviceBindingValidator;
         private readonly BootstrapSessionService _bootstrapSessionService;
         private readonly HostedDiagnosticsBlobService _hostedDiagnostics;
 
@@ -65,6 +66,7 @@ namespace AutopilotMonitor.Functions.Functions.Diagnostics
             CorporateIdentifierValidator corporateIdentifierValidator,
             DeviceAssociationValidator deviceAssociationValidator,
             CloudPcDeviceValidator cloudPcDeviceValidator,
+            IntuneDeviceBindingValidator intuneDeviceBindingValidator,
             BootstrapSessionService bootstrapSessionService,
             HostedDiagnosticsBlobService hostedDiagnostics)
         {
@@ -76,6 +78,7 @@ namespace AutopilotMonitor.Functions.Functions.Diagnostics
             _corporateIdentifierValidator = corporateIdentifierValidator;
             _deviceAssociationValidator = deviceAssociationValidator;
             _cloudPcDeviceValidator = cloudPcDeviceValidator;
+            _intuneDeviceBindingValidator = intuneDeviceBindingValidator;
             _bootstrapSessionService = bootstrapSessionService;
             _hostedDiagnostics = hostedDiagnostics;
         }
@@ -127,7 +130,8 @@ namespace AutopilotMonitor.Functions.Functions.Diagnostics
                     _logger,
                     bootstrapSessionService: _bootstrapSessionService,
                     deviceAssociationValidator: _deviceAssociationValidator,
-                    cloudPcDeviceValidator: _cloudPcDeviceValidator
+                    cloudPcDeviceValidator: _cloudPcDeviceValidator,
+                    intuneDeviceBindingValidator: _intuneDeviceBindingValidator
                 );
 
                 if (errorResponse != null)

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

@@ -1,4 +1,4 @@
-using System.IO;
+using System.IO;
 using System.Net;
 using AutopilotMonitor.Functions.Helpers;
 using AutopilotMonitor.Functions.Security;
@@ -51,6 +51,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
         private readonly CorporateIdentifierValidator _corporateIdentifierValidator;
         private readonly DeviceAssociationValidator _deviceAssociationValidator;
         private readonly CloudPcDeviceValidator _cloudPcDeviceValidator;
+        private readonly IntuneDeviceBindingValidator _intuneDeviceBindingValidator;
         private readonly BootstrapSessionService _bootstrapSessionService;
         private readonly KillSwitchEvaluator _killSwitchEvaluator;
         private readonly SessionDeletionGuard _deletionGuard;
@@ -69,6 +70,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
             CorporateIdentifierValidator corporateIdentifierValidator,
             DeviceAssociationValidator deviceAssociationValidator,
             CloudPcDeviceValidator cloudPcDeviceValidator,
+            IntuneDeviceBindingValidator intuneDeviceBindingValidator,
             BootstrapSessionService bootstrapSessionService,
             KillSwitchEvaluator killSwitchEvaluator,
             SessionDeletionGuard deletionGuard)
@@ -86,6 +88,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
             _corporateIdentifierValidator = corporateIdentifierValidator;
             _deviceAssociationValidator = deviceAssociationValidator;
             _cloudPcDeviceValidator = cloudPcDeviceValidator;
+            _intuneDeviceBindingValidator = intuneDeviceBindingValidator;
             _bootstrapSessionService = bootstrapSessionService;
             _killSwitchEvaluator = killSwitchEvaluator;
             _deletionGuard = deletionGuard;
@@ -116,7 +119,8 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
                     _logger,
                     bootstrapSessionService: _bootstrapSessionService,
                     deviceAssociationValidator: _deviceAssociationValidator,
-                    cloudPcDeviceValidator: _cloudPcDeviceValidator);
+                    cloudPcDeviceValidator: _cloudPcDeviceValidator,
+                    intuneDeviceBindingValidator: _intuneDeviceBindingValidator);
 
                 if (errorResponse != null) return AsOutput(errorResponse);
 

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

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Collections.Generic;
 using System.Linq;
 using System.Net;
@@ -35,6 +35,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
         private readonly CorporateIdentifierValidator _corporateIdentifierValidator;
         private readonly DeviceAssociationValidator _deviceAssociationValidator;
         private readonly CloudPcDeviceValidator _cloudPcDeviceValidator;
+        private readonly IntuneDeviceBindingValidator _intuneDeviceBindingValidator;
         private readonly TelemetryClient _telemetryClient;
         private readonly BootstrapSessionService _bootstrapSessionService;
         private readonly ISessionRepository _sessionRepo;
@@ -49,6 +50,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
             CorporateIdentifierValidator corporateIdentifierValidator,
             DeviceAssociationValidator deviceAssociationValidator,
             CloudPcDeviceValidator cloudPcDeviceValidator,
+            IntuneDeviceBindingValidator intuneDeviceBindingValidator,
             TelemetryClient telemetryClient,
             BootstrapSessionService bootstrapSessionService,
             ISessionRepository sessionRepo,
@@ -62,6 +64,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
             _corporateIdentifierValidator = corporateIdentifierValidator;
             _deviceAssociationValidator = deviceAssociationValidator;
             _cloudPcDeviceValidator = cloudPcDeviceValidator;
+            _intuneDeviceBindingValidator = intuneDeviceBindingValidator;
             _telemetryClient = telemetryClient;
             _bootstrapSessionService = bootstrapSessionService;
             _sessionRepo = sessionRepo;
@@ -98,7 +101,8 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
                     _logger,
                     bootstrapSessionService: _bootstrapSessionService,
                     deviceAssociationValidator: _deviceAssociationValidator,
-                    cloudPcDeviceValidator: _cloudPcDeviceValidator
+                    cloudPcDeviceValidator: _cloudPcDeviceValidator,
+                    intuneDeviceBindingValidator: _intuneDeviceBindingValidator
                 );
 
                 if (errorResponse != null)

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

@@ -0,0 +1,232 @@
+using System.Globalization;
+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>
+    /// Operator-only verdict calibration matrix (docs/backend/verdict-calibration.md): per verdict
+    /// path, how many sessions it produced in the window, its share, how often it was overridden
+    /// (admin / late completion / other), the 7-day re-enrollment proxy, and a 7d-vs-28d trend.
+    /// Platform scope ("global" partition) or one tenant via ?tenantId=. Serves the admin
+    /// metrics page and the <c>get_verdict_calibration</c> MCP tool. Never customer-facing —
+    /// these are platform-internal classifier diagnostics, not product features.
+    /// </summary>
+    public class GetVerdictCalibrationFunction
+    {
+        private readonly ILogger<GetVerdictCalibrationFunction> _logger;
+        private readonly IMetricsRepository _metricsRepo;
+        private readonly IHardwareRejectionNotificationTracker _tracker;
+
+        public GetVerdictCalibrationFunction(
+            ILogger<GetVerdictCalibrationFunction> logger,
+            IMetricsRepository metricsRepo,
+            IHardwareRejectionNotificationTracker tracker)
+        {
+            _logger = logger;
+            _metricsRepo = metricsRepo;
+            _tracker = tracker;
+        }
+
+        [Function("GetVerdictCalibration")]
+        public async Task<HttpResponseData> Run(
+            [HttpTrigger(AuthorizationLevel.Anonymous, "get", Route = "global/metrics/verdict-calibration")] HttpRequestData req)
+        {
+            try
+            {
+                // Authentication + GlobalReadOrAdmin authorization enforced by PolicyEnforcementMiddleware
+                var query = System.Web.HttpUtility.ParseQueryString(req.Url.Query);
+                var tenantIdFilter = query["tenantId"];
+                var days = VerdictCalibrationResponse.ClampDays(query["days"]);
+                var partition = "global";
+                if (!string.IsNullOrWhiteSpace(tenantIdFilter))
+                {
+                    if (!Guid.TryParse(tenantIdFilter, out var parsed))
+                    {
+                        var bad = req.CreateResponse(HttpStatusCode.BadRequest);
+                        await bad.WriteAsJsonAsync(new { success = false, message = "tenantId must be a GUID" });
+                        return bad;
+                    }
+                    partition = parsed.ToString("D");
+                }
+
+                var today = DateTime.UtcNow.Date;
+                // One read covers both the requested window and the trend horizon (7d window + 28d baseline).
+                var readDays = Math.Max(days, VerdictCalibrationResponse.TrendHorizonDays);
+                var daily = await _metricsRepo.GetVerdictCalibrationAggregatesAsync(
+                    partition, VerdictCalibrationResponse.InclusiveWindowStart(today, readDays), today);
+                // Active drift episodes of this partition (tracker keyspace; fail-soft empty).
+                var alerts = await _tracker.GetVerdictCalibrationAlertsAsync(partition);
+
+                var response = req.CreateResponse(HttpStatusCode.OK);
+                await response.WriteAsJsonAsync(VerdictCalibrationResponse.Build(daily, partition, today, days, alerts));
+                return response;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(ex, "Error fetching verdict calibration");
+                var errorResponse = req.CreateResponse(HttpStatusCode.InternalServerError);
+                await errorResponse.WriteAsJsonAsync(new { success = false, message = "Internal server error" });
+                return errorResponse;
+            }
+        }
+    }
+
+    /// <summary>Pure response builder — internal static so the matrix arithmetic is pinned by tests.</summary>
+    internal static class VerdictCalibrationResponse
+    {
+        internal const int DefaultWindowDays = 30;
+        internal const int MaxWindowDays = 180; // aggregate retention
+        internal const int TrendWindowDays = 7;
+        internal const int TrendBaselineDays = 28;
+        internal const int TrendHorizonDays = TrendWindowDays + TrendBaselineDays;
+
+        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 inclusive).</summary>
+        internal static DateTime InclusiveWindowStart(DateTime today, int days) => today.AddDays(-(days - 1));
+
+        internal sealed class PathRow
+        {
+            public string VerdictPath { get; set; } = string.Empty;
+            public string Status { get; set; } = string.Empty;
+            public int Count { get; set; }
+            public double SharePct { get; set; }
+            public int DerivedCount { get; set; }
+            public int Eligible7d { get; set; }
+            public int ReEnrolled7d { get; set; }
+            /// <summary>Null below the minimum eligible count — never a rate on a handful of sessions.</summary>
+            public double? ReEnrollRatePct { get; set; }
+            public int OverriddenByAdmin { get; set; }
+            public int OverriddenByLateCompletion { get; set; }
+            public int OverriddenOther { get; set; }
+            public TrendWindow Window7 { get; set; } = new();
+            public TrendWindow Baseline28 { get; set; } = new();
+            /// <summary>Window share ÷ baseline share; null when the baseline share is 0 (a new path has no finite lift — never invented).</summary>
+            public double? Lift { get; set; }
+        }
+
+        internal sealed class TrendWindow
+        {
+            public int Count { get; set; }
+            public int Sessions { get; set; }
+            public double SharePct { get; set; }
+        }
+
+        /// <summary>Minimum eligible sessions before a re-enrollment rate is stated (truthfulness: no rate on n&lt;20).</summary>
+        internal const int MinEligibleForRate = 20;
+
+        internal static object Build(
+            IReadOnlyList<VerdictCalibrationDailyAggregate> daily, string partition, DateTime today, int days,
+            IReadOnlyList<VerdictCalibrationAlert>? alerts = null)
+        {
+            var windowStart = Key(InclusiveWindowStart(today, days));
+            var trendWindowStart = Key(InclusiveWindowStart(today, TrendWindowDays));
+            var baselineStart = Key(InclusiveWindowStart(today, TrendHorizonDays));
+
+            var rows = new Dictionary<(string Path, string Status), PathRow>();
+            int sessions = 0, terminal = 0, derived = 0;
+            int trendSessions = 0, baselineSessions = 0;
+            DateTime? computedAt = null;
+            var versions = new HashSet<int>();
+
+            PathRow Row(VerdictCalibrationBucket b)
+            {
+                var key = (b.VerdictPath, b.Status);
+                if (!rows.TryGetValue(key, out var row))
+                {
+                    row = new PathRow { VerdictPath = b.VerdictPath, Status = b.Status };
+                    rows[key] = row;
+                }
+                return row;
+            }
+
+            foreach (var day in daily)
+            {
+                var inWindow = string.CompareOrdinal(day.Date, windowStart) >= 0;
+                var inTrendWindow = string.CompareOrdinal(day.Date, trendWindowStart) >= 0;
+                var inBaseline = !inTrendWindow && string.CompareOrdinal(day.Date, baselineStart) >= 0;
+
+                if (inWindow)
+                {
+                    sessions += day.SessionCount;
+                    terminal += day.TerminalSessionCount;
+                    versions.Add(day.Version);
+                    if (computedAt == null || day.ComputedAt > computedAt) computedAt = day.ComputedAt;
+                }
+                if (inTrendWindow) trendSessions += day.SessionCount;
+                if (inBaseline) baselineSessions += day.SessionCount;
+
+                foreach (var b in day.Buckets)
+                {
+                    // Overrides are attributed to the prior path and can exist on a bucket with Count=0.
+                    var row = Row(b);
+                    if (inWindow)
+                    {
+                        row.Count += b.Count;
+                        row.DerivedCount += b.DerivedCount;
+                        row.Eligible7d += b.Eligible7d;
+                        row.ReEnrolled7d += b.ReEnrolled7d;
+                        row.OverriddenByAdmin += b.OverriddenByAdmin;
+                        row.OverriddenByLateCompletion += b.OverriddenByLateCompletion;
+                        row.OverriddenOther += b.OverriddenOther;
+                        derived += b.DerivedCount;
+                    }
+                    if (inTrendWindow) row.Window7.Count += b.Count;
+                    if (inBaseline) row.Baseline28.Count += b.Count;
+                }
+            }
+
+            var list = rows.Values
+                .Where(r => r.Count > 0 || r.OverriddenByAdmin + r.OverriddenByLateCompletion + r.OverriddenOther > 0)
+                .OrderByDescending(r => r.Count)
+                .ThenBy(r => r.VerdictPath, StringComparer.Ordinal)
+                .ThenBy(r => r.Status, StringComparer.Ordinal)
+                .ToList();
+
+            foreach (var r in list)
+            {
+                r.SharePct = Pct(r.Count, sessions);
+                r.ReEnrollRatePct = r.Eligible7d >= MinEligibleForRate ? Pct(r.ReEnrolled7d, r.Eligible7d) : null;
+                r.Window7.Sessions = trendSessions;
+                r.Window7.SharePct = Pct(r.Window7.Count, trendSessions);
+                r.Baseline28.Sessions = baselineSessions;
+                r.Baseline28.SharePct = Pct(r.Baseline28.Count, baselineSessions);
+                r.Lift = r.Baseline28.SharePct > 0 && trendSessions > 0
+                    ? Math.Round(r.Window7.SharePct / r.Baseline28.SharePct, 2)
+                    : null;
+            }
+
+            return new
+            {
+                success = true,
+                tenantId = partition,
+                windowDays = days,
+                windowStart,
+                windowEnd = Key(today),
+                computedAt,
+                versions = versions.OrderBy(v => v).ToArray(),
+                totals = new { sessions, terminal, derived, days = daily.Count(d => string.CompareOrdinal(d.Date, windowStart) >= 0) },
+                trend = new { windowDays = TrendWindowDays, baselineDays = TrendBaselineDays, windowSessions = trendSessions, baselineSessions },
+                paths = list,
+                // Wording contract shared with the rule radar: a dimension is correlated, not causal.
+                alerts = (alerts ?? Array.Empty<VerdictCalibrationAlert>())
+                    .OrderByDescending(a => a.FirstNotifiedAt)
+                    .ToList(),
+            };
+        }
+
+        private static string Key(DateTime d) => d.ToString("yyyy-MM-dd", CultureInfo.InvariantCulture);
+
+        private static double Pct(int part, int whole) => whole <= 0 ? 0 : Math.Round(100.0 * part / whole, 1);
+    }
+}

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

@@ -49,6 +49,7 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
                     tenantId,
                     sessionId,
                     SessionStatus.Failed,
+                    VerdictPaths.ManualFailed,
                     currentPhase: null, // Keep current phase
                     failureReason: "Manually marked as failed by administrator",
                     failureSource: "manual",

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

@@ -49,6 +49,7 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
                     tenantId,
                     sessionId,
                     SessionStatus.Succeeded,
+                    VerdictPaths.ManualSucceeded,
                     currentPhase: null, // Keep current phase
                     failureReason: "Manually marked as succeeded by administrator",
                     adminMarkedAction: "Succeeded"

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

@@ -1,4 +1,4 @@
-using AutopilotMonitor.Shared;
+using AutopilotMonitor.Shared;
 using System.Net;
 using AutopilotMonitor.Functions.Helpers;
 using AutopilotMonitor.Functions.Security;
@@ -32,6 +32,7 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
         private readonly CorporateIdentifierValidator _corporateIdentifierValidator;
         private readonly DeviceAssociationValidator _deviceAssociationValidator;
         private readonly CloudPcDeviceValidator _cloudPcDeviceValidator;
+        private readonly IntuneDeviceBindingValidator _intuneDeviceBindingValidator;
         private readonly BootstrapSessionService _bootstrapSessionService;
         private readonly WebhookNotificationService _webhookNotificationService;
         private readonly SessionDeletionGuard _deletionGuard;
@@ -48,6 +49,7 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
             CorporateIdentifierValidator corporateIdentifierValidator,
             DeviceAssociationValidator deviceAssociationValidator,
             CloudPcDeviceValidator cloudPcDeviceValidator,
+            IntuneDeviceBindingValidator intuneDeviceBindingValidator,
             BootstrapSessionService bootstrapSessionService,
             WebhookNotificationService webhookNotificationService,
             SessionDeletionGuard deletionGuard)
@@ -62,6 +64,7 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
             _corporateIdentifierValidator = corporateIdentifierValidator;
             _deviceAssociationValidator = deviceAssociationValidator;
             _cloudPcDeviceValidator = cloudPcDeviceValidator;
+            _intuneDeviceBindingValidator = intuneDeviceBindingValidator;
             _bootstrapSessionService = bootstrapSessionService;
             _webhookNotificationService = webhookNotificationService;
             _deletionGuard = deletionGuard;
@@ -106,7 +109,8 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
                     registration.SessionId,
                     bootstrapSessionService: _bootstrapSessionService,
                     deviceAssociationValidator: _deviceAssociationValidator,
-                    cloudPcDeviceValidator: _cloudPcDeviceValidator
+                    cloudPcDeviceValidator: _cloudPcDeviceValidator,
+                    intuneDeviceBindingValidator: _intuneDeviceBindingValidator
                 );
 
                 if (errorResponse2 != null)
@@ -305,7 +309,7 @@ namespace AutopilotMonitor.Functions.Functions.Sessions
                                  $"enrollment session at {registration.StartedAt:yyyy-MM-dd HH:mm} UTC before this one " +
                                  "reached a terminal state.";
                     var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                        registration.TenantId, stale.SessionId, SessionStatus.Incomplete,
+                        registration.TenantId, stale.SessionId, SessionStatus.Incomplete, VerdictPaths.RegisterSuperseded,
                         failureReason: reason, failureSource: "superseded_by_reregistration");
                     if (!transitioned)
                         continue;

