A platform script that ran 10 s was shown as "1h 00m" (session 4377911b): its
start line was self-anchored (+120 min) while its result line, re-read after an
overwrite rewind and therefore never fresh, fell back to the agent's cached zone
(+60 min), was rejected as ahead of the clock, and the duration was taken from
the raw pair. The same shape fired script_timeout_suspected daily in one tenant
on 2-9 s scripts read as 17 h. Fleet 2026-09-20..23: about 1 % of script
durations, 28 of 28 inflated ones carrying derivedTimestamp.
Agent: ScriptExecutionState binds the provenance of its start, result and exit
lines (CmTraceLineProvenance, persisted); EmitScriptCompleted judges the pair -
verified (both measured, or both assumed with the same offset from the same
file), unverified, or suppressed with durationSuppressedReason
(rejected-endpoint | offset-grid, CmTraceOffsetCalibrator.IsOnOffsetGrid) - and
always emits durationProvenance plus startSourceOffset*. A held completion's
sourceOffset* tags now describe its own result line; script_timeout_suspected
only judges a measurable duration; the emitted-run marker is capped at the
clock so a future-shifted result no longer swallows the next run's start.
Web: the reducer applies the same rule to stored events from older agents,
pairing a final with the latest script_started of its policy, and the panel
shows "n/a" with the explanation where the run time would be.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
ImeRegistryAppStateObserver reads a fourth IME registry surface,
Win32Apps\OperationalState\<userGuid>\<appGuid>: the install deadline a
downloaded app waits for (ExecutionDeadlineTime, IME 1.106) and the
app-in-use deferral counters (DeferralsUsed, DeferUntilTime,
DeferralMaxDeferrals, DeferralAutoDeferred, IME 1.105). Both travel on the
existing registry_app_state event as data (installDeadlineUtc, deferralsUsed,
deferralMaxDeferrals, deferUntilUtc, deferralAutoDeferred) with the change
triggers installDeadline / deferralsUsed / deferUntil, and the message says
the state in words ("downloaded, install deadline ... UTC", "deferred by user
1/3 until ... UTC", "auto-deferred ..."). No new event type.
ClassifyStatusServiceStatus maps IME's legacy AppInstallStatus explicitly:
1000/1002/1004 installed, 1001/1003 installing, 2000 band notApplicable,
3000 failed, anything else unjudged. The 1000 band was treated as installed,
but IME maps every InProgress enforcement state to 1001, so a running app
counted as terminal success on the registry rail (observability only).
Tests: explicit status mapping, both registry timestamp shapes, the new
change triggers, the wait-state text, and the tick flow for a deadline and a
deferral event. Mutation proof: classifier and trigger broken on purpose
turn exactly the five expected tests red.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
network_state_change and network_connectivity_check join the non-activity
list next to power_state_change. A flapping dock or access point kept the
idle-stopped periodic collectors running for the whole episode and reset
the stall clock of the enrollment it was stalling. The events are emitted
unchanged; the device-info network snapshots stay activity.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
IME re-runs remediation policies on its own schedule for as long as the
device is up, so an enrollment that waits for its user reported every run
of every policy. The tracker now emits a policy's first run and every run
whose result differs from the last reported one (compliance, exit code,
stderr, remediation status, error code); repeats are counted and surface
as one script_recurrence_summary per policy when the tracker stops.
The start signal of a known policy is held until its result is known,
because the running indicators pair a start with a later result; a start
without any result is still emitted after a grace. The counters live in
the tracker state file next to the file positions, so a hard kill rolls
both back together. Platform scripts are untouched. The summary does not
count as enrollment activity.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A process name is not a protected namespace: any user session can keep a
binary named like the agent resident, which made every agent start during
that session exit 1 before registration and let install mode report a
runtime launch that never happened.
The guard and install mode's launch verification now share one rule
(FindSiblingAgentPid): a name match counts only in session 0 and never for
the caller itself. Every runtime start lands there (Scheduled Task, WMI
launch from the SYSTEM install chain, self-update restart), so behaviour
for real agents is unchanged. The session id comes from the process
snapshot, so no handle is opened; a process that cannot be read does not
count, which fails open into two agents rather than none.