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

@@ -0,0 +1,281 @@
+using System;
+using System.Collections.Generic;
+using System.Linq;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Helpers;
+
+/// <summary>
+/// One drift finding of the verdict-calibration radar (pure output; the maintenance partial
+/// turns it into a tracker episode + ops event).
+/// </summary>
+public class VerdictCalibrationFinding
+{
+    public string Kind { get; set; } = string.Empty;
+    public string VerdictPath { get; set; } = string.Empty;
+    public string Status { get; set; } = string.Empty;
+    public int WindowHitCount { get; set; }
+    public int WindowSessionCount { get; set; }
+    public int BaselineHitCount { get; set; }
+    public int BaselineSessionCount { get; set; }
+    public double WindowRatePct { get; set; }
+    public double BaselineRatePct { get; set; }
+    public double? Lift { get; set; }
+    public string WindowStartDate { get; set; } = string.Empty;
+    public string WindowEndDate { get; set; } = string.Empty;
+}
+
+/// <summary>
+/// Pure drift detection over one tenant's verdict-calibration daily rows
+/// (docs/backend/verdict-calibration.md) — the same statistics as the rule-regression radar
+/// (7-day window vs the prior 28 days, lift ≥2×, Wilson-separated via
+/// <see cref="MetricsMath.RateIncreaseSeparated"/>, small-n gates) applied to three questions:
+/// <list type="bullet">
+/// <item><b>share_regression</b> — one verdict path's share of all sessions doubled. Points at a rule
+/// (or an agent path) whose conditions suddenly match much more often.</item>
+/// <item><b>silence_share_regression</b> — the share of sessions the backend had to decide because
+/// the agent went silent (sweep:* + maxlife:*) doubled. An agent-liveness signal, not a classifier one.</item>
+/// <item><b>evidence_gap</b> — absolute gate: the pure fallthrough rule (r6) decides ≥20 % of the
+/// window's classifier verdicts. The evidence the other rules need is not arriving.</item>
+/// </list>
+/// Derived (<c>legacy:*</c>) paths are excluded from all three — pre-instrumentation attribution
+/// is too weak to alert on. Agent-declared paths (<c>agent:*</c>, <c>register:*</c>) are excluded
+/// from the per-path share regression: their share tracks the customer's workflow mix (a
+/// WhiteGlove rollout week doubles <c>agent:whiteglove_pending</c>; a Continue-Anyway tenant
+/// lives on <c>agent:complete_soft</c>) — real signals, but not about the BACKEND classifier
+/// this radar calibrates. First production pass (2026-08-23) fired exactly those three. Re-arm
+/// mirrors the rule radar: share back under 1.5× baseline or the path stops occurring; the
+/// evidence gap re-arms under 15 %.
+/// </summary>
+public static class VerdictCalibrationRadar
+{
+    public const int WindowDays = RuleRegressionRadar.WindowDays;
+    public const int BaselineDays = RuleRegressionRadar.BaselineDays;
+    public const int MinWindowHits = RuleRegressionRadar.MinWindowHitSessions;
+    public const int MinWindowSessions = RuleRegressionRadar.MinWindowSessions;
+    public const double MinLift = RuleRegressionRadar.MinLift;
+    public const double ReArmLiftFactor = RuleRegressionRadar.ReArmLiftFactor;
+
+    /// <summary>Evidence gap fires at ≥20 % r6 share of the window's classifier verdicts (≥20 verdicts) and re-arms under 15 %.</summary>
+    public const double EvidenceGapFirePct = 20.0;
+    public const double EvidenceGapReArmPct = 15.0;
+    public const int EvidenceGapMinVerdicts = 20;
+
+    public const string SilenceGroupPath = "sweep+maxlife";
+    public const string EvidenceGapPath = "r6";
+    public const string GroupStatus = "*";
+
+    private const string LegacyOrigin = "legacy";
+
+    /// <summary>
+    /// True when a path is eligible for the per-path share regression: backend-decided paths only
+    /// (sweep / maxlife / late / retro / rule / manual / ingest). Agent-declared and registration
+    /// paths mirror customer workflow, legacy paths are derived.
+    /// </summary>
+    public static bool IsShareRegressionEligible(string path)
+    {
+        var origin = VerdictPaths.Origin(path);
+        return origin != LegacyOrigin
+            && origin != "agent"
+            && origin != "register";
+    }
+
+    public static List<VerdictCalibrationFinding> Evaluate(IReadOnlyList<VerdictCalibrationDailyAggregate> tenantRows, DateTime targetDateUtc)
+    {
+        var findings = new List<VerdictCalibrationFinding>();
+        var horizon = Summarize(tenantRows, targetDateUtc);
+        var windowStart = targetDateUtc.Date.AddDays(-(WindowDays - 1)).ToString("yyyy-MM-dd");
+        var windowEnd = targetDateUtc.Date.ToString("yyyy-MM-dd");
+
+        // 1. Per-path share regression.
+        foreach (var pair in horizon.Paths)
+        {
+            if (!IsShareRegressionEligible(pair.Key.Path)) continue;
+            var f = TryShareRegression(VerdictCalibrationAlertKinds.ShareRegression, pair.Key.Path, pair.Key.Status,
+                pair.Value.WindowHits, horizon.WindowSessions, pair.Value.BaselineHits, horizon.BaselineSessions, windowStart, windowEnd);
+            if (f != null) findings.Add(f);
+        }
+
+        // 2. Silence-share regression (group).
+        {
+            var f = TryShareRegression(VerdictCalibrationAlertKinds.SilenceShareRegression, SilenceGroupPath, GroupStatus,
+                horizon.SilenceWindowHits, horizon.WindowSessions, horizon.SilenceBaselineHits, horizon.BaselineSessions, windowStart, windowEnd);
+            if (f != null) findings.Add(f);
+        }
+
+        // 3. Evidence gap (absolute).
+        if (horizon.ClassifierWindowVerdicts >= EvidenceGapMinVerdicts)
+        {
+            var pct = 100.0 * horizon.R6WindowHits / horizon.ClassifierWindowVerdicts;
+            if (pct >= EvidenceGapFirePct)
+            {
+                findings.Add(new VerdictCalibrationFinding
+                {
+                    Kind = VerdictCalibrationAlertKinds.EvidenceGap,
+                    VerdictPath = EvidenceGapPath,
+                    Status = GroupStatus,
+                    WindowHitCount = horizon.R6WindowHits,
+                    WindowSessionCount = horizon.ClassifierWindowVerdicts,
+                    BaselineHitCount = horizon.R6BaselineHits,
+                    BaselineSessionCount = horizon.ClassifierBaselineVerdicts,
+                    WindowRatePct = Math.Round(pct, 1),
+                    BaselineRatePct = horizon.ClassifierBaselineVerdicts > 0
+                        ? Math.Round(100.0 * horizon.R6BaselineHits / horizon.ClassifierBaselineVerdicts, 1) : 0,
+                    Lift = null,
+                    WindowStartDate = windowStart,
+                    WindowEndDate = windowEnd,
+                });
+            }
+        }
+
+        return findings
+            .OrderByDescending(f => f.WindowRatePct)
+            .ThenBy(f => f.Kind, StringComparer.Ordinal)
+            .ThenBy(f => f.VerdictPath, StringComparer.Ordinal)
+            .ToList();
+    }
+
+    /// <summary>True when an ACTIVE alert may re-arm given the current horizon sums (or when its path is no longer eligible — episodes from an earlier, broader gate clear themselves).</summary>
+    public static bool ShouldReArm(VerdictCalibrationAlert alert, IReadOnlyList<VerdictCalibrationDailyAggregate> tenantRows, DateTime targetDateUtc)
+    {
+        if (alert.Kind == VerdictCalibrationAlertKinds.ShareRegression && !IsShareRegressionEligible(alert.VerdictPath))
+            return true;
+
+        var horizon = Summarize(tenantRows, targetDateUtc);
+        switch (alert.Kind)
+        {
+            case VerdictCalibrationAlertKinds.EvidenceGap:
+                if (horizon.ClassifierWindowVerdicts < EvidenceGapMinVerdicts) return true;
+                return 100.0 * horizon.R6WindowHits / horizon.ClassifierWindowVerdicts < EvidenceGapReArmPct;
+
+            case VerdictCalibrationAlertKinds.SilenceShareRegression:
+                return ReArmByRate(horizon.SilenceWindowHits, horizon.WindowSessions, horizon.SilenceBaselineHits, horizon.BaselineSessions);
+
+            default:
+                horizon.Paths.TryGetValue((alert.VerdictPath, alert.Status), out var sums);
+                return ReArmByRate(sums.WindowHits, horizon.WindowSessions, sums.BaselineHits, horizon.BaselineSessions);
+        }
+    }
+
+    /// <summary>Current numbers for an ongoing episode (refresh without re-firing).</summary>
+    public static (int WindowHits, int WindowSessions, int BaselineHits, int BaselineSessions) CurrentSums(
+        VerdictCalibrationAlert alert, IReadOnlyList<VerdictCalibrationDailyAggregate> tenantRows, DateTime targetDateUtc)
+    {
+        var horizon = Summarize(tenantRows, targetDateUtc);
+        switch (alert.Kind)
+        {
+            case VerdictCalibrationAlertKinds.EvidenceGap:
+                return (horizon.R6WindowHits, horizon.ClassifierWindowVerdicts, horizon.R6BaselineHits, horizon.ClassifierBaselineVerdicts);
+            case VerdictCalibrationAlertKinds.SilenceShareRegression:
+                return (horizon.SilenceWindowHits, horizon.WindowSessions, horizon.SilenceBaselineHits, horizon.BaselineSessions);
+            default:
+                horizon.Paths.TryGetValue((alert.VerdictPath, alert.Status), out var sums);
+                return (sums.WindowHits, horizon.WindowSessions, sums.BaselineHits, horizon.BaselineSessions);
+        }
+    }
+
+    private static bool ReArmByRate(int windowHits, int windowSessions, int baselineHits, int baselineSessions)
+    {
+        if (windowHits == 0 || windowSessions == 0) return true;
+        if (baselineSessions == 0 || baselineHits == 0) return false; // zero-baseline episode re-arms only via hits-stopped
+        var windowRate = (double)windowHits / windowSessions;
+        var baselineRate = (double)baselineHits / baselineSessions;
+        return windowRate < ReArmLiftFactor * baselineRate;
+    }
+
+    private static VerdictCalibrationFinding? TryShareRegression(string kind, string path, string status,
+        int windowHits, int windowSessions, int baselineHits, int baselineSessions, string windowStart, string windowEnd)
+    {
+        if (windowHits < MinWindowHits) return null;
+        if (windowSessions < MinWindowSessions) return null;
+        if (baselineSessions <= 0) return null;
+
+        var windowRate = (double)windowHits / windowSessions;
+        var baselineRate = (double)baselineHits / baselineSessions;
+        if (windowRate < MinLift * baselineRate) return null;
+        if (!MetricsMath.RateIncreaseSeparated(windowHits, windowSessions, baselineHits, baselineSessions)) return null;
+
+        return new VerdictCalibrationFinding
+        {
+            Kind = kind,
+            VerdictPath = path,
+            Status = status,
+            WindowHitCount = windowHits,
+            WindowSessionCount = windowSessions,
+            BaselineHitCount = baselineHits,
+            BaselineSessionCount = baselineSessions,
+            WindowRatePct = Math.Round(windowRate * 100, 1),
+            BaselineRatePct = Math.Round(baselineRate * 100, 1),
+            Lift = baselineRate > 0 ? Math.Round(windowRate / baselineRate, 1) : null,
+            WindowStartDate = windowStart,
+            WindowEndDate = windowEnd,
+        };
+    }
+
+    internal readonly record struct PathSums(int WindowHits, int BaselineHits);
+
+    internal sealed class Horizon
+    {
+        public int WindowSessions;
+        public int BaselineSessions;
+        public int SilenceWindowHits;
+        public int SilenceBaselineHits;
+        public int R6WindowHits;
+        public int R6BaselineHits;
+        public int ClassifierWindowVerdicts;
+        public int ClassifierBaselineVerdicts;
+        public readonly Dictionary<(string Path, string Status), PathSums> Paths = new();
+    }
+
+    /// <summary>
+    /// Sums the daily rows into the trailing window [target−6d, target] and the prior baseline
+    /// [target−34d, target−7d] (ISO-string date compare, rows outside the horizon ignored).
+    /// Silence group = sweep:* + maxlife:* (incl. non-rule sweep paths like sweep:stalled — the
+    /// question is "how often did the backend have to decide", not which rule did). Classifier
+    /// verdicts = stamped rule paths (<see cref="VerdictPaths.IsClassifierPath"/>); r6 = their
+    /// fallthrough subset.
+    /// </summary>
+    internal static Horizon Summarize(IReadOnlyList<VerdictCalibrationDailyAggregate> rows, 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");
+
+        var h = new Horizon();
+        foreach (var row in rows)
+        {
+            if (string.CompareOrdinal(row.Date, target) > 0) continue;
+            bool inWindow;
+            if (string.CompareOrdinal(row.Date, windowStart) >= 0) inWindow = true;
+            else if (string.CompareOrdinal(row.Date, baselineStart) >= 0) inWindow = false;
+            else continue;
+
+            if (inWindow) h.WindowSessions += row.SessionCount; else h.BaselineSessions += row.SessionCount;
+
+            foreach (var b in row.Buckets)
+            {
+                if (b.Count == 0) continue;
+                var key = (b.VerdictPath, b.Status);
+                h.Paths.TryGetValue(key, out var sums);
+                h.Paths[key] = inWindow
+                    ? new PathSums(sums.WindowHits + b.Count, sums.BaselineHits)
+                    : new PathSums(sums.WindowHits, sums.BaselineHits + b.Count);
+
+                var origin = VerdictPaths.Origin(b.VerdictPath);
+                if (origin == VerdictPaths.OriginSweep || origin == VerdictPaths.OriginMaxLifetime)
+                {
+                    if (inWindow) h.SilenceWindowHits += b.Count; else h.SilenceBaselineHits += b.Count;
+                }
+                if (VerdictPaths.IsClassifierPath(b.VerdictPath))
+                {
+                    if (inWindow) h.ClassifierWindowVerdicts += b.Count; else h.ClassifierBaselineVerdicts += b.Count;
+                    if (b.VerdictPath.EndsWith(":" + ClassifierRules.R6Fallthrough, StringComparison.Ordinal))
+                    {
+                        if (inWindow) h.R6WindowHits += b.Count; else h.R6BaselineHits += b.Count;
+                    }
+                }
+            }
+        }
+        return h;
+    }
+}

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

@@ -0,0 +1,127 @@
+using System;
+using AutopilotMonitor.Shared.Models;
+
+namespace AutopilotMonitor.Functions.Helpers
+{
+    /// <summary>
+    /// Read-side attribution for sessions written BEFORE <see cref="SessionSummary.VerdictPath"/>
+    /// was stamped. No mass backfill: the calibration aggregate calls <see cref="Derive"/> and
+    /// counts the result with <c>Derived=true</c>, so pre-instrumentation history still shows up
+    /// in the matrix and is visibly weaker. The derivation reads the fixed reason literals the
+    /// writers emit (classifier rule prefixes, the max-lifetime / retro suffixes, the manual and
+    /// rule attributions) — when the origin is ambiguous but the rule is known the path is
+    /// <c>legacy:{rule}</c>; when nothing is attributable it is <see cref="VerdictPaths.LegacyUnknown"/>.
+    /// A stamped row is returned verbatim with <c>Derived=false</c>.
+    /// </summary>
+    public static class VerdictPathDerivation
+    {
+        private const string OriginLegacy = "legacy";
+
+        // Literals are owned by EnrollmentTimeoutClassifier / the writers; pinned by
+        // VerdictPathDerivationTests so a wording change there fails here, not silently.
+        private const string R1Prefix = "Enrollment reported an explicit failure before timeout";
+        private const string R1bAwaitingPrefix = "Pre-provisioning (WhiteGlove) completed; no user sign-in since Part 2 resumed";
+        private const string R1bSucceededPrefix = "Reconciled at timeout: pre-provisioning (WhiteGlove Part 1) completed";
+        private const string R1cPrefix = "Reconciled: self-deploying profile";
+        private const string R2Prefix = "Reconciled at timeout: Account Setup completed";
+        private const string R3Prefix = "Agent emergency break fired";
+        private const string R4Prefix = "Reconciled at timeout: user completed setup";
+        private const string R5AwaitingPrefix = "Device Setup completed; awaiting user / Account Setup phase";
+        private const string R5IncompletePrefix = "No completion signal within";
+        private const string R6Prefix = "No Device Setup completion or explicit failure signal observed before timeout";
+        private const string SupersededPrefix = "Superseded by session";
+        private const string SweepStalledPrefix = "Agent silent for";
+        private const string AgentStallProbePrefix = "Agent reported stall after";
+        private const string EspFallbackLiteral = "ESP failure (backend fallback)";
+        /// <summary>Pre-classifier blanket timeout verdict (LegacyReclassificationService) — never agent-reported.</summary>
+        private const string LegacyBlanketTimeoutPrefix = "Session timed out";
+        private const string MaxLifetimeSuffix = "max-lifetime watchdog";
+        private const string RetroSuffix = "Retro-reclassified";
+
+        public static (string Path, bool Derived) Derive(SessionSummary s)
+        {
+            if (s == null) throw new ArgumentNullException(nameof(s));
+            if (!string.IsNullOrWhiteSpace(s.VerdictPath))
+                return (s.VerdictPath!, false);
+
+            return (DeriveLegacy(s), true);
+        }
+
+        private static string DeriveLegacy(SessionSummary s)
+        {
+            var failureReason = s.FailureReason ?? string.Empty;
+            var reconcileReason = s.ReconcileReason ?? string.Empty;
+            var failureSource = s.FailureSource ?? string.Empty;
+
+            // Unambiguous attributions first — independent of status text.
+            if (string.Equals(s.AdminMarkedAction, "Succeeded", StringComparison.OrdinalIgnoreCase))
+                return VerdictPaths.ManualSucceeded;
+            if (string.Equals(s.AdminMarkedAction, "Failed", StringComparison.OrdinalIgnoreCase))
+                return VerdictPaths.ManualFailed;
+            if (failureSource.StartsWith("rule:", StringComparison.OrdinalIgnoreCase))
+                return VerdictPaths.Compose(VerdictPaths.OriginRule, failureSource.Substring(5));
+
+            switch (s.Status)
+            {
+                case SessionStatus.Succeeded:
+                    if (reconcileReason.Length == 0
+                        || reconcileReason.StartsWith("Late completion report received", StringComparison.Ordinal))
+                        return s.EspSoftFailure ? VerdictPaths.AgentCompleteSoft : VerdictPaths.AgentComplete;
+                    if (Starts(reconcileReason, R1bSucceededPrefix)) return Classifier(reconcileReason, ClassifierRules.R1bWhiteGloveSucceeded);
+                    if (Starts(reconcileReason, R1cPrefix)) return Classifier(reconcileReason, ClassifierRules.R1cSelfDeploying);
+                    if (Starts(reconcileReason, R2Prefix)) return Classifier(reconcileReason, ClassifierRules.R2AccountSetupComplete);
+                    if (Starts(reconcileReason, R4Prefix)) return Classifier(reconcileReason, ClassifierRules.R4DesktopHello);
+                    return VerdictPaths.LegacyUnknown;
+
+                case SessionStatus.Failed:
+                    if (Starts(failureReason, R1Prefix)) return Classifier(failureReason, ClassifierRules.R1ExplicitFailure);
+                    if (string.Equals(failureSource, "max_lifetime_watchdog", StringComparison.OrdinalIgnoreCase))
+                        return VerdictPaths.Compose(VerdictPaths.OriginMaxLifetime, ClassifierRules.R1ExplicitFailure);
+                    if (failureReason.Contains(EspFallbackLiteral, StringComparison.Ordinal)) return VerdictPaths.AgentEspFailureFallback;
+                    if (Starts(failureReason, LegacyBlanketTimeoutPrefix)) return VerdictPaths.LegacyUnknown;
+                    if (failureReason.Length > 0 && failureSource.Length == 0) return VerdictPaths.AgentFailed;
+                    return VerdictPaths.LegacyUnknown;
+
+                case SessionStatus.Incomplete:
+                    if (Starts(failureReason, R3Prefix)) return Classifier(failureReason, ClassifierRules.R3EmergencyBreak);
+                    if (Starts(failureReason, R5IncompletePrefix)) return Classifier(failureReason, ClassifierRules.R5DeviceSetupIncomplete);
+                    if (Starts(failureReason, R6Prefix)) return Classifier(failureReason, ClassifierRules.R6Fallthrough);
+                    if (Starts(failureReason, SupersededPrefix)) return VerdictPaths.Compose(OriginLegacy, "superseded");
+                    return VerdictPaths.LegacyUnknown;
+
+                case SessionStatus.AwaitingUser:
+                    if (Starts(failureReason, R1bAwaitingPrefix)) return VerdictPaths.Compose(OriginLegacy, "wg_awaiting");
+                    if (Starts(failureReason, R5AwaitingPrefix)) return Classifier(failureReason, ClassifierRules.R5DeviceSetupAwaiting);
+                    return VerdictPaths.LegacyUnknown;
+
+                case SessionStatus.Stalled:
+                    if (Starts(failureReason, SweepStalledPrefix)) return VerdictPaths.SweepStalled;
+                    if (Starts(failureReason, AgentStallProbePrefix)) return VerdictPaths.AgentStallProbe;
+                    return VerdictPaths.LegacyUnknown;
+
+                case SessionStatus.Pending:
+                    return VerdictPaths.AgentWhiteGlovePending;
+
+                default:
+                    return VerdictPaths.LegacyUnknown;
+            }
+        }
+
+        /// <summary>
+        /// Classifier verdict with a known rule: the origin is recoverable only from the suffixes
+        /// the max-lifetime and retro writers append; sweep / late-reconcile share the bare text
+        /// and stay <c>legacy:{rule}</c>.
+        /// </summary>
+        private static string Classifier(string reason, string rule)
+        {
+            if (reason.Contains(MaxLifetimeSuffix, StringComparison.Ordinal))
+                return VerdictPaths.Compose(VerdictPaths.OriginMaxLifetime, rule);
+            if (reason.Contains(RetroSuffix, StringComparison.Ordinal))
+                return VerdictPaths.Compose(VerdictPaths.OriginRetro, rule);
+            return VerdictPaths.Compose(OriginLegacy, rule);
+        }
+
+        private static bool Starts(string text, string prefix)
+            => text.StartsWith(prefix, StringComparison.Ordinal);
+    }
+}

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

@@ -1,9 +1,11 @@
 using System.Diagnostics;
 using Microsoft.ApplicationInsights;
+using Microsoft.ApplicationInsights.Channel;
 using Microsoft.ApplicationInsights.DataContracts;
 using Microsoft.Azure.Functions.Worker;
 using Microsoft.Azure.Functions.Worker.Middleware;
 using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Functions.Security;
 
 namespace AutopilotMonitor.Functions.Middleware;
 
@@ -113,6 +115,14 @@ public class RequestTelemetryMiddleware : IFunctionsWorkerMiddleware
             if (context.Items.TryGetValue("CorrelationId", out var corrId) && corrId is string correlationId)
                 requestTelemetry.Properties["CorrelationId"] = correlationId;
 
+            // CERT-TENANT-BINDING-SHADOW — set by SecurityValidator.ObserveCertTenantBinding.
+            // Carried on the request row rather than a trace line: worker-side LogInformation never
+            // reaches App Insights (provider default rule is Warning+), so the bulk "Match" outcome
+            // would be invisible and the shadow telemetry would have no denominator. This row already
+            // exists per request and is unsampled, so the outcome costs no additional telemetry.
+            if (context.Items.TryGetValue(CertTenantBinding.RequestItemKey, out var binding) && binding is string bindingOutcome)
+                requestTelemetry.Properties["CertTenantBinding"] = bindingOutcome;
+
             var reqCtx = context.GetRequestContext();
             var tenantId = reqCtx.TenantId;
             if (string.IsNullOrEmpty(tenantId))
@@ -140,6 +150,15 @@ public class RequestTelemetryMiddleware : IFunctionsWorkerMiddleware
             if (caughtException != null)
                 requestTelemetry.Properties["ExceptionType"] = caughtException.GetType().Name;
 
+            // Bypass the worker's adaptive sampling for this item. The SDK sampling processor
+            // passes through any item whose SamplingPercentage is already set, so this is the
+            // per-item opt-out that host.json "excludedTypes" (host process only) cannot provide.
+            // Live-verified 2026-08-23 BEFORE this change: worker request rows carried
+            // ItemCount 2-5 on weekdays (~30% of individual requests missing), so count()-based
+            // queries undercounted. This copy is the single canonical request record — the host
+            // duplicate is dropped by a workspace DCR transformation on AppRequests.
+            ((ISupportSampling)requestTelemetry).SamplingPercentage = 100;
+
             try
             {
                 _telemetryClient.TrackRequest(requestTelemetry);

src/Backend/AutopilotMonitor.Functions/Program.cs

@@ -1,4 +1,4 @@
-using System.IO.Compression;
+using System.IO.Compression;
 using System.Text.Json;
 using System.Text.Json.Serialization;
 using AutopilotMonitor.Functions.DataAccess;
@@ -71,6 +71,10 @@ builder.Services
 // unchanged host duplicate). The host request item is emitted by the HOST process and cannot be
 // dropped from worker code or by host.json "Host.Results" level. Re-attempt only with a mechanism
 // verified against the isolated-worker telemetry pipeline — do not re-add the manual chain.
+// 2026-08-23: solved differently — RequestTelemetryMiddleware sets SamplingPercentage=100 on its
+// own item (per-item sampling bypass), and the host duplicate is removed by a workspace DCR
+// transformation on AppRequests (keeps rows with Properties.Source=='WorkerMiddleware' plus
+// non-HTTP host rows with empty Url, i.e. timer/queue invocations). See docs/backend/telemetry-ingest-shaping.md.
 
 // Drop successful Azure Storage dependencies (Table/Queue/Blob) from the worker telemetry
 // pipeline to curb AppDependencies ingestion cost — this backend is storage-I/O heavy and those
@@ -226,6 +230,7 @@ builder.Services.AddHostedService<AutopilotMonitor.Functions.Services.Offboardin
 // no AddHostedService call needed. DI for its dependencies comes from the singletons above.
 
 builder.Services.AddHostedService<TableInitializerService>(); // Initialize all tables at startup
+builder.Services.AddHostedService<StartupTelemetryService>();  // BackendStartupMs / BackendTableInitMs metrics on ApplicationStarted
 
 // Data Access Layer — repository interfaces backed by Table Storage.
 // To switch to Cosmos DB: replace AddTableStorageDataAccess() with AddCosmosDataAccess().
@@ -419,6 +424,7 @@ builder.Services.AddSingleton<AutopilotMonitor.Functions.Security.AutopilotDevic
 builder.Services.AddSingleton<AutopilotMonitor.Functions.Security.CorporateIdentifierValidator>();
 builder.Services.AddSingleton<AutopilotMonitor.Functions.Security.DeviceAssociationValidator>();
 builder.Services.AddSingleton<AutopilotMonitor.Functions.Security.CloudPcDeviceValidator>();
+builder.Services.AddSingleton<AutopilotMonitor.Functions.Security.IntuneDeviceBindingValidator>();
 
 // Graph add-on permission detection + script display-name resolution.
 // Detector parses the `roles` claim of the SP access token; Resolver fetches

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

@@ -0,0 +1,93 @@
+using System;
+
+namespace AutopilotMonitor.Functions.Security
+{
+    /// <summary>
+    /// Compares the Entra TenantId stamped into the agent's mTLS client certificate against the
+    /// tenant the request claims to belong to.
+    /// <para>
+    /// Without this, a validated certificate only proves "issued by the Intune MDM Device CA to
+    /// some tenant" — the chain is pinned to Microsoft's roots, which every Intune tenant on the
+    /// planet shares. The tenant scoping comes solely from the downstream device validators, so an
+    /// attacker with their own Intune tenant and a known serial number of the victim tenant would
+    /// pass certificate validation. Binding the cert's tenant to the requested tenant closes that.
+    /// </para>
+    /// </summary>
+    /// <remarks>
+    /// SHADOW MODE (stage 1): the outcome is observed and logged only — it never changes an
+    /// authorization decision. Enforcement is stage 2, once telemetry shows how many field
+    /// certificates actually carry the extension. Grep marker for the stage-2 change:
+    /// <c>CERT-TENANT-BINDING-SHADOW</c>.
+    /// </remarks>
+    public static class CertTenantBinding
+    {
+        /// <summary>
+        /// <see cref="Microsoft.Azure.Functions.Worker.FunctionContext"/> item key under which
+        /// <c>SecurityValidator</c> hands the outcome to <c>RequestTelemetryMiddleware</c>, which
+        /// stamps it onto the request row as the <c>CertTenantBinding</c> dimension.
+        /// </summary>
+        /// <remarks>
+        /// The request row is the carrier because worker-side <c>LogInformation</c> never reaches
+        /// App Insights (the provider's default rule is Warning+), so the bulk "Match" outcome
+        /// cannot be observed as a trace and the shadow telemetry would have no denominator.
+        /// </remarks>
+        public const string RequestItemKey = "CertTenantBinding";
+
+        /// <summary>
+        /// Stable outcome codes emitted in the <c>AgentCertTenantBinding</c> structured log.
+        /// Keep these strings stable — they are queried by exact match in KQL
+        /// (<c>customDimensions.Outcome == "Mismatch"</c>).
+        /// </summary>
+        public static class Outcome
+        {
+            /// <summary>Certificate tenant equals the requested tenant — the expected case.</summary>
+            public const string Match = "Match";
+
+            /// <summary>Certificate belongs to a different tenant than the request claims.</summary>
+            public const string Mismatch = "Mismatch";
+
+            /// <summary>Certificate carries no tenant extension (older enrollment / unexpected CA).</summary>
+            public const string ExtensionMissing = "ExtensionMissing";
+
+            /// <summary>Tenant extension present but undecodable.</summary>
+            public const string Unparseable = "Unparseable";
+
+            /// <summary>Requested tenant id was not a GUID, so no meaningful comparison is possible.</summary>
+            public const string RequestTenantNotAGuid = "RequestTenantNotAGuid";
+        }
+
+        /// <summary>
+        /// Evaluates the binding. Pure — no logging, no I/O, no side effects.
+        /// </summary>
+        /// <param name="certTenantId">Tenant GUID decoded from the client certificate, if any.</param>
+        /// <param name="status">Whether the certificate carried a decodable tenant extension.</param>
+        /// <param name="requestedTenantId">The tenant id the request is scoped to.</param>
+        /// <returns>One of the <see cref="Outcome"/> codes.</returns>
+        public static string Evaluate(Guid? certTenantId, CertTenantIdStatus status, string? requestedTenantId)
+        {
+            switch (status)
+            {
+                case CertTenantIdStatus.ExtensionMissing:
+                    return Outcome.ExtensionMissing;
+                case CertTenantIdStatus.Unparseable:
+                    return Outcome.Unparseable;
+            }
+
+            if (certTenantId == null)
+                return Outcome.ExtensionMissing;
+
+            if (!Guid.TryParse(requestedTenantId, out var requested))
+                return Outcome.RequestTenantNotAGuid;
+
+            return certTenantId.Value == requested ? Outcome.Match : Outcome.Mismatch;
+        }
+
+        /// <summary>
+        /// Whether an outcome would block the request once stage-2 enforcement is switched on.
+        /// Nothing consumes this for authorization yet — it exists so the shadow telemetry can be
+        /// read as "would this have been rejected?" without re-deriving the rule at query time.
+        /// </summary>
+        public static bool WouldRejectUnderEnforcement(string outcome) =>
+            outcome == Outcome.Mismatch;
+    }
+}

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

@@ -6,6 +6,7 @@ using System.Linq;
 using System.Reflection;
 using System.Security.Cryptography;
 using System.Security.Cryptography.X509Certificates;
+using AutopilotMonitor.Shared.Security;
 using Microsoft.Extensions.Logging;
 
 namespace AutopilotMonitor.Functions.Security
@@ -253,6 +254,12 @@ namespace AutopilotMonitor.Functions.Security
                     };
                 }
 
+                // 4. Extract the Entra TenantId the CA stamped into the cert (OID 1.2.840.113556.5.14).
+                // Pure function of the cert, so it is cached alongside the validation result.
+                // Read only - the tenant binding it enables is evaluated (and currently only observed)
+                // by SecurityValidator; see CertTenantBinding.
+                var (certTenantId, certTenantIdStatus) = ExtractEntraTenantId(certificate);
+
                 // All checks passed - create and cache success result
                 logger?.LogInformation($"Certificate validated successfully: Thumbprint={thumbprint}");
                 var successResult = new CertificateValidationResult
@@ -260,7 +267,9 @@ namespace AutopilotMonitor.Functions.Security
                     IsValid = true,
                     Thumbprint = thumbprint,
                     Subject = certificate.Subject,
-                    Issuer = certificate.Issuer
+                    Issuer = certificate.Issuer,
+                    CertTenantId = certTenantId,
+                    CertTenantIdStatus = certTenantIdStatus
                 };
 
                 // Cache successful validation
@@ -288,6 +297,30 @@ namespace AutopilotMonitor.Functions.Security
             }
         }
 
+        /// <summary>
+        /// Reads the Entra TenantId the Intune MDM Device CA stamps into the client certificate
+        /// (OID <c>1.2.840.113556.5.14</c>). Never throws — a cert without the extension, or with an
+        /// undecodable one, yields the corresponding status instead of an error, because this value
+        /// is currently observed only and must never turn a valid cert into a rejected one.
+        /// </summary>
+        /// <remarks>
+        /// Deliberately NOT OID <c>1.2.840.113556.5.6</c>: that one carries the Intune <b>Account</b>
+        /// ID, which is a different GUID from the Entra tenant ID.
+        /// </remarks>
+        private static (Guid? TenantId, CertTenantIdStatus Status) ExtractEntraTenantId(X509Certificate2 certificate)
+        {
+            var ext = certificate.Extensions
+                .OfType<X509Extension>()
+                .FirstOrDefault(e => e.Oid?.Value == MsDeviceCertificateOids.IntuneCertEntraTenantIdOid);
+
+            if (ext?.RawData == null)
+                return (null, CertTenantIdStatus.ExtensionMissing);
+
+            return MsDeviceCertificateOids.TryParseGuid(ext.RawData, out var tenantId)
+                ? (tenantId, CertTenantIdStatus.Present)
+                : (null, CertTenantIdStatus.Unparseable);
+        }
+
         /// <summary>
         /// Loads all certs whose embedded resource filename starts with the given prefix.
         /// Multiple PEM blocks per file are supported (cross-signed bundles). Duplicates
@@ -378,5 +411,36 @@ namespace AutopilotMonitor.Functions.Security
         /// Error message if validation failed
         /// </summary>
         public string? ErrorMessage { get; set; }
+
+        /// <summary>
+        /// Entra TenantId stamped into the certificate by the Intune MDM Device CA
+        /// (OID <c>1.2.840.113556.5.14</c>), or <c>null</c> when
+        /// <see cref="CertTenantIdStatus"/> is not <see cref="Security.CertTenantIdStatus.Present"/>.
+        /// </summary>
+        public Guid? CertTenantId { get; set; }
+
+        /// <summary>
+        /// Whether the certificate carried a decodable Entra TenantId extension.
+        /// </summary>
+        public CertTenantIdStatus CertTenantIdStatus { get; set; }
+
+    }
+
+    /// <summary>
+    /// Whether a validated client certificate carried a decodable Entra TenantId extension.
+    /// </summary>
+    public enum CertTenantIdStatus
+    {
+        /// <summary>Certificate was not validated far enough to look, or validation failed.</summary>
+        NotEvaluated = 0,
+
+        /// <summary>Extension present and decoded into a GUID.</summary>
+        Present = 1,
+
+        /// <summary>Certificate carries no such extension (expected on older enrollments).</summary>
+        ExtensionMissing = 2,
+
+        /// <summary>Extension present but its value could not be decoded as a GUID.</summary>
+        Unparseable = 3
     }
 }

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

@@ -1,4 +1,4 @@
-using System;
+using System;
 
 namespace AutopilotMonitor.Functions.Security
 {
@@ -55,5 +55,32 @@ namespace AutopilotMonitor.Functions.Security
         public string? ManagedDeviceId { get; set; }
 
         public string? ErrorMessage { get; set; }
+
+        /// <summary>
+        /// True when this result came from the in-memory cache rather than a fresh Graph lookup.
+        /// The shadow log uses it to emit one line per real lookup instead of one per request:
+        /// a single enrollment produced 263 identical lines off a handful of Graph calls.
+        /// </summary>
+        public bool ServedFromCache { get; set; }
+
+        /// <summary>
+        /// Copy of this result marked as a cache hit, so the shared cached instance is never
+        /// mutated by a caller.
+        /// </summary>
+        internal DeviceAssociationResult AsCacheHit() => new()
+        {
+            IsValid = IsValid,
+            IsTransient = IsTransient,
+            SerialNumber = SerialNumber,
+            AssociationState = AssociationState,
+            DevicePreparationPolicyId = DevicePreparationPolicyId,
+            PreAssociatedByUserPrincipalName = PreAssociatedByUserPrincipalName,
+            AssignedToUserPrincipalName = AssignedToUserPrincipalName,
+            PreAssociationDateTime = PreAssociationDateTime,
+            AssociationDateTime = AssociationDateTime,
+            ManagedDeviceId = ManagedDeviceId,
+            ErrorMessage = ErrorMessage,
+            ServedFromCache = true
+        };
     }
 }

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

@@ -1,4 +1,4 @@
-using AutopilotMonitor.Shared;
+using AutopilotMonitor.Shared;
 using System;
 using System.Linq;
 using System.Net.Http;
@@ -63,7 +63,7 @@ namespace AutopilotMonitor.Functions.Security
 
             if (_cache.TryGetValue(cacheKey, out DeviceAssociationResult? cached) && cached != null)
             {
-                return cached;
+                return cached.AsCacheHit();
             }
 
             const int maxAttempts = 2;

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

@@ -464,6 +464,7 @@ public static class EndpointAccessPolicyCatalog
         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/metrics/verdict-calibration", 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),