Two interactive --console runs in the same user session no longer block
each other; an interactive run next to the session-0 runtime still does.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A configured diagnostics entry is resolved once to folder, pattern and recursion, and
that target is what gets validated and enumerated; recursion asks the guard before
entering a subfolder, and the user-profile token is expanded before validation.
Security, PowerShell and Sysmon event logs are refused per file, by name and by
resolved channel, in every mode; the portal flags such a path before it is saved.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Test-VM session c3ecb568 (2.0.1460): the tracker loop was not scheduled for 26 s at 100 %
VM CPU while the rest of the agent kept running; five results were read before their
executor end blocks — held and completed correctly (D-247) — and nothing in the fleet
telemetry could have shown the pause (ime_backlog_bytes is an instantaneous value).
- ImeLogTracker.RunPollPassAsync hosts the loop body (file reads, the pending-completion
flush, the token-failure window) and times it on a monotonic clock (MonotonicMillisProvider,
Stopwatch in production): PassMaxMs = longest single pass (time INSIDE — reads, ledger
verification, regex), PassGapMaxMs = longest pause between the end of one pass and the
start of the next (the 100 ms delay, SaveState and any scheduling delay). Restart-safe
like the other health counters (TrackerHealthData); ImeTrackerHealth carries both.
- agent_metrics_snapshot: ime_pass_max_ms / ime_pass_gap_max_ms; ime_pattern_hits:
passMaxMs / passGapMaxMs.
- MCP session coverage: sessionTotals.passMaxMs / passGapMaxMs, and from 5 s on a note that
says the lines were read late, not lost — every event keeps its log line's timestamp; the
overwrite note and the pause note join when both apply.
- Tests: ImeLogTrackerPollTimingTests (maxima over three passes with a fake clock, fresh
tracker at zero, restart keeps and never lowers them); MCP coverage tests for the note
threshold. Agent suite 2941 green, solution build 0 warnings, MCP tsc clean.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Test-VM session 24dc69d1 (2.0.1459) had the overwrite geometry the result hold (D-247)
was built for, yet the hold never fired: "Powershell execution is done, exitCode = 0" —
written by IME's health-script worker (AgentCommon ScriptWorker) to both
IntuneManagementExtension.log and HealthScripts.log — was read as PS-SCRIPT-EXITCODE and
attributed to the pending platform script c81b8053 1.2 s before its executor even started
(IME-log marker: the last PS-SCRIPT-GENERATED; HealthScripts.log: the no-marker fallback to
the last platform start). The completion went out at the result line with the foreign exit
code, and the run's own end block, recovered 150 ms later by the rewind, was dropped as
already emitted. Both codes were 0 by luck; a failing health script would have turned a
succeeded platform script into script_failed. The worker's launch line flipped the pending
script's context the same way (338d9434, twice in one run).
- rules: PS-SCRIPT-EXITCODE deleted — the line is health-script-only (3 hits = 3 cycles,
platform scripts never produce it), HS-EXITCODE already carries the value, and a line that
lands inside a parallel platform run's window cannot be attributed by position.
HS-INVOCATION (new): the worker's command line `"…\agentexecutor.exe" -remediationScript`
(identical from IME 1.104.102 to 1.105.152) as agentInvocationArgument of a non-powershell
kind — the launch line it writes next stays with that invocation. Both take effect on the
next reseed for every agent from 2.0.1458 on.
- tracker: a file without markers resolves to the platform run in flight only when a
platform run has ever opened in that file (IntuneManagementExtension.log after a rollover,
AgentExecutor.log); HealthScripts.log and the app logs never qualify. Test seams:
ProcessLogMessageForTest takes the source file, ClearInvocationMarkersForTest.
- tests: ImeLogTrackerPlatformScriptContaminationTests +5 with the session's real lines —
the worker's exit line pre-fills nothing, its launch line flips no context, HealthScripts.log
lines resolve to no platform script, the context line after an IME-log rollover still finds
the run in flight, and no shipped pattern reads the worker's exit line.
- rules/dist regenerated (node rules/scripts/combine.js); analyze-rules.json catches up with
ANALYZE-OFFICE-001 1.1.1 (6657c6b9 was committed without its dist).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The first agent with the ledger (2.0.1458, session e7f3c910) lost nothing — 16/16 results,
20/20 end blocks, 20 rewinds — yet 3 of 16 completions carried no exit code and one policy
got an agentexecutor_fallback duplicate. The IME result line reaches the tracker before the
AgentExecutor.log end block of the same run: always when the block was overwritten (the check
the result triggers recovers it one pass later), often without any overwrite (the
AgentExecutor.log read of a pass precedes the IME log read by tens of milliseconds while the
exit line and the result line are 8–30 ms apart), and whenever the poll loop stalls (2 s at
100 % VM CPU). Emitting at the result line dropped every recovered end block, and the late
start line reopened the slot, cleared the emitted-marker and the fallback emitted the same
run again 15 s later.
- CompletePlatformScriptFromImeResult (one core for the handler and the test seam) emits at
the result line only when the slot has its exit code; otherwise the result stays on the
slot (ResultObservedAtUtc, ResultPatternId, ResultHeldSinceUtc) and
FlushPendingPlatformScriptResults — the former FlushStalePlatformScriptResults, still the
home of the 15 s exit-code fallback — emits it at the end of the pass in which the exit
code arrived (the block's output lines come with it), at shutdown, or after
PlatformScriptEndBlockGrace (5 s, wall clock).
- The emitted-marker is a Dictionary keyed by policy id with the run's source timestamp
(state file: PlatformScriptResultEmittedAt; the old list is read as MinValue). A start
line strictly older than the emitted result is the late start line of that run: it owns
the lines behind it in its file, opens no slot, emits no script_started. A start line at
or after a held result closes the hold as the next run; a start line that meets a slot
the result opened stamps StartedAtUtc.
- The adapter binds the completion to the result line (ResultObservedAtUtc, patternId =
ResultPatternId) the way the fallback binds to the exit line, so a held emit keeps its
timestamp, pattern and duration.
- Exit-code lines whose owner has no pending slot any more log why they are dropped;
ApplyExitCode is shared by the handler and the test seam.
- Tests: ImeLogTrackerResultHoldTests (11) and ImeLogTrackerAdapterResultHoldTests (1); six
existing cases that encoded the immediate emit without an exit code now flush explicitly.
Agent suite 2933 green, solution build 0 warnings.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
IME's trace listener opens every log once per process with FileMode.Append and FileShare.ReadWrite
and writes at its own stream position afterwards. AgentExecutor.exe (two processes per user-context
script, one per detection script) has such a listener on AgentExecutor.log and on
IntuneManagementExtension.log, so a process that resumes logging after another one appended
overwrites those bytes: the final file holds the last writer, a byte-position tailer has read the
first version. Session 46749560 lost 6 of 16 platform-script results and 4 of 16 executor end
blocks with every tracker health counter at zero — results fell back to the exit code (one Success
where IME said Failed), one finished script stayed Incomplete. The agentexecutor_fallback results
in the fleet are this race, not late IME logging.
The tracker now keeps a fingerprint ledger of every entry it processed from the two multi-writer
files (FNV-1a 64 over the decoded bytes, one CmTraceEntryAssembler shared by the read loop and the
verification so entry boundaries are identical by construction), verifies it on every script end
signal, on a byte-budgeted cadence while an invocation is in flight (256 KB/s, at least 1 s
apart), in a 4 KB guard zone on every growth pass and on a fragment at the bookmark, and rewinds
to the first changed entry; entries whose (offset, hash) still read unchanged are skipped,
everything else is processed as backlog (never fresh, never a calibration anchor). The
policyId-less executor lines are owned by the nearest preceding invocation marker in their file
(PS-AGENT-ARG opens, PS-AGENT-SCRIPT-START upgrades to the platform script, PS-AGENT-INVOCATION and
PS-AGENT-COMPLETED close) instead of a last-started pointer, so an end block a rewind uncovers
reaches its own script. Restart seeds the ledger from the 64 KB behind the persisted bookmark;
a rollover clears it. Four restart-safe counters (ime_overwrite_rewinds,
ime_overwrite_bytes_reprocessed, ime_verify_passes, ime_verified_bytes) ride agent_metrics_snapshot
and ime_pattern_hits — recovered data, not skipped work, no new event type.