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

@@ -0,0 +1,352 @@
+using AutopilotMonitor.Shared;
+using System;
+using System.Net.Http.Headers;
+using System.Threading.Tasks;
+using Microsoft.Extensions.Caching.Memory;
+using Microsoft.Extensions.Logging;
+using Newtonsoft.Json;
+using Newtonsoft.Json.Linq;
+
+namespace AutopilotMonitor.Functions.Security
+{
+    /// <summary>
+    /// Binds the agent's client certificate to a device object that actually exists in the tenant
+    /// the request claims: the certificate's Subject CN carries the Intune managedDevice id, and
+    /// this validator looks exactly that id up in the tenant's own
+    /// <c>deviceManagement/managedDevices</c> inventory.
+    /// <para>
+    /// <see cref="CertTenantBinding"/> already proves the certificate was issued to this tenant.
+    /// This adds the second half: that the specific device is still one of the tenant's enrolled
+    /// devices, so a certificate lifted from a decommissioned or wiped machine no longer resolves.
+    /// </para>
+    /// </summary>
+    /// <remarks>
+    /// Built as a near-sibling of <see cref="CloudPcDeviceValidator"/> — same DI dependencies and
+    /// the same cache/retry/transient contract (30 min positive, 5 min negative, 2 attempts with a
+    /// 2s back-off, transient on Graph 5xx / token failure / exception). Two deliberate differences:
+    /// <list type="bullet">
+    /// <item><description>A point-GET on <c>managedDevices/{id}</c> rather than an OData
+    /// <c>$filter</c>. There is no filter literal to inject into, and a 404 is an unambiguous
+    /// "not this tenant's device".</description></item>
+    /// <item><description><see cref="IntuneDeviceBindingResult.EnrolledDateTime"/> is carried out
+    /// so callers can record how old the device object was at request time. That is the open
+    /// question this check must answer before it could ever gate enrollment: a device whose object
+    /// appears only after the agent's first call is a race, not an attack.</description></item>
+    /// </list>
+    /// SHADOW ONLY (stage 3): nothing consumes the result for an authorization decision. Grep
+    /// marker for the enforcement change: <c>CERT-DEVICE-BINDING-SHADOW</c>.
+    /// </remarks>
+    public class IntuneDeviceBindingValidator
+    {
+        private static readonly TimeSpan PositiveCacheTtl = TimeSpan.FromMinutes(30);
+        private static readonly TimeSpan NegativeCacheTtl = TimeSpan.FromMinutes(5);
+
+        private readonly ILogger<IntuneDeviceBindingValidator> _logger;
+        private readonly IHttpClientFactory _httpClientFactory;
+        private readonly IMemoryCache _cache;
+        private readonly GraphTokenService _graphTokenService;
+
+        public IntuneDeviceBindingValidator(
+            ILogger<IntuneDeviceBindingValidator> logger,
+            IHttpClientFactory httpClientFactory,
+            IMemoryCache cache,
+            GraphTokenService graphTokenService)
+        {
+            _logger = logger;
+            _httpClientFactory = httpClientFactory;
+            _cache = cache;
+            _graphTokenService = graphTokenService;
+        }
+
+        /// <summary>
+        /// Resolves the Intune device id (from the client certificate's Subject CN) against the
+        /// tenant's managedDevices inventory. Never throws.
+        /// </summary>
+        public async Task<IntuneDeviceBindingResult> ValidateAsync(
+            string tenantId,
+            string? intuneDeviceId,
+            string? sessionId = null)
+        {
+            // The id becomes a URL path segment. The strict GUID gate keeps it from being anything
+            // else - the same rule the rest of the security path applies.
+            if (!SecurityValidator.IsValidGuid(intuneDeviceId))
+            {
+                return new IntuneDeviceBindingResult
+                {
+                    Outcome = IntuneDeviceBindingOutcome.NoDeviceIdInCert,
+                    ErrorMessage = "Certificate subject does not carry a valid Intune device id"
+                };
+            }
+
+            var normalizedId = intuneDeviceId!.Trim().ToLowerInvariant();
+            var cacheKey = BuildCacheKey(tenantId, normalizedId);
+
+            if (_cache.TryGetValue(cacheKey, out IntuneDeviceBindingResult? cached) && cached != null)
+                return cached.AsCacheHit();
+
+            const int maxAttempts = 2;
+            IntuneDeviceBindingResult? lastTransient = null;
+
+            for (int attempt = 1; attempt <= maxAttempts; attempt++)
+            {
+                var result = await TryValidateViaGraphAsync(tenantId, normalizedId, sessionId, cacheKey, attempt);
+
+                if (!result.IsTransient)
+                    return result;
+
+                lastTransient = result;
+                if (attempt < maxAttempts)
+                {
+                    _logger.LogWarning(
+                        "Intune device binding transient failure for tenant {TenantId}, deviceId {IntuneDeviceId} (attempt {Attempt}/{MaxAttempts}). Retrying...",
+                        tenantId, normalizedId, attempt, maxAttempts);
+                    await Task.Delay(TimeSpan.FromSeconds(2));
+                }
+            }
+
+            return lastTransient!;
+        }
+
+        private async Task<IntuneDeviceBindingResult> TryValidateViaGraphAsync(
+            string tenantId, string normalizedId, string? sessionId, string cacheKey, int attempt)
+        {
+            try
+            {
+                var tokenResult = await _graphTokenService.GetAccessTokenAsync(tenantId);
+                if (string.IsNullOrEmpty(tokenResult.AccessToken))
+                {
+                    return new IntuneDeviceBindingResult
+                    {
+                        Outcome = IntuneDeviceBindingOutcome.Transient,
+                        IntuneDeviceId = normalizedId,
+                        ErrorMessage = "Graph access token could not be acquired"
+                    };
+                }
+
+                var graphClient = _httpClientFactory.CreateClient();
+                graphClient.DefaultRequestHeaders.Authorization = new AuthenticationHeaderValue("Bearer", tokenResult.AccessToken);
+
+                // Point-GET: normalizedId passed the GUID gate, so the path segment is safe.
+                var graphUrl = $"{Constants.GraphBaseUrl}/v1.0/deviceManagement/managedDevices/{normalizedId}"
+                               + "?$select=id,deviceName,enrolledDateTime,azureADDeviceId,managementState";
+
+                var response = await graphClient.GetAsync(graphUrl);
+                var responseBody = await response.Content.ReadAsStringAsync();
+
+                if (response.StatusCode == System.Net.HttpStatusCode.NotFound)
+                {
+                    // Definitive: this tenant has no managedDevice with that id. Either the device
+                    // belongs elsewhere, or its object has not appeared yet - the race this shadow
+                    // pass exists to measure.
+                    return CacheAndReturn(cacheKey, new IntuneDeviceBindingResult
+                    {
+                        Outcome = IntuneDeviceBindingOutcome.NotFound,
+                        IntuneDeviceId = normalizedId,
+                        ErrorMessage = $"Device '{normalizedId}' is not an enrolled Intune device in this tenant"
+                    }, isPositive: false);
+                }
+
+                if (response.StatusCode == System.Net.HttpStatusCode.Forbidden)
+                {
+                    // Configuration state, not an outage: retrying cannot fix a missing grant.
+                    // Cached briefly so a fresh grant is picked up quickly.
+                    return CacheAndReturn(cacheKey, new IntuneDeviceBindingResult
+                    {
+                        Outcome = IntuneDeviceBindingOutcome.PermissionMissing,
+                        IntuneDeviceId = normalizedId,
+                        ErrorMessage = "Managed device lookup not permitted - grant the optional 'IntuneDeviceBinding' Graph feature (DeviceManagementManagedDevices.Read.All)"
+                    }, isPositive: false);
+                }
+
+                if (!response.IsSuccessStatusCode)
+                {
+                    _logger.LogWarning(
+                        "Intune device binding Graph query failed for tenant {TenantId} (attempt {Attempt}). Status: {StatusCode}. Body: {ResponseBody}",
+                        tenantId, attempt, (int)response.StatusCode, responseBody);
+
+                    return new IntuneDeviceBindingResult
+                    {
+                        Outcome = IntuneDeviceBindingOutcome.Transient,
+                        IntuneDeviceId = normalizedId,
+                        ErrorMessage = $"Graph query failed with status {(int)response.StatusCode}"
+                    };
+                }
+
+                var result = ParseManagedDeviceResponse(responseBody, normalizedId);
+                return CacheAndReturn(cacheKey, result, isPositive: result.IsValid);
+            }
+            catch (Exception ex)
+            {
+                _logger.LogError(
+                    ex,
+                    "Error during Intune device binding for tenant {TenantId}, session {SessionId}, deviceId {IntuneDeviceId} (attempt {Attempt})",
+                    tenantId, sessionId ?? "<none>", normalizedId, attempt);
+
+                return new IntuneDeviceBindingResult
+                {
+                    Outcome = IntuneDeviceBindingOutcome.Transient,
+                    IntuneDeviceId = normalizedId,
+                    ErrorMessage = $"Error during Intune device binding: {ex.Message}"
+                };
+            }
+        }
+
+        /// <summary>
+        /// Pure-function: maps a <c>managedDevices/{id}</c> Graph response body to a result.
+        /// Re-checks the returned <c>id</c> so a redirected or widened lookup can never produce a
+        /// false positive.
+        /// </summary>
+        internal static IntuneDeviceBindingResult ParseManagedDeviceResponse(string responseBody, string normalizedId)
+        {
+            var notFound = new IntuneDeviceBindingResult
+            {
+                Outcome = IntuneDeviceBindingOutcome.NotFound,
+                IntuneDeviceId = normalizedId,
+                ErrorMessage = $"Device '{normalizedId}' is not an enrolled Intune device in this tenant"
+            };
+
+            JObject? device;
+            try
+            {
+                device = JsonConvert.DeserializeObject<JObject>(responseBody);
+            }
+            catch (JsonException)
+            {
+                return notFound;
+            }
+
+            var returnedId = device?["id"]?.ToString()?.Trim();
+            if (string.IsNullOrEmpty(returnedId) ||
+                !string.Equals(returnedId, normalizedId, StringComparison.OrdinalIgnoreCase))
+            {
+                return notFound;
+            }
+
+            return new IntuneDeviceBindingResult
+            {
+                Outcome = IntuneDeviceBindingOutcome.Match,
+                IntuneDeviceId = normalizedId,
+                DeviceName = device?["deviceName"]?.ToString(),
+                AzureAdDeviceId = device?["azureADDeviceId"]?.ToString(),
+                ManagementState = device?["managementState"]?.ToString(),
+                EnrolledDateTime = ParseEnrolledDateTime(device?["enrolledDateTime"]?.ToString())
+            };
+        }
+
+        /// <summary>
+        /// Parses Graph's <c>enrolledDateTime</c>. Intune returns <c>0001-01-01T00:00:00Z</c> for
+        /// devices it has no enrollment timestamp for; that is absence, not a date, so it maps to
+        /// <c>null</c> rather than a bogus age of two thousand years.
+        /// </summary>
+        internal static DateTimeOffset? ParseEnrolledDateTime(string? raw)
+        {
+            if (string.IsNullOrWhiteSpace(raw)) return null;
+
+            if (!DateTimeOffset.TryParse(
+                    raw,
+                    System.Globalization.CultureInfo.InvariantCulture,
+                    System.Globalization.DateTimeStyles.AdjustToUniversal
+                        | System.Globalization.DateTimeStyles.AssumeUniversal,
+                    out var parsed))
+            {
+                return null;
+            }
+
+            return parsed.Year <= 1 ? null : parsed;
+        }
+
+        internal static string BuildCacheKey(string tenantId, string intuneDeviceId)
+            => $"intune-device-binding:{tenantId}:{intuneDeviceId}";
+
+        private IntuneDeviceBindingResult CacheAndReturn(
+            string cacheKey, IntuneDeviceBindingResult result, bool isPositive)
+        {
+            _cache.Set(cacheKey, result, new MemoryCacheEntryOptions
+            {
+                AbsoluteExpirationRelativeToNow = isPositive ? PositiveCacheTtl : NegativeCacheTtl
+            });
+            return result;
+        }
+    }
+
+    /// <summary>
+    /// Stable outcome codes for the device-binding check. Emitted verbatim in the
+    /// <c>AgentCertDeviceBinding</c> log and as the <c>CertDeviceBinding</c> request dimension —
+    /// operators match on them in KQL, so keep the names stable.
+    /// </summary>
+    public enum IntuneDeviceBindingOutcome
+    {
+        /// <summary>Device id from the certificate resolves to a managedDevice in this tenant.</summary>
+        Match = 0,
+
+        /// <summary>No managedDevice with that id in this tenant (foreign device, or not yet created).</summary>
+        NotFound = 1,
+
+        /// <summary>Certificate subject carried no usable Intune device id.</summary>
+        NoDeviceIdInCert = 2,
+
+        /// <summary>DeviceManagementManagedDevices.Read.All not granted in this tenant.</summary>
+        PermissionMissing = 3,
+
+        /// <summary>Graph outage, token failure or exception — says nothing about the device.</summary>
+        Transient = 4
+    }
+
+    public class IntuneDeviceBindingResult
+    {
+        public IntuneDeviceBindingOutcome Outcome { get; set; }
+
+        /// <summary>True only for <see cref="IntuneDeviceBindingOutcome.Match"/>.</summary>
+        public bool IsValid => Outcome == IntuneDeviceBindingOutcome.Match;
+
+        /// <summary>
+        /// True when the failure says nothing about the device (Graph error, token issue).
+        /// Transient results are never cached.
+        /// </summary>
+        public bool IsTransient => Outcome == IntuneDeviceBindingOutcome.Transient;
+
+        /// <summary>Intune device id taken from the client certificate's Subject CN.</summary>
+        public string? IntuneDeviceId { get; set; }
+
+        /// <summary>Intune device name — diagnostic only, never a gate.</summary>
+        public string? DeviceName { get; set; }
+
+        /// <summary>Entra device id of the same device, if Intune knows one — diagnostic only.</summary>
+        public string? AzureAdDeviceId { get; set; }
+
+        /// <summary>Intune management state — diagnostic only.</summary>
+        public string? ManagementState { get; set; }
+
+        /// <summary>
+        /// When Intune recorded the enrollment. The age of this value at request time is what
+        /// tells a genuine race (object created seconds ago) apart from a foreign certificate.
+        /// </summary>
+        public DateTimeOffset? EnrolledDateTime { get; set; }
+
+        public string? ErrorMessage { get; set; }
+
+        /// <summary>
+        /// True when this result came from the in-memory cache rather than a fresh Graph lookup.
+        /// Callers use it to log once per real lookup instead of once per request: a single
+        /// enrollment produced 136 requests but only one Graph call, so logging per request said
+        /// the same thing 136 times.
+        /// </summary>
+        public bool ServedFromCache { get; set; }
+
+        /// <summary>
+        /// Copy of this result marked as a cache hit, so the shared cached instance is never
+        /// mutated by a caller.
+        /// </summary>
+        internal IntuneDeviceBindingResult AsCacheHit() => new()
+        {
+            Outcome = Outcome,
+            IntuneDeviceId = IntuneDeviceId,
+            DeviceName = DeviceName,
+            AzureAdDeviceId = AzureAdDeviceId,
+            ManagementState = ManagementState,
+            EnrolledDateTime = EnrolledDateTime,
+            ErrorMessage = ErrorMessage,
+            ServedFromCache = true
+        };
+    }
+}

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

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Net;
 using System.Threading.Tasks;
 using AutopilotMonitor.Functions.Services;
@@ -34,9 +34,10 @@ namespace AutopilotMonitor.Functions.Security
             string? sessionId = null,
             BootstrapSessionService? bootstrapSessionService = null,
             DeviceAssociationValidator? deviceAssociationValidator = null,
-            CloudPcDeviceValidator? cloudPcDeviceValidator = null)
+            CloudPcDeviceValidator? cloudPcDeviceValidator = null,
+            IntuneDeviceBindingValidator? intuneDeviceBindingValidator = null)
         {
-            var validator = new SecurityValidator(configService, adminConfigService, rateLimitService, autopilotDeviceValidator, corporateIdentifierValidator, logger, bootstrapSessionService, deviceAssociationValidator, cloudPcDeviceValidator);
+            var validator = new SecurityValidator(configService, adminConfigService, rateLimitService, autopilotDeviceValidator, corporateIdentifierValidator, logger, bootstrapSessionService, deviceAssociationValidator, cloudPcDeviceValidator, intuneDeviceBindingValidator);
             var validation = await validator.ValidateRequestAsync(req, tenantId, sessionId);
 
             if (!validation.IsValid)

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

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Linq;
 using System.Net;
 using System.Text.RegularExpressions;
@@ -48,6 +48,7 @@ namespace AutopilotMonitor.Functions.Security
         private readonly CorporateIdentifierValidator _corporateIdentifierValidator;
         private readonly DeviceAssociationValidator? _deviceAssociationValidator;
         private readonly CloudPcDeviceValidator? _cloudPcDeviceValidator;
+        private readonly IntuneDeviceBindingValidator? _intuneDeviceBindingValidator;
         private readonly BootstrapSessionService? _bootstrapSessionService;
         private readonly ILogger _logger;
 
@@ -60,7 +61,8 @@ namespace AutopilotMonitor.Functions.Security
             ILogger logger,
             BootstrapSessionService? bootstrapSessionService = null,
             DeviceAssociationValidator? deviceAssociationValidator = null,
-            CloudPcDeviceValidator? cloudPcDeviceValidator = null)
+            CloudPcDeviceValidator? cloudPcDeviceValidator = null,
+            IntuneDeviceBindingValidator? intuneDeviceBindingValidator = null)
         {
             _configService = configService;
             _adminConfigService = adminConfigService;
@@ -69,6 +71,7 @@ namespace AutopilotMonitor.Functions.Security
             _corporateIdentifierValidator = corporateIdentifierValidator;
             _deviceAssociationValidator = deviceAssociationValidator;
             _cloudPcDeviceValidator = cloudPcDeviceValidator;
+            _intuneDeviceBindingValidator = intuneDeviceBindingValidator;
             _bootstrapSessionService = bootstrapSessionService;
             _logger = logger;
         }
@@ -275,6 +278,12 @@ namespace AutopilotMonitor.Functions.Security
                 };
             }
 
+            // 1b. CERT-TENANT-BINDING-SHADOW — observe only, never decide.
+            // The cert chain is pinned to Microsoft's Intune roots, which every Intune tenant shares,
+            // so a valid cert alone does not prove the caller belongs to THIS tenant. Stage 1 measures
+            // how reliably the certificate's own tenant stamp matches before stage 2 enforces it.
+            ObserveCertTenantBinding(certValidation, tenantId, req, sessionId);
+
             // 2. Check rate limit (DoS protection)
             // Effective limit = per-tenant override if set, otherwise the global default.
             var rateLimitValue = await ResolveDeviceRateLimitAsync(config);
@@ -444,6 +453,14 @@ namespace AutopilotMonitor.Functions.Security
                     try
                     {
                         var devPrepResult = await validator.LookupAsync(capturedTenant, capturedSerial, capturedSession);
+
+                        // One line per real Graph lookup, not per request — the same gate the
+                        // cert-device-binding shadow uses. Measured 2026-08-24: a single enrollment
+                        // produced 263 identical lines because the cache answered nearly every
+                        // request. Uncached outcomes (transient failures) still log every time.
+                        if (devPrepResult.ServedFromCache)
+                            return;
+
                         logger.LogWarning(
                             "DevPrep association lookup (shadow) for tenant {TenantId}, session {SessionId}, serial {SerialNumber}: matched={Matched}, transient={Transient}, state={State}, policy={PolicyId}",
                             capturedTenant, capturedSession ?? "<none>", capturedSerial,
@@ -457,6 +474,69 @@ namespace AutopilotMonitor.Functions.Security
                 });
             }
 