Tests: ImeLogTrackerOverwriteRewindTests (15) drive the positional-overwrite write pattern against
both files, including the two field scenarios with the shipped patterns; agent suite 2921 green,
backend known-actions list extended for the two new patterns.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Revises 14b185db after the pre-release review. The fleet measurement (30 days,
all tenants) showed 87 of 821 sessions with a sleep episode had their verdict
inside the episode — 59 of them one tenant's self-deploying devices closed two
minutes before the device-only detection came due and reopened 5–7 days later.
Stamping the firing clock would have turned those outages into session
duration and pushed the tenant's mean duration from minutes to days.
- OnDeadlineFired posts OccurredAtUtc = DueAtUtc again; the due time stays in the
payload and the two stale-fire guards keep reading it (any stamp stays safe).
- Network Timeline: a timeout verdict (helloOutcome timeout, device-only detection
window) is drawn as a hollow marker "Completed (timeout)" with a legend line and a
plain-language tooltip — a wait expired, nothing observed on the device; the
Asleep block is never split at it. finalizing_grace alone is not an indicator.
- Session Info standby note clips to the two WhiteGlove spans when the card shows
the combined WhiteGlove duration; the dedup key is consumed only by a usable
observation (review findings).
- Stale comments describing the firing-clock contract corrected.
Tests: DeadlineFireTimeStampTests (reducer stays correct for any stamp),
EnrollmentOrchestratorTests (due-time stamp + payload), networkTimelineModel
timeout markers (3), standby windows, WhiteGlove windows.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A HelloSafety deadline due during Modern Standby fired after wake but was stamped
with its due time, so enrollment_complete and CompletedAt landed inside the sleep
window: the Network Timeline drew "Completed" in the middle of the Asleep block
and three views named three different standby durations.
- OnDeadlineFired posts OccurredAtUtc = FiredAtUtc and carries the due time under
SignalPayloadKeys.DeadlineDueAtUtc; the stale-fire guards (RealmJoin re-arm,
SelfDeploying guard C) identify the incarnation via DecisionEngine.DeadlineDueAtUtc,
which falls back to the signal time for pre-key signal logs.
- SignalSerializer renders Date-tokenized payload strings as ISO-8601 round-trip text
instead of the culture-formatted JValue.ToString() (a replayed deadlineDueAtUtc came
back as "4/20/2026 9:55:00 AM").
- Web: the Session Info standby note is clipped to the same window as the duration
next to it (lib/standby.ts) and printed with seconds, matching the attribution chip.
Tests: DeadlineFireTimeStampTests (late HelloSafety fire, guard C, RealmJoin re-arm,
helper fallback), SerializationTests date-like payload round-trip,
EnrollmentOrchestratorTests late fire, lib/__tests__/standby.test.ts.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
ANALYZE-ESP-004 fired on every ESP terminal failure of a Continue-anyway
profile and rendered the configured FirstSync limit as elapsed time — the
mayHaveContinuedAnyway flag only says the profile has the button. The rule
now requires the agent's own verdict (app_install_failed with
failureType=esp_apps_timeout), names that app via a same-event filter, reads
failedSubcategory from esp_failure_settle_started and words the limit as a
limit. Verified with test_analyze_rule: cb7036a6 fires, 683f1eff (0x80070652
after 19 min) stays silent.
Rule engine: not_exists with a dataField on event_type/event_data
conditions matches when no event of the type carries a non-empty value —
including an absent type — instead of being dead behind the fieldValue null
short-circuit. Schema and MCP authoring guide say so.
Agent: a Shell-Core 62407 failure that lands inside the provisioning
tracker's 30-s settle window (a third of all ESP terminal failures) now
inherits the pending errorCode / failedSubcategory / category /
likelyCulpritApps, so enrollment_failed, Sessions.FailureReason and the
termination-time app classification see the HRESULT the registry had
already named. failureType stays the Shell-Core value.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Gather rules and admin-configured diagnostics paths could never reach a
folder below a user profile. The %LOGGED_ON_USER_PROFILE% token lifts the
C:\Users hard block for AppData\Local and AppData\Roaming, but lifting is
not admitting: the path then fell into filePrefixes, which is matched
literally against the already-expanded path, and the profile name is not
known when that list is authored. The exception only ever worked in
unrestricted mode, and no agent test passed a non-null userProfilePath.
D-228 introduces userProfileFilePrefixes, holding entries relative to the
profile root (AppData\Local\RealmJoin). Both guards resolve them against
the detected profile and match segment-bounded; entries that are rooted,
contain "..", or sit outside AppData\Local / AppData\Roaming are dropped at
load, so configuration can only narrow the hard block's exception, never
widen it. The allowlist is anchored to the detected profile, so a rule
cannot reach another user's AppData. The AppData subdirectory list existed
in both guard classes with the GatherRuleGuards copy unused; it is now one
list and DiagnosticsPathGuards delegates. filePrefixes additionally gains
C:\ProgramData\RealmJoin\choco\logs.
D-229 fixes the pre-flight validators, which must only ever reject what the
agent rejects:
- the portal returned "allowed" on the lifted hard block without consulting
the allowlist, so a rule the agent refuses passed the enable gate and then
failed silently on the device with a security_warning;
- the MCP validator offered logparser the diagnosticsPathPrefixes list that
no collector consults (four paths reported allowed that the agent blocks);
- the MCP validator did not expand environment variables at all, reporting
%ProgramData% targets as blocked although the device reads them.
Measured before changing the portal: no rule in any tenant used the token,
so nothing that was working gets disabled.
Verified: solution build 0 warnings, agent 2897 tests, web 1257 + tsc +
eslint 0, MCP 779, guardrails.json validates against its schema, rules/dist
regenerates without drift. The six new agent tests were green against the
unchanged guard first, which is what proves the exception was inert.
Takes effect on devices only after an agent release: guardrails.json is an
EmbeddedResource in the agent assembly.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The portal showed the two-stage bootstrap output ending in ...[truncated]. That marker
comes from the agent, not from the web view: the stdout is cut at capture time, so no
amount of UI space brings the rest back.
Measured over 1,486 script outputs in the fleet (2026-08-25..09-08): median 39, p90 644,
max 1,501 chars, and the 2,048 cap had never been hit -- the maximum was our own
one-stage bootstrap. The loader preamble of ~460 chars pushed exactly that output over
the edge, i.e. this is a consequence of D-224 and it hit the one output an operator
actually reads.
8,192 covers the realistic worst case (download retries plus relax logging, ~2,700) and
stays 4x below the 32K-char Table Storage property limit that bounds the event
downstream. Other scripts are unaffected in practice: p90 is 644 chars.
Reaches the fleet with the next agent release.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Invoke-BootstrapDownload.ps1 and Start-AutopilotMonitor.ps1 had become the same script:
download the current installer, run it, log, always exit 0. The MSI copy additionally
downloaded and executed WITHOUT verifying what it got, which is exactly the gap the loader
was written to close. The MSI is now packaging only -- it installs and runs the same file
that is assigned directly in Intune, and both delivery channels get publisher verification
from one implementation.
Dropped from the MSI path: its own Deployed-marker check. That guard lives in stage 2 and
runs there anyway, so the only cost is one download on a device that is already deployed.
Kept: the log file name. bootstrap-msi.log is what the diagnostics tooling keys on to tell
the MDM-LOB channel apart from the Intune script channel, so the loader takes -LogFileName
(default bootstrap-loader.log) and the custom action passes bootstrap-msi.log. A Pester
test now pins that contract across the two files -- renaming the parameter on either side
would leave the MSI channel logging under the wrong name.
Package version 1.0.2 -> 1.1.0 so Intune LOB update detection sees the new payload, and a
new component GUID because the key path file changed.
Loader 1.0 -> 1.1. Bootstrap suite on Windows PowerShell 5.1: 44/44.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- IME-EXITCODE-CLASS captures the return-code class of the installer exit code ("lpExitCode is defined as <Success|SoftReboot|HardReboot|Retry|Failed>", verified against the IME string literal) as exitCodeClass on AppPackageState; new action captureExitCodeClass, persisted across the agent restart a reboot causes, last value wins across a Retry loop. IME-EXITCODE-MAPPED-FAILED keeps the error path; both match the same line.