+            // 6. CERT-DEVICE-BINDING-SHADOW - cert-to-device binding (Global-Admin-only preview).
+            // Stage 1 (CertTenantBinding) proves the certificate was issued to THIS tenant; this
+            // proves the specific device is one the tenant actually enrolled, so a certificate
+            // lifted from a decommissioned machine stops resolving. Observation only: it never
+            // touches deviceValidated or the returned result.
+            //
+            // Fire-and-forget for the same reason as the DevPrep lookup above - a cold Graph
+            // round-trip can take seconds and must never sit in the agent's request path. That
+            // rules out the request-row dimension stage 1 uses (the row is written the moment the
+            // function returns, before this task finishes), so the outcome is logged at Warning
+            // like the DevPrep shadow. Volume is bounded: only Global Admins can enable the
+            // toggle. Widening it beyond preview should move to an inline call plus the
+            // CertTenantBinding-style request dimension.
+            if (config.ValidateIntuneDeviceBinding && _intuneDeviceBindingValidator != null)
+            {
+                var bindingValidator = _intuneDeviceBindingValidator;
+                var bindingTenant = tenantId;
+                var bindingSession = sessionId;
+                var bindingSubject = certValidation.Subject;
+                var bindingThumbprint = certValidation.Thumbprint;
+                var bindingLogger = _logger;
+
+                _ = Task.Run(async () =>
+                {
+                    try
+                    {
+                        TryGetIntuneDeviceIdFromCertSubject(bindingSubject, out var certDeviceId);
+                        var binding = await bindingValidator.ValidateAsync(bindingTenant, certDeviceId, bindingSession);
+
+                        // One line per real Graph lookup, not per request. Measured on the first
+                        // live enrollment: 136 requests produced exactly ONE Graph call (the 30 min
+                        // positive cache did its job), so per-request logging repeated the same
+                        // finding 136 times. Cached repeats add nothing - the age just counts up
+                        // mechanically from the same enrolledDateTime. Uncached outcomes
+                        // (Transient, NoDeviceIdInCert) still log every time, which is what we want
+                        // for anomalies.
+                        if (binding.ServedFromCache)
+                            return;
+
+                        // Age of the device object at request time: the discriminator between a
+                        // genuine enrollment race (object created moments ago, or not yet) and a
+                        // certificate that never belonged to this tenant.
+                        var enrolledAgeSeconds = binding.EnrolledDateTime.HasValue
+                            ? (long)Math.Round((DateTimeOffset.UtcNow - binding.EnrolledDateTime.Value).TotalSeconds)
+                            : -1;
+
+                        bindingLogger.LogWarning(
+                            "AgentCertDeviceBinding outcome={Outcome} enforced={Enforced} tenant={TenantId} "
+                            + "certDeviceId={CertDeviceId} device={DeviceName} enrolledAgeSeconds={EnrolledAgeSeconds} "
+                            + "mgmtState={ManagementState} thumbprint={Thumbprint} session={SessionId} detail={Detail}",
+                            binding.Outcome, false, bindingTenant,
+                            certDeviceId ?? "n/a", binding.DeviceName ?? "n/a", enrolledAgeSeconds,
+                            binding.ManagementState ?? "n/a", bindingThumbprint ?? "n/a",
+                            bindingSession ?? "n/a", binding.ErrorMessage ?? "n/a");
+                    }
+                    catch (Exception ex)
+                    {
+                        bindingLogger.LogWarning(ex,
+                            "AgentCertDeviceBinding (shadow) failed for tenant {TenantId} - ignored.", bindingTenant);
+                    }
+                });
+            }
+
             // All checks passed
             return new SecurityValidationResult
             {
@@ -514,6 +594,73 @@ namespace AutopilotMonitor.Functions.Security
             return false;
         }
 
+        /// <summary>
+        /// CERT-TENANT-BINDING-SHADOW — stage 1 observability for the client-certificate tenant
+        /// binding. Evaluates whether the tenant stamped into the agent's certificate matches the
+        /// tenant the request claims, and logs the outcome. It deliberately returns nothing and
+        /// changes no state: enforcement is stage 2, after this telemetry shows the miss rate.
+        /// </summary>
+        /// <remarks>
+        /// Two carriers, because they answer different questions and have different costs:
+        /// <list type="bullet">
+        /// <item><description>Every outcome is stamped onto the request row via
+        /// <see cref="CertTenantBinding.RequestItemKey"/>. That row already exists per request and is
+        /// unsampled, so the denominator ("how many requests matched") costs no extra telemetry.
+        /// A trace line could not do this: worker-side <c>LogInformation</c> never reaches App
+        /// Insights, so the bulk "Match" case would simply be invisible.</description></item>
+        /// <item><description>Outcomes that need an enforcement decision (mismatch, missing or
+        /// undecodable extension) are additionally logged at Warning with the full context —
+        /// thumbprint, cert tenant, session — because each one is individually actionable.</description></item>
+        /// </list>
+        /// </remarks>
+        private void ObserveCertTenantBinding(
+            CertificateValidationResult certValidation,
+            string tenantId,
+            HttpRequestData req,
+            string? sessionId)
+        {
+            try
+            {
+                var outcome = CertTenantBinding.Evaluate(
+                    certValidation.CertTenantId,
+                    certValidation.CertTenantIdStatus,
+                    tenantId);
+
+                // Denominator: rides along on the request row for every outcome, including Match.
+                var items = req.FunctionContext?.Items;
+                if (items != null)
+                    items[CertTenantBinding.RequestItemKey] = outcome;
+
+                // Everything below is the detail record for outcomes someone has to act on.
+                if (outcome == CertTenantBinding.Outcome.Match)
+                    return;
+
+                var agentVersion = req.Headers.Contains("X-Agent-Version")
+                    ? req.Headers.GetValues("X-Agent-Version").FirstOrDefault() ?? "n/a"
+                    : "n/a";
+
+                // Only the cert's tenant is logged, never the raw certificate. On a mismatch this is
+                // the attacker-controlled side of the comparison and the single most useful field.
+                const string template =
+                    "AgentCertTenantBinding outcome={Outcome} enforced={Enforced} wouldReject={WouldReject} "
+                    + "tenant={TenantId} certTenant={CertTenantId} status={Status} thumbprint={Thumbprint} "
+                    + "session={SessionId} ver={AgentVersion}";
+
+                var certTenant = certValidation.CertTenantId?.ToString() ?? "n/a";
+                var wouldReject = CertTenantBinding.WouldRejectUnderEnforcement(outcome);
+
+                _logger.LogWarning(template,
+                    outcome, false, wouldReject, tenantId, certTenant, certValidation.CertTenantIdStatus,
+                    certValidation.Thumbprint ?? "n/a", sessionId ?? "n/a", agentVersion);
+            }
+            catch (Exception ex)
+            {
+                // Shadow-mode observability must never be able to fail a request that would
+                // otherwise have been accepted.
+                _logger.LogWarning(ex, "AgentCertTenantBinding observation failed for tenant {TenantId}", tenantId);
+            }
+        }
+
         /// <summary>
         /// Emits a structured warning with request-side context (path, hardware headers, agent
         /// version) at every fail-fast rejection point. Pairs with the cert-side warnings emitted

src/Backend/AutopilotMonitor.Functions/Services/DistressRateLimitService.cs

@@ -1,5 +1,4 @@
 using System;
-using System.Collections.Concurrent;
 using System.Collections.Generic;
 using System.Linq;
 using Microsoft.Extensions.Caching.Memory;
@@ -21,8 +20,15 @@ namespace AutopilotMonitor.Functions.Services
         private readonly IMemoryCache _cache;
         private readonly ILogger<DistressRateLimitService> _logger;
 
-        // Per-key locks (same proven pattern as RateLimitService)
-        private readonly ConcurrentDictionary<string, object> _locks = new();
+        // History + lock in one cache entry (same pattern as RateLimitService): same list ⇒ same lock.
+        private sealed class RequestBucket
+        {
+            public readonly List<DateTime> History = new();
+            public readonly object Sync = new();
+        }
+
+        // Serializes bucket creation (GetOrCreate is not atomic); hot path is a lock-free TryGetValue.
+        private readonly object _createLock = new();
 
         // Layer 1: Per-IP
         private const int MaxPerIp = 5;
@@ -91,17 +97,12 @@ namespace AutopilotMonitor.Functions.Services
         /// </summary>
         private bool CheckSlidingWindow(string cacheKey, int maxRequests, TimeSpan window, DateTime now)
         {
-            var lockObj = _locks.GetOrAdd(cacheKey, _ => new object());
-            var history = _cache.GetOrCreate(cacheKey, entry =>
-            {
-                entry.SlidingExpiration = window;
-                entry.AbsoluteExpirationRelativeToNow = TimeSpan.FromHours(2);
-                entry.RegisterPostEvictionCallback((key, _, _, _) => _locks.TryRemove((string)key, out _));
-                return new List<DateTime>();
-            })!;
+            var bucket = GetOrCreateBucket(cacheKey, window);
 
-            lock (lockObj)
+            lock (bucket.Sync)
             {
+                var history = bucket.History;
+
                 // Remove entries outside the window
                 var windowStart = now.Subtract(window);
                 history.RemoveAll(ts => ts < windowStart);
@@ -113,16 +114,26 @@ namespace AutopilotMonitor.Functions.Services
 
                 history.Add(now);
 
-                _cache.Set(cacheKey, history, new MemoryCacheEntryOptions
-                {
-                    SlidingExpiration = window,
-                    AbsoluteExpirationRelativeToNow = TimeSpan.FromHours(2)
-                });
-
                 return true;
             }
         }
 
+        private RequestBucket GetOrCreateBucket(string cacheKey, TimeSpan window)
+        {
+            if (_cache.TryGetValue(cacheKey, out RequestBucket? existing) && existing != null)
+                return existing;
+
+            lock (_createLock)
+            {
+                return _cache.GetOrCreate(cacheKey, entry =>
+                {
+                    entry.SlidingExpiration = window;
+                    entry.AbsoluteExpirationRelativeToNow = TimeSpan.FromHours(2);
+                    return new RequestBucket();
+                })!;
+            }
+        }
+
         private bool IsCircuitOpen(DateTime now)
         {
             lock (_globalLock)

src/Backend/AutopilotMonitor.Functions/Services/EnrollmentTimeoutClassifier.cs

@@ -273,6 +273,34 @@ namespace AutopilotMonitor.Functions.Services
                    $"device presumably sealed or powered off awaiting the end user{acct}";
         }
 
+        /// <summary>
+        /// Self-deploying (kiosk / shared-device) profile gate. Such a profile has NO user phase:
+        /// Device ESP reaching all-succeeded IS the enrollment's end, and the agent's own
+        /// SelfDeploying terminal only adds a 5-min confirmation window on top of it. When the
+        /// agent goes silent after that point (reboot into the kiosk autologon, powered off and
+        /// boxed for the school term, WiFi never re-associating for the relaunch — tenant
+        /// aebdce78, 836 sessions 2026-08-17..21), "awaiting user / Account Setup" is a factual
+        /// error and the later Incomplete("Account Setup 0/5") is worse: nobody will ever sign
+        /// in, the devices are finished and in service. Demands the registration-time
+        /// registry-confirmed profile flag (CloudAssignedOobeConfig 0x20|0x40), Device Setup
+        /// all-succeeded and no explicit failure. Shared by the sweep (stage 1 + 2), the
+        /// max-lifetime ingest verdict and the retro-reclassification so one evidence rule applies.
+        /// </summary>
+        public static bool IsSelfDeployingProvisioned(in EspProvisioningRollup rollup, bool isSelfDeployingProfile)
+            => isSelfDeployingProfile
+               && rollup.DeviceSetupAllSucceeded
+               && !rollup.HasExplicitFailure;
+
+        /// <summary>
+        /// Customer-facing Succeeded-reconcile reason for <see cref="IsSelfDeployingProvisioned"/>.
+        /// Carries the silence-transparency clause like every other reconcile reason.
+        /// </summary>
+        public static string SelfDeployingReconcileReason(DateTime startedAtUtc, DateTime nowUtc, DateTime? lastEventAtUtc)
+            => AppendReconcileTiming(
+                "Reconciled: self-deploying profile — Device Setup fully provisioned (this profile has no user / Account Setup phase); " +
+                "agent went silent before confirming completion",
+                startedAtUtc, nowUtc, lastEventAtUtc);
+
         /// <summary>
         /// Decide the honest terminal (or non-terminal) state for a timed-out session. See the
         /// decision table in the design note. Returns the target status and a human-readable
@@ -291,14 +319,16 @@ namespace AutopilotMonitor.Functions.Services
         /// </param>
         /// <param name="isPreProvisioned">Session's IsPreProvisioned flag (WhiteGlove Part-2 gate input).</param>
         /// <param name="resumedAt">Session's ResumedAt — set exclusively by the whiteglove_resumed ingest.</param>
-        public static (SessionStatus Status, string Reason) ClassifyTimedOutSession(
+        /// <param name="isSelfDeployingProfile">Session's IsSelfDeployingProfile flag (rule 1c gate input).</param>
+        public static (SessionStatus Status, string Reason, string Rule) ClassifyTimedOutSession(
             EspProvisioningRollup rollup, DateTime startedAtUtc, DateTime nowUtc, int graceHours,
-            DateTime? lastEventAtUtc = null, bool isPreProvisioned = false, DateTime? resumedAt = null)
+            DateTime? lastEventAtUtc = null, bool isPreProvisioned = false, DateTime? resumedAt = null,
+            bool isSelfDeployingProfile = false)
         {
             // 1. An explicit terminal failure event is a real failure (defensive — such a session
             //    would normally already be terminal via ingest and not reach the sweep).
             if (rollup.HasExplicitFailure)
-                return (SessionStatus.Failed, "Enrollment reported an explicit failure before timeout");
+                return (SessionStatus.Failed, "Enrollment reported an explicit failure before timeout", ClassifierRules.R1ExplicitFailure);
 
             // 1b. WhiteGlove Part 2 with zero user evidence since the resume: the device is parked
             //     between technician and end user (sealed box / powered off at the logon screen).
@@ -311,13 +341,22 @@ namespace AutopilotMonitor.Functions.Services
             if (IsWhiteGloveAwaitingUser(rollup, isPreProvisioned, resumedAt))
             {
                 if ((nowUtc - startedAtUtc).TotalHours < graceHours)
-                    return (SessionStatus.AwaitingUser, WhiteGloveAwaitingUserReason(rollup));
+                    return (SessionStatus.AwaitingUser, WhiteGloveAwaitingUserReason(rollup), ClassifierRules.R1bWhiteGloveAwaiting);
 
                 return (SessionStatus.Succeeded, AppendReconcileTiming(
                     $"Reconciled at timeout: pre-provisioning (WhiteGlove Part 1) completed — no user sign-in observed within {graceHours}h grace",
-                    startedAtUtc, nowUtc, lastEventAtUtc));
+                    startedAtUtc, nowUtc, lastEventAtUtc), ClassifierRules.R1bWhiteGloveSucceeded);
             }
 
+            // 1c. Self-deploying profile with Device Setup fully provisioned: there is no user phase
+            //     to wait for, so neither AwaitingUser nor the grace window apply — the device is
+            //     done (see IsSelfDeployingProvisioned). Runs before rule 2 so the reason names the
+            //     profile instead of claiming "Account Setup completed"; real user evidence cannot
+            //     contradict it (a kiosk autologon is not a user) and an explicit failure already
+            //     returned in rule 1.
+            if (IsSelfDeployingProvisioned(rollup, isSelfDeployingProfile))
+                return (SessionStatus.Succeeded, SelfDeployingReconcileReason(startedAtUtc, nowUtc, lastEventAtUtc), ClassifierRules.R1cSelfDeploying);
+
             // 2. Account Setup fully succeeded (or a terminal completion event) → reconcile to success.
             //    Once Part 2 has resumed, the Part-1 whiteglove_complete in the same stream no longer
             //    proves the session's outcome — only a real enrollment_complete does. Without this,
@@ -329,14 +368,15 @@ namespace AutopilotMonitor.Functions.Services
             if (rollup.AccountSetupAllSucceeded || terminalCompleteProvesOutcome)
                 return (SessionStatus.Succeeded, AppendReconcileTiming(
                     "Reconciled at timeout: Account Setup completed (all subcategories succeeded / enrollment_complete observed)",
-                    startedAtUtc, nowUtc, lastEventAtUtc));
+                    startedAtUtc, nowUtc, lastEventAtUtc), ClassifierRules.R2AccountSetupComplete);
 
             // 3. The agent's absolute-age emergency break fired — the agent has cleaned up and exited, so
             //    nothing more will ever arrive for this session. Terminalize NOW (skip the AwaitingUser grace):
             //    it did not complete (2) and did not explicitly fail (1), so the honest verdict is Incomplete.
             if (rollup.HasAgentEmergencyBreak)
                 return (SessionStatus.Incomplete,
-                    "Agent emergency break fired (absolute session-age cap) — agent gone without completion");
+                    "Agent emergency break fired (absolute session-age cap) — agent gone without completion",
+                    ClassifierRules.R3EmergencyBreak);
 
             // 4. The user demonstrably finished setup: a real-user desktop was observed AND
             //    Windows Hello reached a positive terminal (provisioned or skipped). Those are
@@ -371,7 +411,7 @@ namespace AutopilotMonitor.Functions.Services
                     : "agent went silent before reporting completion";
                 return (SessionStatus.Succeeded, AppendReconcileTiming(
                     $"Reconciled at timeout: user completed setup ({evidence}) — {detail}",
-                    startedAtUtc, nowUtc, lastEventAtUtc));
+                    startedAtUtc, nowUtc, lastEventAtUtc), ClassifierRules.R4DesktopHello);
             }
 
             // 5. Device Setup fully provisioned (device is AADJ + MDM-enrolled), user phase pending.
@@ -384,7 +424,8 @@ namespace AutopilotMonitor.Functions.Services
                         ? $" (Account Setup {rollup.AccountSetupSucceededCount}/{rollup.AccountSetupTotal})"
                         : " (Account Setup not yet started)";
                     return (SessionStatus.AwaitingUser,
-                        $"Device Setup completed; awaiting user / Account Setup phase — agent silent, within {graceHours}h grace{acct}");
+                        $"Device Setup completed; awaiting user / Account Setup phase — agent silent, within {graceHours}h grace{acct}",
+                        ClassifierRules.R5DeviceSetupAwaiting);
                 }
 
                 // "0/0" would suggest a parsed rollup that never existed — when no Account Setup
@@ -393,12 +434,14 @@ namespace AutopilotMonitor.Functions.Services
                     ? $"last Account Setup {rollup.AccountSetupSucceededCount}/{rollup.AccountSetupTotal}"
                     : "Account Setup progress never observed";
                 return (SessionStatus.Incomplete,
-                    $"No completion signal within {graceHours}h grace after Device Setup completed ({acctDetail})");
+                    $"No completion signal within {graceHours}h grace after Device Setup completed ({acctDetail})",
+                    ClassifierRules.R5DeviceSetupIncomplete);
             }
 
             // 6. Silent before Device Setup completed, with no explicit failure → unknown, not a failure.
             return (SessionStatus.Incomplete,
-                "No Device Setup completion or explicit failure signal observed before timeout");
+                "No Device Setup completion or explicit failure signal observed before timeout",
+                ClassifierRules.R6Fallthrough);
         }
 
         /// <summary>

src/Backend/AutopilotMonitor.Functions/Services/EventIngestProcessor.Classification.cs

@@ -286,9 +286,10 @@ namespace AutopilotMonitor.Functions.Services
             var effectiveStart = session?.ResumedAt ?? session?.StartedAt ?? triggerEvent.Timestamp;
 
             var rollup = EnrollmentTimeoutClassifier.ExtractRollup(sessionEvents);
-            var (targetStatus, reason) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+            var (targetStatus, reason, rule) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
                 rollup, effectiveStart, now, graceHours, session?.LastEventAt ?? c.LatestEventTimestamp,
-                isPreProvisioned: session?.IsPreProvisioned == true, resumedAt: session?.ResumedAt);
+                isPreProvisioned: session?.IsPreProvisioned == true, resumedAt: session?.ResumedAt,
+                isSelfDeployingProfile: session?.IsSelfDeployingProfile == true);
 
             // Keep the max-lifetime trigger visible: the Succeeded reasons already carry their own
             // silence-transparency clause, everything else gets the watchdog context appended.
@@ -314,7 +315,8 @@ namespace AutopilotMonitor.Functions.Services
             }
 
             var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                request.TenantId, request.SessionId, targetStatus, triggerEvent.Phase, reason,
+                request.TenantId, request.SessionId, targetStatus, VerdictPaths.Compose(VerdictPaths.OriginMaxLifetime, rule),
+                triggerEvent.Phase, reason,
                 completedAt: isTerminalNonSuccess ? triggerEvent.Timestamp : (DateTime?)null,
                 earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                 failureSource: isTerminalNonSuccess ? "max_lifetime_watchdog" : null,
@@ -356,15 +358,17 @@ namespace AutopilotMonitor.Functions.Services
 
                 var now = DateTime.UtcNow;
                 var effectiveStart = session.ResumedAt ?? session.StartedAt;
-                var (targetStatus, reason) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+                var (targetStatus, reason, rule) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
                     rollup, effectiveStart, now, graceHours, session.LastEventAt ?? c.LatestEventTimestamp,
-                    isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt);
+                    isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt,
+                    isSelfDeployingProfile: session.IsSelfDeployingProfile);
 
                 if (targetStatus != SessionStatus.Succeeded)
                     return;
 
                 var healed = await _sessionRepo.UpdateSessionStatusAsync(
                     request.TenantId, request.SessionId, SessionStatus.Succeeded,
+                    VerdictPaths.Compose(VerdictPaths.OriginLateReconcile, rule),
                     failureReason: reason,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp);
                 if (healed)
@@ -404,7 +408,7 @@ namespace AutopilotMonitor.Functions.Services
                     return false;
 
                 var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                    request.TenantId, request.SessionId, SessionStatus.AwaitingUser,
+                    request.TenantId, request.SessionId, SessionStatus.AwaitingUser, VerdictPaths.IngestWhiteGloveAwaiting,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                     failureReason: EnrollmentTimeoutClassifier.WhiteGloveAwaitingUserReason(rollup));
                 if (transitioned)
@@ -463,7 +467,9 @@ namespace AutopilotMonitor.Functions.Services
                     ? c.CompletionEvent.Data["completionSource"]?.ToString() : null;
 
                 statusTransitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                    request.TenantId, request.SessionId, SessionStatus.Succeeded, c.CompletionEvent.Phase,
+                    request.TenantId, request.SessionId, SessionStatus.Succeeded,
+                    espSoftFailure ? VerdictPaths.AgentCompleteSoft : VerdictPaths.AgentComplete,
+                    c.CompletionEvent.Phase,
                     completedAt: c.CompletionEvent.Timestamp,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                     espSoftFailure: espSoftFailure, completionSource: completionSource);
@@ -490,7 +496,8 @@ namespace AutopilotMonitor.Functions.Services
                         : c.FailureEvent.Message;
 
                     statusTransitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                        request.TenantId, request.SessionId, SessionStatus.Failed, c.FailureEvent.Phase, failureReason,
+                        request.TenantId, request.SessionId, SessionStatus.Failed, VerdictPaths.AgentFailed,
+                        c.FailureEvent.Phase, failureReason,
                         completedAt: c.FailureEvent.Timestamp,
                         earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp);
                     _logger.LogWarning("{SessionPrefix} Status: Failed - {FailureReason} (transitioned={Transitioned})", sessionPrefix, failureReason, statusTransitioned);
@@ -500,7 +507,8 @@ namespace AutopilotMonitor.Functions.Services
             {
                 failureReason = c.EspFailureEvent.Message ?? "ESP failure (backend fallback)";
                 statusTransitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                    request.TenantId, request.SessionId, SessionStatus.Failed, c.EspFailureEvent.Phase, failureReason,
+                    request.TenantId, request.SessionId, SessionStatus.Failed, VerdictPaths.AgentEspFailureFallback,
+                    c.EspFailureEvent.Phase, failureReason,
                     completedAt: c.EspFailureEvent.Timestamp,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp);
                 _logger.LogWarning("{SessionPrefix} Status: Failed via esp_failure fallback - {FailureReason} (transitioned={Transitioned})",
@@ -509,7 +517,8 @@ namespace AutopilotMonitor.Functions.Services
             else if (c.GatherCompletionEvent != null)
             {
                 await _sessionRepo.UpdateSessionStatusAsync(
-                    request.TenantId, request.SessionId, SessionStatus.Succeeded, c.GatherCompletionEvent.Phase,
+                    request.TenantId, request.SessionId, SessionStatus.Succeeded, VerdictPaths.AgentGatherComplete,
+                    c.GatherCompletionEvent.Phase,
                     completedAt: c.GatherCompletionEvent.Timestamp,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp);
                 _logger.LogInformation("{SessionPrefix} Status: Succeeded (gather_rules)", sessionPrefix);
@@ -517,7 +526,8 @@ namespace AutopilotMonitor.Functions.Services
             else if (c.WhiteGloveEvent != null)
             {
                 whiteGloveStatusTransitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                    request.TenantId, request.SessionId, SessionStatus.Pending, EnrollmentPhase.AppsDevice,
+                    request.TenantId, request.SessionId, SessionStatus.Pending, VerdictPaths.AgentWhiteGlovePending,
+                    EnrollmentPhase.AppsDevice,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                     isPreProvisioned: true, isUserDriven: false);
 
@@ -538,7 +548,7 @@ namespace AutopilotMonitor.Functions.Services
                         _logger.LogWarning("{SessionPrefix} WhiteGlove UpdateSessionStatusAsync failed, attempting unconditional fallback for IsPreProvisioned + Status", sessionPrefix);
                         try
                         {
-                            await _sessionRepo.SetSessionPreProvisionedAsync(request.TenantId, request.SessionId, true, SessionStatus.Pending, isUserDriven: false);
+                            await _sessionRepo.SetSessionPreProvisionedAsync(request.TenantId, request.SessionId, true, SessionStatus.Pending, isUserDriven: false, verdictPath: VerdictPaths.AgentWhiteGlovePending);
                             whiteGloveStatusTransitioned = true;
                             _logger.LogInformation("{SessionPrefix} WhiteGlove fallback succeeded: IsPreProvisioned + Status=Pending set via unconditional merge", sessionPrefix);
                         }
@@ -578,7 +588,8 @@ namespace AutopilotMonitor.Functions.Services
                 if (currentSession?.Status == SessionStatus.Pending)
                 {
                     await _sessionRepo.UpdateSessionStatusAsync(
-                        request.TenantId, request.SessionId, SessionStatus.InProgress, c.WhiteGloveResumedEvent.Phase,
+                        request.TenantId, request.SessionId, SessionStatus.InProgress, VerdictPaths.AgentWhiteGloveResumed,
+                        c.WhiteGloveResumedEvent.Phase,
                         earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                         isUserDriven: true, resumedAt: c.WhiteGloveResumedEvent.Timestamp);
                     _logger.LogInformation("{SessionPrefix} Status: InProgress (WhiteGlove Part 2 resumed, IsUserDriven=true)", sessionPrefix);
@@ -624,7 +635,7 @@ namespace AutopilotMonitor.Functions.Services
 
                 var stalledReason = "Agent reported stall after 60min without progress (stall_probe)";
                 var stalledTransitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                    request.TenantId, request.SessionId, SessionStatus.Stalled,
+                    request.TenantId, request.SessionId, SessionStatus.Stalled, VerdictPaths.AgentStallProbe,
                     earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                     stalledAt: c.SessionStalledEvent.Timestamp, failureReason: stalledReason);
                 if (stalledTransitioned)
@@ -647,7 +658,7 @@ namespace AutopilotMonitor.Functions.Services
                 if (currentStatus == SessionStatus.Stalled)
                 {
                     var healed = await _sessionRepo.UpdateSessionStatusAsync(
-                        request.TenantId, request.SessionId, SessionStatus.InProgress,
+                        request.TenantId, request.SessionId, SessionStatus.InProgress, VerdictPaths.AgentStallHeal,
                         earliestEventTimestamp: c.EarliestEventTimestamp, latestEventTimestamp: c.LatestEventTimestamp,
                         clearStalledAt: true, clearFailureReason: true);
                     if (healed)

src/Backend/AutopilotMonitor.Functions/Services/LegacyReclassificationService.cs

@@ -18,6 +18,11 @@ namespace AutopilotMonitor.Functions.Services
     /// registered a LATER session (Part 2 under a fresh session id, or re-provisioning) are
     /// resolved as Incomplete("Superseded by ..."). Forward-looking coverage lives in the
     /// registration supersede pass; this mode clears the existing backlog.</item>
+    /// <item><b>self_deploying_silent</b> — self-deploying-profile rows parked as
+    /// Incomplete / AwaitingUser / Stalled by the pre-2026-08-23 classifier (which had no
+    /// notion of a profile without a user phase) are re-run through the classifier; only a
+    /// Succeeded verdict (<see cref="EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned"/>)
+    /// is applied, everything else keeps its verdict.</item>
     /// </list>
     /// Hand-marked sessions (AdminMarkedAction) are never touched.
     /// </summary>
@@ -30,6 +35,7 @@ namespace AutopilotMonitor.Functions.Services
 
         public const string ModeLegacyTimeouts = "legacy_timeouts";
         public const string ModePendingOrphans = "pending_orphans";
+        public const string ModeSelfDeployingSilent = "self_deploying_silent";
         private const int MaxSamplesInResponse = 100;
 
         public LegacyReclassificationService(
@@ -108,9 +114,10 @@ namespace AutopilotMonitor.Functions.Services
                         var events = await _sessionRepo.GetSessionEventsAsync(tenantId, session.SessionId, maxResults: 1000);
                         var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
                         var effectiveStart = session.ResumedAt ?? session.StartedAt;
-                        var (target, reason) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+                        var (target, reason, rule) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
                             rollup, effectiveStart, now, graceHours, session.LastEventAt,
-                            isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt);
+                            isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt,
+                            isSelfDeployingProfile: session.IsSelfDeployingProfile);
 
                         if (target == SessionStatus.Failed)
                         {
@@ -138,7 +145,7 @@ namespace AutopilotMonitor.Functions.Services
                         }
 
                         var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                            tenantId, session.SessionId, target,
+                            tenantId, session.SessionId, target, VerdictPaths.Compose(VerdictPaths.OriginRetro, rule),
                             failureReason: reason,
                             failureSource: target == SessionStatus.Incomplete ? "legacy_timeout_retro_reconcile" : null,
                             allowTerminalReclassification: true);
@@ -165,6 +172,99 @@ namespace AutopilotMonitor.Functions.Services
             return result;
         }
 
+        public async Task<ReclassificationResult> ReconcileSelfDeployingSilentAsync(
+            string? tenantIdScope, bool dryRun, int maxSessions, string triggeredBy)
+        {
+            var result = new ReclassificationResult { Mode = ModeSelfDeployingSilent, DryRun = dryRun };
+            var now = DateTime.UtcNow;
+
+            foreach (var tenantId in await ResolveTenantsAsync(tenantIdScope))
+            {
+                if (result.SessionsExamined >= maxSessions) { result.CapReached = true; break; }
+                result.TenantsExamined++;
+
+                int? tenantGraceHours = null, absoluteMaxHours = null;
+                try
+                {
+                    var config = await _configService.GetConfigurationAsync(tenantId);
+                    tenantGraceHours = config?.SessionGraceHours;
+                    absoluteMaxHours = config?.AbsoluteMaxSessionHours;
+                }
+                catch { /* defaults */ }
+                var graceHours = EnrollmentTimeoutClassifier.ResolveGraceHours(tenantGraceHours, absoluteMaxHours);
+
+                var candidates = await _maintenanceRepo.GetSelfDeployingSilentSessionsAsync(
+                    tenantId, maxSessions - result.SessionsExamined + 1);
+                if (candidates.Count > maxSessions - result.SessionsExamined)
+                {
+                    result.CapReached = true;
+                    candidates = candidates.Take(maxSessions - result.SessionsExamined).ToList();
+                }
+
+                foreach (var session in candidates)
+                {
+                    result.SessionsExamined++;
+                    try
+                    {
+                        if (!string.IsNullOrEmpty(session.AdminMarkedAction))
+                        {
+                            result.Skipped++;
+                            continue;
+                        }
+
+                        var events = await _sessionRepo.GetSessionEventsAsync(tenantId, session.SessionId, maxResults: 1000);
+                        var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
+                        var effectiveStart = session.ResumedAt ?? session.StartedAt;
+                        var (target, reason, _) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+                            rollup, effectiveStart, now, graceHours, session.LastEventAt,
+                            isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt,
+                            isSelfDeployingProfile: session.IsSelfDeployingProfile);
+
+                        // Heal-only: a verdict other than Succeeded means the device never reached
+                        // Device ESP all-succeeded (or failed explicitly) — the existing label stands.
+                        if (target != SessionStatus.Succeeded)
+                        {
+                            result.Skipped++;
+                            continue;
+                        }
+
+                        RecordSample(result, tenantId, session, target, reason, oldStatus: session.Status.ToString());
+
+                        if (dryRun)
+                        {
+                            result.WouldChange++;
+                            continue;
+                        }
+
+                        // Succeeded may upgrade Incomplete / AwaitingUser / Stalled through the normal
+                        // reconcile gate (no terminal-reclassification override needed); the storage
+                        // layer additionally refuses admin-marked rows.
+                        var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
+                            tenantId, session.SessionId, SessionStatus.Succeeded, VerdictPaths.RetroSelfDeployingReconcile,
+                            failureReason: reason);
+                        if (transitioned)
+                        {
+                            result.Changed++;
+                            result.ToSucceeded++;
+                        }
+                        else
+                        {
+                            result.Skipped++;
+                        }
+                    }
+                    catch (Exception ex)
+                    {
+                        result.Errors++;
+                        _logger.LogWarning(ex, "Self-deploying retro-reconcile failed for session {SessionId} (tenant {TenantId})",
+                            session.SessionId, tenantId);
+                    }
+                }
+            }
+
+            await WriteAuditAsync(result, tenantIdScope, triggeredBy);
+            return result;
+        }
+
         public async Task<ReclassificationResult> ResolvePendingOrphansAsync(
             string? tenantIdScope, bool dryRun, int maxSessions, string triggeredBy)
         {
@@ -214,7 +314,7 @@ namespace AutopilotMonitor.Functions.Services
 
                         // Pending is non-terminal — the standard transition guard covers this write.
                         var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
-                            tenantId, pending.SessionId, SessionStatus.Incomplete,
+                            tenantId, pending.SessionId, SessionStatus.Incomplete, VerdictPaths.RetroSuperseded,
                             failureReason: reason, failureSource: "superseded_by_reregistration");
                         if (transitioned) { result.Changed++; result.ToIncomplete++; }
                         else result.Skipped++;

src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.Aggregation.cs

@@ -289,14 +289,13 @@ namespace AutopilotMonitor.Functions.Services
         /// Sessions with a deletion cascade in flight are skipped so the sweep cannot write a
         /// breakdown row after the session's deletion manifest was snapshotted.
         /// </summary>
-        private async Task SweepTimeAttributionAsync()
+        private async Task SweepTimeAttributionAsync(IReadOnlyList<SessionSummary> sweepWindow)
         {
             try
             {
                 var sw = Stopwatch.StartNew();
                 var today = DateTime.UtcNow.Date;
-                var sessions = await _maintenanceRepo.GetSessionsByDateRangeAsync(
-                    today.AddDays(-TimeAttributionSweepDays), today.AddDays(1));
+                var sessions = SliceSweepWindow(sweepWindow, today.AddDays(-TimeAttributionSweepDays));
 
                 var pairs = new List<(SessionSummary Session, SessionTimeBreakdown Breakdown)>();
                 var missing = new List<SessionSummary>();
@@ -1102,6 +1101,19 @@ namespace AutopilotMonitor.Functions.Services
             {
                 _logger.LogWarning(ex, "Failed to cleanup old device-journey aggregates");
             }
+
+            // Verdict-calibration daily aggregates: same 180d window (regenerable; the radar
+            // needs 35d, the matrix serves up to 180d).
+            try
+            {
+                var deleted = await _metricsRepo.DeleteVerdictCalibrationAggregatesOlderThanAsync(now.AddDays(-usageMetricsRetentionDays));
+                if (deleted > 0)
+                    _logger.LogInformation("Verdict-calibration aggregate cleanup: deleted {Count} rows older than {Days} days", deleted, usageMetricsRetentionDays);
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "Failed to cleanup old verdict-calibration aggregates");
+            }
         }
         /// <summary>
         /// Detects and cleans up orphaned events — events stored in the Events table

src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.DeviceJourneys.cs

@@ -36,7 +36,7 @@ namespace AutopilotMonitor.Functions.Services
         /// per window device per tick — same order as the F1 sweep; negligible at current fleet
         /// size (the scaling lever is gating the per-device RMW on new terminal sessions).
         /// </summary>
-        private async Task SweepDeviceJourneysAsync()
+        private async Task SweepDeviceJourneysAsync(IReadOnlyList<SessionSummary> sweepWindow)
         {
             try
             {
@@ -84,7 +84,7 @@ namespace AutopilotMonitor.Functions.Services
                 }
 
                 // ---- Phase 2: merge the window's terminal sessions into their device chains. ----
-                var sessions = await _maintenanceRepo.GetSessionsByDateRangeAsync(windowStart, windowEnd);
+                var sessions = SliceSweepWindow(sweepWindow, windowStart);
                 var excludedSessions = new List<SessionSummary>();
                 var byDevice = new Dictionary<(string TenantId, string SerialKey), List<SessionSummary>>();
 