- Adapter: exitCodeClass whenever known, exitCode on every terminal state (Installed/Postponed/Skipped/Error) instead of Error only.
- ANALYZE-APP-018 (warning, apps): app_install_completed.exitCodeClass in SoftReboot,HardReboot; +25 HardReboot, +15 observed system_reboot_detected; never fails the session.
- Web: completed-app exit badge amber (failed stays red), "mapped as SoftReboot/HardReboot/Retry" on timeline and Install Progress.
- Tests: pack end to end (phase, current app, class lines, negatives, restart round trip), adapter payloads, linearity samples, backend rule engine + known actions.
- internal gitlink: plan, D-214, backlog.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
MCP: coverage block (window, IME tracker, collectors, upload, diagnostics, gaps) built from the
health events the summary already fetches (session-coverage.ts, data.<key> slices); validate_rule
warns on a required not_exists of an IME-log-derived event without the ime_tracker_degraded
precondition; authoring guide 'absence'; instructions and prompts point at coverage.gaps.
Rules: guardrails.json imeLogDerivedEventTypes (schema + combine). Backend: catalog invariant test —
absence rules carry the coverage gate or a documented exception (ANALYZE-ID-004).
Agent: diagnostics_uploaded/_upload_failed carry packaging counters (truncated, includedFiles,
skippedFiles, skippedByReason, problemsByKind) — counts only, never paths.
Also: shared-manifests.json regenerated (was stale at HEAD; McpAccessPolicy doc string) + codegen.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
TelemetryItemKind moves to Shared; TelemetryItem.ToWire() maps onto TelemetryItemDto
and the uploader serialises the DTO, so both ends share one wire class. The spool line
keeps writing Kind by name (StringEnumConverter registered in TelemetryItemSerializer).
IngestTelemetryFunction.PartitionBatch routes by exact kind name and collects every
item it cannot route or parse; a non-empty rejection list answers the batch with 422 +
TelemetryItemsRejectedResponse (poison, rejectedRowKeys, reason) before anything is
persisted, logs a Warning and writes the new TelemetryItemsRejected ops event. The
agent's existing poison protocol drops exactly the named RowKeys and re-uploads the
rest; before this the ingest skipped the items, answered 200 and the agent cleared
its spool.
Contract tests from both sides: TelemetryWireContractTests (agent) freezes the real
batch body under tests/fixtures/telemetry-wire/ and parses the pinned poison body;
TelemetryWireFixtureTests (backend) dispatches every frozen body with zero rejections,
requires an arm per kind and pins the poison body.
Agent-facing responses implement IApiResponse and go through ResponseHelper, request
types carry [WireContract]; plain ingest refusals use the error envelope, the
register-session and upload-url refusals keep their DTO bodies (the agent reads
ErrorCode/Message). Manifests and generated wire types regenerated (also picks up
four types HEAD had left out of the manifest).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
hybrid_login_pending now measures the absence of a real-user desktop after a reboot
(cancelled by the DesktopArrivalDetector or a real JoinInfo user), is never armed on
WhiteGlove devices (resume marker, Part-2 flag, .part1- archive) and reports the
JoinInfo placeholder as a fact only — on a Hybrid join the AD sign-in never replaces it.
New EntraUserAffinityDetector: armed at the real-user desktop on Hybrid devices, the
first IME-TOKEN-SUCCESS line after the desktop posts ime_user_token_acquired; ten
minutes without one post entra_user_affinity_pending with the IME token-failure codes.
ImeLogTracker exposes raw per-line token callbacks past the historic-replay gate;
HelloTracker surfaces User Device Registration 304/305 as device_registration_event.
ANALYZE-ID-004 v2.0.0 keys on those events (event_type not_exists gained a branch),
GATHER-DEVICE-006 follows, GATHER-DEVICE-007/008 collect the Entra operational log.