src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.VerdictCalibration.cs

@@ -0,0 +1,259 @@
+using System;
+using System.Collections.Generic;
+using System.Diagnostics;
+using System.Globalization;
+using System.Linq;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.Models;
+using Microsoft.Extensions.Logging;
+
+namespace AutopilotMonitor.Functions.Services
+{
+    /// <summary>
+    /// Partial: verdict-calibration daily aggregates (docs/backend/verdict-calibration.md) — the
+    /// operator thermometer for the rule classifier: which verdict path produced how many
+    /// sessions per day, how many of those were later overridden (admin / late completion /
+    /// other), and how many devices re-enrolled within 7 days.
+    /// </summary>
+    public partial class MaintenanceService
+    {
+        /// <summary>App setting kill switch: "true" skips the sweep. Fail-open — the sweep only writes its own regenerable table.</summary>
+        internal const string VerdictCalibrationKillSwitchSetting = "VerdictCalibrationSweepDisabled";
+
+        /// <summary>Rolling window: sessions started in the last 30 days (same as the F1/F2 sweeps).</summary>
+        private const int VerdictCalibrationSweepDays = 30;
+
+        /// <summary>Re-enrollment proxy horizon: another terminal session of the same device within this many days.</summary>
+        internal const int VerdictCalibrationReEnrollDays = 7;
+
+        /// <summary>
+        /// Recomputes the verdict-calibration rows for the rolling window — idempotent per
+        /// (tenant × StartedAt date) plus "global" mirror rows. Slices the tick's shared window
+        /// scan (no own drain) and reads each window tenant's device histories (one partition
+        /// scan per tenant) for the re-enrollment proxy. Stale date rows a run did not regenerate
+        /// are deleted so window sums never serve ghost counts.
+        /// </summary>
+        private async Task SweepVerdictCalibrationAsync(IReadOnlyList<SessionSummary> sweepWindow)
+        {
+            try
+            {
+                if (string.Equals(_configuration[VerdictCalibrationKillSwitchSetting], "true", StringComparison.OrdinalIgnoreCase))
+                {
+                    _logger.LogInformation("Verdict-calibration sweep skipped ({Setting}=true)", VerdictCalibrationKillSwitchSetting);
+                    return;
+                }
+
+                var sw = Stopwatch.StartNew();
+                var now = DateTime.UtcNow;
+                var today = now.Date;
+                var windowStart = today.AddDays(-VerdictCalibrationSweepDays);
+                var windowEnd = today.AddDays(1);
+
+                var sessions = SliceSweepWindow(sweepWindow, windowStart);
+
+                var historiesByTenant = new Dictionary<string, List<DeviceHistory>>(StringComparer.OrdinalIgnoreCase);
+                foreach (var tenantId in sessions.Select(s => s.TenantId).Distinct(StringComparer.OrdinalIgnoreCase))
+                {
+                    historiesByTenant[tenantId] = await _metricsRepo.GetDeviceHistoriesByTenantAsync(tenantId);
+                }
+
+                var aggregates = BuildVerdictCalibrationAggregates(sessions, historiesByTenant, now);
+                var saved = 0;
+                foreach (var aggregate in aggregates)
+                {
+                    if (await _metricsRepo.SaveVerdictCalibrationAggregateAsync(aggregate))
+                        saved++;
+                }
+
+                var generatedKeys = new HashSet<(string TenantId, string Date)>(aggregates.Select(a => (a.TenantId, a.Date)));
+                var removedStale = 0;
+                foreach (var partition in aggregates.Select(a => a.TenantId).Distinct().ToList())
+                {
+                    var existingRows = await _metricsRepo.GetVerdictCalibrationAggregatesAsync(partition, windowStart, windowEnd.AddDays(-1));
+                    foreach (var row in existingRows)
+                    {
+                        if (generatedKeys.Contains((partition, row.Date))) continue;
+                        await _metricsRepo.DeleteVerdictCalibrationAggregateAsync(partition, row.Date);
+                        removedStale++;
+                    }
+                }
+
+                sw.Stop();
+                _logger.LogInformation(
+                    "Verdict-calibration sweep: {Sessions} window sessions across {Tenants} tenants, {Rows} rows written, {Removed} stale rows removed in {Ms}ms",
+                    sessions.Count, historiesByTenant.Count, saved, removedStale, sw.ElapsedMilliseconds);
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "Verdict-calibration sweep failed (non-fatal)");
+            }
+        }
+
+        /// <summary>
+        /// Pure aggregation core behind <see cref="SweepVerdictCalibrationAsync"/> — internal
+        /// static so the contract is pinned by unit tests. Per session: the current
+        /// (VerdictPath, Status) bucket gets <c>Count</c> (+ <c>DerivedCount</c> when the path was
+        /// derived read-side), terminal sessions old enough get <c>Eligible7d</c> and, if the
+        /// device's history chain shows another terminal session starting within 7 days of this
+        /// one's end, <c>ReEnrolled7d</c>; a session carrying a prior verdict additionally
+        /// increments the PRIOR bucket's <c>Overridden*</c> counter. Sessions inside a deletion
+        /// cascade are skipped. Rows are emitted per (tenant, StartedAt date) and mirrored into
+        /// the "global" partition; buckets are sorted for deterministic output.
+        /// </summary>
+        internal static List<VerdictCalibrationDailyAggregate> BuildVerdictCalibrationAggregates(
+            IReadOnlyList<SessionSummary> sessions,
+            IReadOnlyDictionary<string, List<DeviceHistory>> historiesByTenant,
+            DateTime nowUtc)
+        {
+            var nextStartBySession = BuildNextStartLookup(historiesByTenant);
+            var eligibleCutoff = nowUtc.AddDays(-VerdictCalibrationReEnrollDays);
+
+            var rows = new Dictionary<(string TenantId, string Date), RowAccumulator>();
+
+            RowAccumulator Row(string tenantId, string date)
+            {
+                var key = (tenantId, date);
+                if (!rows.TryGetValue(key, out var acc))
+                {
+                    acc = new RowAccumulator();
+                    rows[key] = acc;
+                }
+                return acc;
+            }
+
+            foreach (var session in sessions)
+            {
+                if (session == null) continue;
+                if (!string.IsNullOrEmpty(session.DeletionState) && session.DeletionState != "None")
+                    continue;
+
+                var date = session.StartedAt.ToString("yyyy-MM-dd", CultureInfo.InvariantCulture);
+                var (path, derived) = VerdictPathDerivation.Derive(session);
+                var statusName = session.Status.ToString();
+                var isTerminal = DeviceJourneyCalculator.IsTerminal(session.Status);
+
+                var eligible = false;
+                var reEnrolled = false;
+                if (isTerminal)
+                {
+                    var end = session.CompletedAt ?? session.LastEventAt ?? session.StartedAt;
+                    eligible = end <= eligibleCutoff;
+                    if (eligible
+                        && nextStartBySession.TryGetValue((session.TenantId, session.SessionId), out var next)
+                        && next.NextStart.HasValue
+                        && (next.NextStart.Value - next.End) < TimeSpan.FromDays(VerdictCalibrationReEnrollDays))
+                    {
+                        reEnrolled = true;
+                    }
+                }
+
+                var overridden = ClassifyOverride(session);
+
+                foreach (var partition in new[] { session.TenantId, "global" })
+                {
+                    var acc = Row(partition, date);
+                    acc.SessionCount++;
+                    if (isTerminal) acc.TerminalSessionCount++;
+
+                    var bucket = acc.Bucket(path, statusName);
+                    bucket.Count++;
+                    if (derived) bucket.DerivedCount++;
+                    if (eligible) bucket.Eligible7d++;
+                    if (reEnrolled) bucket.ReEnrolled7d++;
+
+                    if (overridden.HasValue)
+                    {
+                        var prior = acc.Bucket(session.PriorVerdictPath!, session.PriorStatus!);
+                        switch (overridden.Value)
+                        {
+                            case OverrideKind.Admin: prior.OverriddenByAdmin++; break;
+                            case OverrideKind.LateCompletion: prior.OverriddenByLateCompletion++; break;
+                            default: prior.OverriddenOther++; break;
+                        }
+                    }
+                }
+            }
+
+            return rows
+                .OrderBy(r => r.Key.TenantId, StringComparer.Ordinal)
+                .ThenBy(r => r.Key.Date, StringComparer.Ordinal)
+                .Select(r => new VerdictCalibrationDailyAggregate
+                {
+                    TenantId = r.Key.TenantId,
+                    Date = r.Key.Date,
+                    Version = VerdictCalibrationDailyAggregate.CurrentVersion,
+                    SessionCount = r.Value.SessionCount,
+                    TerminalSessionCount = r.Value.TerminalSessionCount,
+                    Buckets = r.Value.Buckets.Values
+                        .OrderBy(b => b.VerdictPath, StringComparer.Ordinal)
+                        .ThenBy(b => b.Status, StringComparer.Ordinal)
+                        .ToList(),
+                    ComputedAt = nowUtc,
+                })
+                .ToList();
+        }
+
+        private enum OverrideKind { Admin, LateCompletion, Other }
+
+        /// <summary>
+        /// A session whose current status overrode a prior verdict names the writer that did it:
+        /// an admin mark (AdminMarkedAction set), a late agent completion (now Succeeded without
+        /// admin involvement), or anything else (retro-reclassification, grace expiry, supersede).
+        /// </summary>
+        private static OverrideKind? ClassifyOverride(SessionSummary session)
+        {
+            if (string.IsNullOrEmpty(session.PriorVerdictPath) || string.IsNullOrEmpty(session.PriorStatus))
+                return null;
+            if (!string.IsNullOrEmpty(session.AdminMarkedAction))
+                return OverrideKind.Admin;
+            if (session.Status == SessionStatus.Succeeded)
+                return OverrideKind.LateCompletion;
+            return OverrideKind.Other;
+        }
+
+        /// <summary>
+        /// (tenant, sessionId) → this session's chain end and the next chain entry's StartedAt —
+        /// the re-enrollment proxy source. Chains are ascending by StartedAt and capped at 20, so
+        /// a window session older than the cap simply has no entry (counts as not re-enrolled
+        /// only when eligible — the cap is a known, small bias toward under-counting).
+        /// </summary>
+        internal static Dictionary<(string TenantId, string SessionId), (DateTime End, DateTime? NextStart)> BuildNextStartLookup(
+            IReadOnlyDictionary<string, List<DeviceHistory>> historiesByTenant)
+        {
+            var lookup = new Dictionary<(string, string), (DateTime, DateTime?)>();
+            foreach (var pair in historiesByTenant)
+            {
+                foreach (var history in pair.Value)
+                {
+                    var chain = history.Chain;
+                    for (var i = 0; i < chain.Count; i++)
+                    {
+                        var end = chain[i].CompletedAt ?? chain[i].StartedAt;
+                        DateTime? nextStart = i + 1 < chain.Count ? chain[i + 1].StartedAt : null;
+                        lookup[(pair.Key, chain[i].SessionId)] = (end, nextStart);
+                    }
+                }
+            }
+            return lookup;
+        }
+
+        private sealed class RowAccumulator
+        {
+            public int SessionCount;
+            public int TerminalSessionCount;
+            public readonly Dictionary<(string Path, string Status), VerdictCalibrationBucket> Buckets = new();
+
+            public VerdictCalibrationBucket Bucket(string path, string status)
+            {
+                var key = (path, status);
+                if (!Buckets.TryGetValue(key, out var bucket))
+                {
+                    bucket = new VerdictCalibrationBucket { VerdictPath = path, Status = status };
+                    Buckets[key] = bucket;
+                }
+                return bucket;
+            }
+        }
+    }
+}

src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.VerdictCalibrationRadar.cs

@@ -0,0 +1,206 @@
+using System;
+using System.Collections.Generic;
+using System.Diagnostics;
+using System.Globalization;
+using System.Linq;
+using System.Threading.Tasks;
+using AutopilotMonitor.Functions.Helpers;
+using AutopilotMonitor.Shared.Models;
+using Microsoft.Extensions.Logging;
+
+namespace AutopilotMonitor.Functions.Services
+{
+    /// <summary>
+    /// Partial: verdict-calibration drift radar (docs/backend/verdict-calibration.md). Same
+    /// episode/tracker pattern as the rule-regression radar, over the verdict-calibration daily
+    /// rows (<see cref="VerdictCalibrationRadar"/>). Operator-only: fires a VerdictCalibrationDrift
+    /// ops event once per episode and surfaces the episode in the calibration endpoint's
+    /// <c>alerts[]</c>; there is deliberately no tenant bell.
+    /// </summary>
+    public partial class MaintenanceService
+    {
+        /// <summary>App setting kill switch: "true" skips the radar. Fail-open — it only notifies.</summary>
+        internal const string VerdictCalibrationRadarKillSwitchSetting = "VerdictCalibrationRadarDisabled";
+
+        /// <summary>
+        /// Evaluates every partition that has calibration rows in the horizon — each tenant plus
+        /// the "global" mirror (a platform-wide drift is its own signal; per-tenant alerts catch
+        /// the local ones). Anchored on <paramref name="targetDate"/> (timer path: yesterday —
+        /// whole days only). Idempotent per anchor via the tracker dedup; fail-soft per partition.
+        /// </summary>
+        private async Task RunVerdictCalibrationRadarAsync(DateTime targetDate, IReadOnlyList<SessionSummary> sweepWindow)
+        {
+            try
+            {
+                if (string.Equals(_configuration[VerdictCalibrationRadarKillSwitchSetting], "true", StringComparison.OrdinalIgnoreCase))
+                {
+                    _logger.LogInformation("Verdict-calibration radar skipped ({Setting}=true)", VerdictCalibrationRadarKillSwitchSetting);
+                    return;
+                }
+
+                var sw = Stopwatch.StartNew();
+                var horizonStart = targetDate.Date.AddDays(-(VerdictCalibrationRadar.WindowDays - 1 + VerdictCalibrationRadar.BaselineDays));
+
+                // Tenants are discovered from the tick's shared window scan (the calibration
+                // sweep wrote a row per tenant it saw) — no own drain.
+                var tenantIds = sweepWindow
+                    .Select(s => s.TenantId)
+                    .Where(t => !string.IsNullOrEmpty(t))
+                    .Distinct(StringComparer.OrdinalIgnoreCase)
+                    .ToList();
+                tenantIds.Add("global");
+
+                int fired = 0, refreshed = 0, rearmed = 0;
+                foreach (var partition in tenantIds)
+                {
+                    try
+                    {
+                        var rows = await _metricsRepo.GetVerdictCalibrationAggregatesAsync(partition, horizonStart, targetDate.Date);
+                        if (rows.Count == 0) continue;
+                        var (f, r, a) = await EvaluateVerdictCalibrationPartitionAsync(partition, rows, targetDate);
+                        fired += f; refreshed += r; rearmed += a;
+                    }
+                    catch (Exception ex)
+                    {
+                        _logger.LogWarning(ex, "Verdict-calibration radar failed for partition {Partition} (non-fatal)", partition);
+                    }
+                }
+
+                sw.Stop();
+                _logger.LogInformation(
+                    "Verdict-calibration radar: {Partitions} partitions, {Fired} fired, {Refreshed} refreshed, {Rearmed} re-armed in {Ms}ms (anchor {Anchor})",
+                    tenantIds.Count, fired, refreshed, rearmed, sw.ElapsedMilliseconds, targetDate.ToString("yyyy-MM-dd"));
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "Verdict-calibration radar failed (non-fatal)");
+            }
+        }
+
+        private async Task<(int Fired, int Refreshed, int Rearmed)> EvaluateVerdictCalibrationPartitionAsync(
+            string partition, List<VerdictCalibrationDailyAggregate> rows, DateTime targetDate)
+        {
+            var findings = VerdictCalibrationRadar.Evaluate(rows, targetDate);
+            var active = await _hardwareRejectionTracker.GetVerdictCalibrationAlertsAsync(partition);
+            var activeByKey = active.ToDictionary(a => AlertKey(a.Kind, a.VerdictPath, a.Status), StringComparer.OrdinalIgnoreCase);
+
+            int fired = 0, refreshed = 0, rearmed = 0;
+            var firedKeys = new HashSet<string>(StringComparer.OrdinalIgnoreCase);
+
+            foreach (var finding in findings)
+            {
+                var key = AlertKey(finding.Kind, finding.VerdictPath, finding.Status);
+                firedKeys.Add(key);
+
+                if (activeByKey.TryGetValue(key, out var existing))
+                {
+                    await _hardwareRejectionTracker.RefreshVerdictCalibrationAlertAsync(
+                        partition, BuildVerdictAlert(partition, finding, existing.Dimension, existing.FirstNotifiedAt));
+                    refreshed++;
+                    continue;
+                }
+
+                var dimension = await TryCorrelateVerdictDimensionsAsync(partition, finding);
+                var alert = BuildVerdictAlert(partition, finding, dimension, DateTime.UtcNow);
+                if (!await _hardwareRejectionTracker.TryRegisterVerdictCalibrationAlertAsync(partition, alert))
+                    continue;
+
+                fired++;
+                await _opsEventService.RecordVerdictCalibrationDriftAsync(
+                    partition, finding.Kind, finding.VerdictPath, finding.Status,
+                    finding.WindowHitCount, finding.WindowSessionCount, finding.WindowRatePct,
+                    finding.BaselineHitCount, finding.BaselineSessionCount, finding.BaselineRatePct,
+                    finding.Lift, DescribeDimension(dimension));
+            }
+
+            foreach (var alert in active)
+            {
+                if (firedKeys.Contains(AlertKey(alert.Kind, alert.VerdictPath, alert.Status)))
+                    continue;
+
+                if (VerdictCalibrationRadar.ShouldReArm(alert, rows, targetDate))
+                {
+                    await _hardwareRejectionTracker.DeleteVerdictCalibrationAlertAsync(partition, alert.Kind, alert.VerdictPath, alert.Status);
+                    rearmed++;
+                }
+                else
+                {
+                    // Elevated but no longer fully separated: keep the episode, refresh the numbers.
+                    var (wh, ws, bh, bs) = VerdictCalibrationRadar.CurrentSums(alert, rows, targetDate);
+                    alert.WindowHitCount = wh;
+                    alert.WindowSessionCount = ws;
+                    alert.BaselineHitCount = bh;
+                    alert.BaselineSessionCount = bs;
+                    alert.WindowRatePct = ws > 0 ? Math.Round(100.0 * wh / ws, 1) : 0;
+                    alert.BaselineRatePct = bs > 0 ? Math.Round(100.0 * bh / bs, 1) : 0;
+                    alert.Lift = alert.Kind == VerdictCalibrationAlertKinds.EvidenceGap || bh == 0 || bs == 0 || ws == 0
+                        ? null
+                        : Math.Round(((double)wh / ws) / ((double)bh / bs), 1);
+                    alert.WindowStartDate = targetDate.Date.AddDays(-(VerdictCalibrationRadar.WindowDays - 1)).ToString("yyyy-MM-dd");
+                    alert.WindowEndDate = targetDate.Date.ToString("yyyy-MM-dd");
+                    alert.LastEvaluatedAt = DateTime.UtcNow;
+                    await _hardwareRejectionTracker.RefreshVerdictCalibrationAlertAsync(partition, alert);
+                    refreshed++;
+                }
+            }
+
+            return (fired, refreshed, rearmed);
+        }
+
+        private static string AlertKey(string kind, string path, string status) => $"{kind}|{path}|{status}";
+
+        internal static VerdictCalibrationAlert BuildVerdictAlert(
+            string partition, VerdictCalibrationFinding finding, RuleRegressionDimension? dimension, DateTime firstNotifiedAt) => new()
+        {
+            TenantId = partition,
+            Kind = finding.Kind,
+            VerdictPath = finding.VerdictPath,
+            Status = finding.Status,
+            WindowHitCount = finding.WindowHitCount,
+            WindowSessionCount = finding.WindowSessionCount,
+            BaselineHitCount = finding.BaselineHitCount,
+            BaselineSessionCount = finding.BaselineSessionCount,
+            WindowRatePct = finding.WindowRatePct,
+            BaselineRatePct = finding.BaselineRatePct,
+            Lift = finding.Lift,
+            WindowStartDate = finding.WindowStartDate,
+            WindowEndDate = finding.WindowEndDate,
+            Dimension = dimension,
+            FirstNotifiedAt = firstNotifiedAt,
+            LastEvaluatedAt = DateTime.UtcNow,
+        };
+
+        /// <summary>
+        /// On-fire dimension correlation for a per-path finding: the window's sessions whose
+        /// (derived) verdict path matches vs all window sessions. Group kinds and the "global"
+        /// partition (a cross-tenant scan would dwarf the signal) carry no dimension claim.
+        /// Fail-soft: any error yields null — never a guessed dimension.
+        /// </summary>
+        private async Task<RuleRegressionDimension?> TryCorrelateVerdictDimensionsAsync(string partition, VerdictCalibrationFinding finding)
+        {
+            if (partition == "global" || finding.Kind != VerdictCalibrationAlertKinds.ShareRegression)
+                return null;
+            try
+            {
+                var windowStart = DateTime.ParseExact(finding.WindowStartDate, "yyyy-MM-dd", CultureInfo.InvariantCulture,
+                    DateTimeStyles.AssumeUniversal | DateTimeStyles.AdjustToUniversal);
+                var windowEndExclusive = DateTime.ParseExact(finding.WindowEndDate, "yyyy-MM-dd", CultureInfo.InvariantCulture,
+                    DateTimeStyles.AssumeUniversal | DateTimeStyles.AdjustToUniversal).AddDays(1);
+
+                var allSessions = await _maintenanceRepo.GetSessionsByDateRangeAsync(windowStart, windowEndExclusive, partition);
+                if (allSessions.Count == 0) return null;
+
+                var hitSessions = allSessions
+                    .Where(s => s.Status.ToString() == finding.Status
+                                && VerdictPathDerivation.Derive(s).Path == finding.VerdictPath)
+                    .ToList();
+                return RuleRegressionRadar.ComputeDimensionConcentration(hitSessions, allSessions);
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "Verdict-calibration dimension correlation failed for {Path} in {Partition}", finding.VerdictPath, partition);
+                return null;
+            }
+        }
+    }
+}

src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.cs

@@ -69,6 +69,7 @@ namespace AutopilotMonitor.Functions.Services
         private readonly IConfiguration _configuration;
         private readonly IHttpClientFactory _httpClientFactory;
         private readonly PreviewWhitelistService _previewWhitelistService;
+        private readonly IStorageInitializer _storageInitializer;
         private readonly ILogger<MaintenanceService> _logger;
 
         private const string PlatformStatsAliasFileName = "platform-stats.json";
@@ -101,6 +102,7 @@ namespace AutopilotMonitor.Functions.Services
             IConfiguration configuration,
             IHttpClientFactory httpClientFactory,
             PreviewWhitelistService previewWhitelistService,
+            IStorageInitializer storageInitializer,
             ILogger<MaintenanceService> logger)
         {
             _maintenanceRepo = maintenanceRepo;
@@ -128,9 +130,22 @@ namespace AutopilotMonitor.Functions.Services
             _configuration = configuration;
             _httpClientFactory = httpClientFactory;
             _previewWhitelistService = previewWhitelistService;
+            _storageInitializer = storageInitializer;
             _logger = logger;
         }
 