Constants + MCP catalog gain the five types (D-072), the guard test learns the
multi-line OnTraceEvent shape, gather_* outputs pass the MCP filter gate by prefix.
CMTrace: ImeLogEraPreScan/CmTraceEraTable anchor the backlog era that holds the
bootstrap execution record against the Deployed install marker (0.13-0.36 s in the
field), whole era or nothing, sibling files by local time; origin era-anchored.
Backend phase_duration/app_install_duration refuse a reader-zone-fallback endpoint.
Decisions D-190, D-191. Session a7140f98.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Reverts the HasAccountSetupActivity tightening from f0a3f043: measured on 485 Classic
(SkipUser=false) sessions, 96 see a single Shell-Core 62407 before any AccountSetup
progress and 75 of them complete AccountSetup only through the user-apps-settled synthesis,
which needs that exit remembered as the edge. Windows confirmed categorySucceeded in 0/485,
so the synthesis is the load-bearing completion path; presence semantics stay and are
documented as such.
UserEspKeepAwakeHost instead gets an additive progress probe
(ProvisioningStatusTracker.HasAccountSetupProgress via EspAndHelloHost) and releases on an
EspExiting that arrives after the AccountSetup registry left notStarted — the real user-ESP
page exit (progress precedes it by >= 1 min fleet-wide; the Device->Account handoff exit
always precedes progress). AccountSetupProvisioningComplete no longer releases; desktop
arrival is deliberately not a trigger (explorer.exe ran before the final exit in all 385
two-exit sessions). Sessions with only the handoff exit hold until the safety cap. Release
reason user_esp_exited replaces account_setup_complete. D-183 rewritten.
Tests: host (release on exit+progress, handoff exit holds, completion no longer releases,
exit before engage ignored), tracker (presence stays true / progress false on the pre-written
all-notStarted JSON). Solution build 0 warnings; agent suite 2823 green.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Agent: ProvisioningStatusTracker.HasAccountSetupActivity now requires at least one
AccountSetup subcategory past notStarted. Windows writes AccountSetupCategory.Status with
every subcategory notStarted BEFORE the Device-ESP page exits, so the presence check
confirmed the Device->Account handoff (Shell-Core 62407) as the final ESP exit; the
user-apps-settled synthesis then completed AccountSetup and released the User-ESP
keep-awake minutes before the real exit (fleet: 77 % of keep-awake sessions, median 3 min;
on battery that is enough for Modern Standby). D-183.
Rule engine: filterField/filterOperator/filterValue now apply to event_data and
event_data_array as a same-event filter (shared ApplyValueFilter with event_count).
ANALYZE-DEV-012 v1.1.0 excludes backfilled pre-session episodes (backfilled equals false);
a second independent condition could not, since each condition scans all events on its
own. D-184. Schema, Shared model, MCP authoring guide and generated types follow.
Tests: ProvisioningStatusTrackerAccountSetupActivityTests (the verbatim all-notStarted
registry JSON), SleepEpisodeRuleTests (+3 backfill cases), RuleEngineEventDataFilterTests.
Solution build 0 warnings; agent suite 2820 and backend suite 5791 green.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- .gitmodules: docs -> internal, repo renamed to Autopilot-Monitor-internal; the OKF bundle now lives at internal/docs/
- .claude/CLAUDE.md rewritten: where knowledge lives, working-file rules, hard rules; imports the private half from internal/CLAUDE.md (silently absent without submodule access)
- .claude/architecture-principles.md moved into the internal bundle (reconciled against the code)
- tasks/ (gitignored) retired; working files live in internal/work/
- doc path references docs/<bundle> -> internal/docs/<bundle> in code comments, skills gitlink, codeql paths, build script
- rule descriptions no longer cite an internal doc path (customer-visible text); rules/dist regenerated
- public tree no longer names private operator tooling
- .claude/commands gitlink: skill docs point at the new paths
The fixture waited 5 s per posted signal and the scenarios 10 s for a
terminal stage. Those are deadlock guards, not performance assertions:
every applied step performs 5-6 WriteThrough + Flush(true) writes and 2-3
atomic renames synchronously on the ingress worker, and on a cold GitHub
runner volume the first scenarios after the parallel phase still blew the
5 s budget even though the collection is already serialised (run
33581374985: the first four scenarios in run order failed, the other nine
passed; the rerun of the same commit was green). One constant,
EnrollmentOrchestratorFixture.SettleTimeoutMs = 30 s, now covers the
per-signal, terminal-stage and uploader waits.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The V2 Core integration scenarios drive the real orchestrator against
wall-clock budgets (5 s per signal, 10 s for Completed / uploader). In
isolation every scenario finishes in under a second; inside the fully
parallel suite on a 4-core runner the same scenarios took 4-7 s locally
and blew the budgets on two slow GitHub runners (CI runs 33564658714 and
33565773888). Serialising them after the parallel phase, like
EnrollmentOrchestratorTests and RecoveryPathTests already are, brings
them back under 700 ms without changing the budgets.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Test code used the real test-tenant GUID, a real UPN, a real VM serial (and
its Intune-normalized form) and a real hostname as literals. None of the
assertions depended on the specific values; they are contoso/1111…/1234… now.