+        private async Task EnsureAllTablesAsync()
+        {
+            try
+            {
+                await _storageInitializer.EnsureAllAsync();
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "Daily table existence pass failed");
+            }
+        }
+
         /// <summary>
         /// Runs all maintenance tasks (used by the daily timer trigger)
         /// </summary>
@@ -141,6 +156,10 @@ namespace AutopilotMonitor.Functions.Services
 
             try
             {
+                // Startup only point-reads the table schema sentinel (see
+                // TableStorageService.InitializeTablesAsync); this daily full pass is what
+                // recreates a table that was deleted out-of-band. Cheap (idempotent creates).
+                await EnsureAllTablesAsync();
                 await MarkStalledSessionsAsTimedOutAsync();
                 await DetectExcessiveEventSessionsAsync();
                 await AggregateMetricsWithCatchUpAsync();
@@ -151,13 +170,24 @@ namespace AutopilotMonitor.Functions.Services
                 // App-version duration regression radar: same episode/tracker pattern over the
                 // install summaries (trailing 35d horizon loaded internally). Fail-soft.
                 await RunAppVersionRegressionRadarAsync();
+                // ONE projected cross-tenant window scan feeds every rolling sweep below (the
+                // StartedAt-only filter is a full-table drain in Table Storage — four of them per
+                // tick was four times the same read). Each sweep slices its own window.
+                var sweepWindow = await LoadSweepWindowSessionsAsync();
                 // F1 PR2: rolling 30d breakdown backfill + daily attribution aggregates. Owns
                 // its own window (NOT the snapshot-gated catch-up above) so late-terminating
                 // sessions still reach their StartedAt-date's aggregate. Fail-soft internally.
-                await SweepTimeAttributionAsync();
+                await SweepTimeAttributionAsync(sweepWindow);
                 // F2 PR4: device-history chain heal (incl. deleted-session ref cleanup) + daily
                 // FTR aggregates over the same rolling window. Fail-soft internally.
-                await SweepDeviceJourneysAsync();
+                await SweepDeviceJourneysAsync(sweepWindow);
+                // Verdict calibration: per-verdict-path daily buckets over the same rolling
+                // window, AFTER the device-journey sweep so the re-enrollment proxy reads
+                // freshly merged chains. Fail-soft internally.
+                await SweepVerdictCalibrationAsync(sweepWindow);
+                // Verdict-calibration drift radar over the rows the sweep just refreshed; anchored
+                // on yesterday like the rule radar (whole days only). Fail-soft.
+                await RunVerdictCalibrationRadarAsync(DateTime.UtcNow.Date.AddDays(-1), sweepWindow);
                 // Plan §5 PR6 / §16 R14: session retention fanout extracted out of the 2h timer
                 // into the dedicated 12h SessionDeletionMaintenanceFunction so cascade-lifecycle
                 // work has independent cadence + kill-switch + OpsEvent watchdogs. The non-session
@@ -199,6 +229,33 @@ namespace AutopilotMonitor.Functions.Services
             }
         }
 
+        /// <summary>
+        /// Shared sweep window: the longest horizon any rolling sweep needs — the calibration
+        /// radar's 7d window + 28d baseline anchored on YESTERDAY (= 35 days back from today) —
+        /// drained once per tick with the maintenance projection. Sweeps with a shorter window
+        /// slice it by StartedAt (<see cref="SliceSweepWindow"/>). Fail-soft: an empty list makes
+        /// every sweep a no-op rather than failing the tick.
+        /// </summary>
+        internal const int SweepWindowDays = Helpers.VerdictCalibrationRadar.WindowDays + Helpers.VerdictCalibrationRadar.BaselineDays;
+
+        private async Task<List<SessionSummary>> LoadSweepWindowSessionsAsync()
+        {
+            try
+            {
+                var today = DateTime.UtcNow.Date;
+                return await _maintenanceRepo.GetMaintenanceWindowSessionsAsync(today.AddDays(-SweepWindowDays), today.AddDays(1));
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex, "Sweep window load failed — rolling sweeps skip this tick (non-fatal)");
+                return new List<SessionSummary>();
+            }
+        }
+
+        /// <summary>Sessions of the shared window that started at or after <paramref name="windowStart"/> (UTC).</summary>
+        internal static List<SessionSummary> SliceSweepWindow(IReadOnlyList<SessionSummary> window, DateTime windowStart)
+            => window.Where(s => s.StartedAt >= windowStart).ToList();
+
         /// <summary>
         /// Manually triggered maintenance with flexible date selection
         /// </summary>
@@ -230,8 +287,11 @@ namespace AutopilotMonitor.Functions.Services
                 // Timer-path parity: manual maintenance also refreshes the attribution
                 // breakdowns + daily aggregates and the device-history/FTR rollups
                 // (rolling 30d windows, cheap once converged).
-                await SweepTimeAttributionAsync();
-                await SweepDeviceJourneysAsync();
+                var sweepWindow = await LoadSweepWindowSessionsAsync();
+                await SweepTimeAttributionAsync(sweepWindow);
+                await SweepDeviceJourneysAsync(sweepWindow);
+                await SweepVerdictCalibrationAsync(sweepWindow);
+                await RunVerdictCalibrationRadarAsync(dateToAggregate, sweepWindow);
 
                 if (!aggregateOnly)
                 {
@@ -350,6 +410,7 @@ namespace AutopilotMonitor.Functions.Services
 
                         int silentMarked = 0;
                         int silentAwaitingUser = 0;
+                        int silentSelfDeployingReconciled = 0;
                         foreach (var silent in silentSessions)
                         {
                             // The query is a permissive two-frame pre-filter; the decision is made
@@ -374,30 +435,44 @@ namespace AutopilotMonitor.Functions.Services
                                 continue;
                             }
 
+                            // Self-deploying profile gate (kiosk tenant aebdce78, 2026-08-23): a
+                            // silent agent after Device ESP all-succeeded is a finished device
+                            // (rebooted into the kiosk autologon / boxed), not a stall and never
+                            // "awaiting user" — reconcile to Succeeded right here instead of
+                            // routing it through Stalled → AwaitingUser → Incomplete.
+                            if (silent.IsSelfDeployingProfile
+                                && await TryReconcileSelfDeployingAsync(silent, lastContactAt, now))
+                            {
+                                silentSelfDeployingReconciled++;
+                                continue;
+                            }
+
                             var silentMinutes = (int)(now - lastContactAt).TotalMinutes;
                             await _sessionRepo.UpdateSessionStatusAsync(
                                 silent.TenantId,
                                 silent.SessionId,
                                 SessionStatus.Stalled,
+                                VerdictPaths.SweepStalled,
                                 stalledAt: now,
                                 failureReason: $"Agent silent for {silentMinutes}min (detected by maintenance sweep)");
                             silentMarked++;
                         }
 
-                        if (silentMarked > 0 || silentAwaitingUser > 0)
+                        if (silentMarked > 0 || silentAwaitingUser > 0 || silentSelfDeployingReconciled > 0)
                         {
                             totalSessionsMarkedStalled += silentMarked;
-                            _logger.LogInformation($"Tenant {tenantId}: Marked {silentMarked} agent-silent sessions as Stalled, {silentAwaitingUser} as AwaitingUser (WhiteGlove Part 2; silence threshold: {agentSilenceHours}h)");
+                            _logger.LogInformation($"Tenant {tenantId}: Marked {silentMarked} agent-silent sessions as Stalled, {silentAwaitingUser} as AwaitingUser (WhiteGlove Part 2), {silentSelfDeployingReconciled} reconciled to Succeeded (self-deploying; silence threshold: {agentSilenceHours}h)");
                             await _maintenanceRepo.LogAuditEntryAsync(
                                 tenantId,
                                 "SessionStalled",
                                 "Session",
-                                $"{silentMarked + silentAwaitingUser} sessions",
+                                $"{silentMarked + silentAwaitingUser + silentSelfDeployingReconciled} sessions",
                                 "System.Maintenance",
                                 new Dictionary<string, string>
                                 {
                                     { "SessionsMarkedStalled", silentMarked.ToString() },
                                     { "SessionsAwaitingUser", silentAwaitingUser.ToString() },
+                                    { "SessionsSelfDeployingReconciled", silentSelfDeployingReconciled.ToString() },
                                     { "AgentSilenceHours", agentSilenceHours.ToString() },
                                     { "SilenceCutoff", silenceCutoff.ToString("yyyy-MM-ddTHH:mm:ss") }
                                 });
@@ -468,9 +543,10 @@ namespace AutopilotMonitor.Functions.Services
                             // Device Setup done + within grace → AwaitingUser, else Incomplete; silent before
                             // Device Setup → Incomplete. Never Failed without an explicit failure signal.
                             var rollup = EnrollmentTimeoutClassifier.ExtractRollup(sessionEvents);
-                            var (targetStatus, reason) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
+                            var (targetStatus, reason, rule) = EnrollmentTimeoutClassifier.ClassifyTimedOutSession(
                                 rollup, effectiveStart, now, graceHours, session.LastEventAt,
-                                isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt);
+                                isPreProvisioned: session.IsPreProvisioned, resumedAt: session.ResumedAt,
+                                isSelfDeployingProfile: session.IsSelfDeployingProfile);
 
                             // No-op if the verdict equals the current (non-terminal) state.
                             if (targetStatus == session.Status)
@@ -480,6 +556,7 @@ namespace AutopilotMonitor.Functions.Services
                                 session.TenantId,
                                 session.SessionId,
                                 targetStatus,
+                                VerdictPaths.Compose(VerdictPaths.OriginSweep, rule),
                                 failureReason: reason,
                                 failureSnapshotJson: snapshotJson
                             );
@@ -609,6 +686,7 @@ namespace AutopilotMonitor.Functions.Services
                     silent.TenantId,
                     silent.SessionId,
                     SessionStatus.AwaitingUser,
+                    VerdictPaths.SweepWhiteGloveAwaiting,
                     failureReason: EnrollmentTimeoutClassifier.WhiteGloveAwaitingUserReason(rollup));
                 if (transitioned)
                     _logger.LogInformation(
@@ -623,6 +701,44 @@ namespace AutopilotMonitor.Functions.Services
             }
         }
 
+        /// <summary>
+        /// Applies the self-deploying gate to one agent-silent candidate (the caller pre-filters on
+        /// <see cref="SessionSummary.IsSelfDeployingProfile"/> so only those pay the event read).
+        /// When <see cref="EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned"/> confirms Device
+        /// ESP all-succeeded with no explicit failure, reconciles the session to Succeeded with the
+        /// shared honest reason. Returns true when the write happened; false (fail-soft, also on
+        /// read errors) lets the caller fall through to the normal Stalled marker.
+        /// </summary>
+        private async Task<bool> TryReconcileSelfDeployingAsync(SessionSummary silent, DateTime lastContactAt, DateTime now)
+        {
+            try
+            {
+                var events = await _sessionRepo.GetSessionEventsAsync(
+                    silent.TenantId, silent.SessionId, maxResults: 1000);
+                var rollup = EnrollmentTimeoutClassifier.ExtractRollup(events);
+                if (!EnrollmentTimeoutClassifier.IsSelfDeployingProvisioned(rollup, silent.IsSelfDeployingProfile))
+                    return false;
+
+                var transitioned = await _sessionRepo.UpdateSessionStatusAsync(
+                    silent.TenantId,
+                    silent.SessionId,
+                    SessionStatus.Succeeded,
+                    VerdictPaths.SweepSelfDeployingReconcile,
+                    failureReason: EnrollmentTimeoutClassifier.SelfDeployingReconcileReason(
+                        silent.ResumedAt ?? silent.StartedAt, now, lastContactAt));
+                if (transitioned)
+                    _logger.LogInformation(
+                        $"Session {silent.SessionId}: self-deploying profile silent after Device Setup provisioned — reconciled to Succeeded instead of Stalled");
+                return transitioned;
+            }
+            catch (Exception ex)
+            {
+                _logger.LogWarning(ex,
+                    $"Self-deploying reconcile gate failed for session {silent.SessionId}; falling back to Stalled marker");
+                return false;
+            }
+        }
+
         /// <summary>
         /// Builds the server-authored <c>session_timeout</c> event injected into the stream when the
         /// maintenance sweep graduates a stalled session to terminal Failed. Static + pure so the field

src/Backend/AutopilotMonitor.Functions/Services/Offboarding/TenantOffboardingHandler.cs

@@ -83,6 +83,8 @@ namespace AutopilotMonitor.Functions.Services.Offboarding
             // whole-row deletion besides chain-emptying.
             Constants.TableNames.DeviceHistories,
             Constants.TableNames.DeviceJourneyAggregates,
+            // Verdict calibration daily buckets (PK=TenantId) — regenerable from Sessions, wiped with the tenant.
+            Constants.TableNames.VerdictCalibrationAggregates,
             // Pre-write config snapshots (PK=tenantId). Wiped in the bulk phase; the
             // TenantConfiguration row itself is deleted LAST via SafeWipe (a delete, not a
             // save), so the wipe cannot re-create a snapshot afterwards. The admin-config

src/Backend/AutopilotMonitor.Functions/Services/OpsEventService.cs

@@ -641,6 +641,40 @@ namespace AutopilotMonitor.Functions.Services
                     lift,
                 });
 
+        /// <summary>
+        /// Verdict-calibration drift radar (docs/backend/verdict-calibration.md): a verdict
+        /// path's share of sessions doubled over its 28d baseline, the silence share
+        /// (sweep+maxlife — agent went quiet, backend had to decide) doubled, or the pure
+        /// fallthrough rule r6 decides ≥20 % of classifier verdicts. Operator-only diagnostic —
+        /// there is deliberately NO tenant bell. Fired ONCE per (kind, path, status) episode
+        /// (tracker-deduped). Dual-registered in the web OPS_EVENT_TYPES catalog
+        /// (memory: feedback_ops_event_types_dual_register).
+        /// </summary>
+        public Task RecordVerdictCalibrationDriftAsync(
+            string tenantId, string kind, string verdictPath, string status,
+            int windowHitCount, int windowSessionCount, double windowRatePct,
+            int baselineHitCount, int baselineSessionCount, double baselineRatePct,
+            double? lift, string? dimensionSummary)
+            => WriteAsync(OpsEventCategory.Maintenance, "VerdictCalibrationDrift", OpsEventSeverity.Warning,
+                $"Verdict calibration [{kind}] {verdictPath}/{status}: {windowHitCount}/{windowSessionCount} ({windowRatePct}%) over 7d " +
+                $"vs {baselineRatePct}% baseline ({baselineHitCount}/{baselineSessionCount} over 28d)" +
+                (lift.HasValue ? $" — lift {lift.Value}x" : string.Empty) +
+                (string.IsNullOrEmpty(dimensionSummary) ? string.Empty : $". {dimensionSummary}"),
+                tenantId, "System.Maintenance",
+                new
+                {
+                    kind,
+                    verdictPath,
+                    status,
+                    windowHitCount,
+                    windowSessionCount,
+                    windowRatePct,
+                    baselineHitCount,
+                    baselineSessionCount,
+                    baselineRatePct,
+                    lift,
+                });
+
         // ── Agent ──────────────────────────────────────────────────────────────
 
         public Task RecordSessionTimeoutsAsync(string tenantId, int sessionCount, int timeoutHours)

src/Backend/AutopilotMonitor.Functions/Services/RateLimitService.cs

@@ -1,5 +1,4 @@
 using System;
-using System.Collections.Concurrent;
 using System.Collections.Generic;
 using System.Linq;
 using Microsoft.Extensions.Caching.Memory;
@@ -17,10 +16,37 @@ namespace AutopilotMonitor.Functions.Services
         private readonly ILogger<RateLimitService> _logger;
         private readonly TimeSpan _windowDuration;
 
-        // Dedicated lock objects per device to avoid race conditions when cache evicts the request history list.
-        // Using the list itself as a lock target is unsafe because cache eviction can create a new list instance,
-        // causing different threads to lock on different objects.
-        private readonly ConcurrentDictionary<string, object> _locks = new();
+        // The request history and its lock travel together in one cache entry. Whoever observes the
+        // same list observes the same lock, so two threads can never mutate one list under different
+        // locks. A thread holding an already-evicted bucket only mutates a discarded list, which is
+        // harmless at the expiry boundary. A separate per-key lock dictionary is deliberately NOT
+        // used: its eviction-callback cleanup raced with lock re-creation and leaked on replaced entries.
+        private sealed class RequestBucket
+        {
+            public readonly List<DateTime> History = new();
+            public readonly object Sync = new();
+        }
+
+        // IMemoryCache.GetOrCreate is not atomic: two first-time callers can each create a bucket and
+        // the loser would count its request into a discarded list. Creation is serialized here
+        // (cheap: only cache misses take it); the hot path is a lock-free TryGetValue.
+        private readonly object _createLock = new();
+
+        private RequestBucket GetOrCreateBucket(string cacheKey)
+        {
+            if (_cache.TryGetValue(cacheKey, out RequestBucket? existing) && existing != null)
+                return existing;
+
+            lock (_createLock)
+            {
+                return _cache.GetOrCreate(cacheKey, entry =>
+                {
+                    entry.SlidingExpiration = _windowDuration;
+                    entry.AbsoluteExpirationRelativeToNow = TimeSpan.FromHours(1);
+                    return new RequestBucket();
+                })!;
+            }
+        }
 
         public RateLimitService(IMemoryCache cache, ILogger<RateLimitService> logger)
         {
@@ -56,18 +82,14 @@ namespace AutopilotMonitor.Functions.Services
             var cacheKey = $"ratelimit_{deviceThumbprint}";
             var now = DateTime.UtcNow;
 
-            // Get or create request history for this device
-            var lockObj = _locks.GetOrAdd(cacheKey, _ => new object());
-            var requestHistory = _cache.GetOrCreate(cacheKey, entry =>
-            {
-                entry.SlidingExpiration = _windowDuration;
-                entry.AbsoluteExpirationRelativeToNow = TimeSpan.FromHours(1);
-                entry.RegisterPostEvictionCallback((key, _, _, _) => _locks.TryRemove((string)key, out _));
-                return new List<DateTime>();
-            })!;
+            // Get or create request history for this device. The bucket is mutated in place; the
+            // cache lookup itself refreshes the sliding expiration, so no re-Set is needed.
+            var bucket = GetOrCreateBucket(cacheKey);
 
-            lock (lockObj)
+            lock (bucket.Sync)
             {
+                var requestHistory = bucket.History;
+
                 // Remove requests outside the sliding window
                 var windowStart = now.Subtract(_windowDuration);
                 requestHistory.RemoveAll(timestamp => timestamp < windowStart);
@@ -104,13 +126,6 @@ namespace AutopilotMonitor.Functions.Services
                 // Add current request to history
                 requestHistory.Add(now);
 
-                // Update cache
-                _cache.Set(cacheKey, requestHistory, new MemoryCacheEntryOptions
-                {
-                    SlidingExpiration = _windowDuration,
-                    AbsoluteExpirationRelativeToNow = TimeSpan.FromHours(1)
-                });
-
                 return new RateLimitResult
                 {
                     IsAllowed = true,

src/Backend/AutopilotMonitor.Functions/Services/RuleEngine.cs

@@ -311,7 +311,7 @@ namespace AutopilotMonitor.Functions.Services
                 // DurationSeconds. Letting UpdateSessionStatusAsync fall back to LastEventAt
                 // anchors duration on when the session actually went silent.
                 await _sessionRepo.UpdateSessionStatusAsync(
-                    tenantId, sessionId, SessionStatus.Failed,
+                    tenantId, sessionId, SessionStatus.Failed, VerdictPaths.Compose(VerdictPaths.OriginRule, rule.RuleId),
                     failureReason: failureReason,
                     failureSource: failureSource);