The one that stays is CertTenantBindingTests: its tenant GUID is stamped
inside device-cert-sample.pem and can only change together with the fixture.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The 2026-09-01 WG-seal soak on post-SignalIngress-fix builds (memory:
project_wg_seal_postfix_soak_residual) found sessions whose DecisionEngine
never sealed Part 1 while the agent kept running for hours - and telemetry
could not distinguish a frozen ingress worker from a deliberate Weak
sealing-classifier verdict, because the metrics snapshots carried no
pipeline counters at all.
Every agent_metrics_snapshot now carries the block. The fields are MONOTONIC
totals plus a monotonic queue peak - deliberately no momentary queue length:
a snapshot samples the integral, so any two snapshots prove whether the
worker ran in between (a frozen worker keeps signal_pending_count growing
while processed/applied stand still; a momentary gauge would alias between
ticks). Same field pattern the snapshot already uses for the spool
(spool_total_enqueued_count / spool_peak_pending_item_count).
- signal_last_ordinal / signal_processed_total (new ProcessedSignalCount) /
signal_pending_count / signal_reentrant_total: SignalIngress worker
progress; signal_queue_peak is a new lock-free Interlocked-max recorded in
Post (fast + slow path; worker-local follow-up items deliberately excluded)
- decision_stage + decision_last_applied_ordinal: from the immutable
DecisionState - answers the WG question directly ("minutes after the
tracker's whiteglove_complete and still not WhiteGloveSealed")
- wg_sealing_level / wg_sealing_score: only after the sealing classifier
produced at least one verdict (key ABSENCE, not Unknown/0 noise, for every
non-WG session) - separates "classifier said Weak" from "never ran"
Wiring is a read-only probe chain following the imeTrackerHealthProbe
pattern: EnrollmentOrchestrator passes () => _processor?.CurrentState into
IComponentFactory.CreateCollectorHosts (new optional decisionStateProbe;
test fakes updated), PeriodicCollectorLifecycleHost combines it with its
existing concrete-SignalIngress reference in BuildSignalPipelineProbe, and
AgentSelfMetricsCollector writes the fields fail-soft (a throwing probe
degrades to a snapshot without the block - pinned by test). The snapshot
event travels the collector/event path, which is independent of the signal
pipeline it measures - field-proven by the freeze cases, whose perf/metrics
events kept flowing while the engine stood still.
No backend/web/MCP change needed: no new event type, DataJson is free-form.
Tests: 5 new collector facts (full block, absence semantics x2, fail-soft,
no-probe compatibility) + 2 new SignalIngress facts (processed total stable
across drain; queue peak survives the queue draining - the property a
momentary gauge would not have). Agent suite 2812 green, solution builds
with 0 warnings.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The shutdown run allowed every logged-in user (explorer.exe owner) dynamically, so a backdoor account created after the startup scan and signed in before enrollment completed was scored 0. The WMI inventory is now read first and a logged-in name that is a local account is rejected for the account and the profile check (reported as logged_in_local_accounts); only Entra/domain identities — the only ones seen in real sessions — remain dynamically allowed for their profile folder.
Built-ins are recognised by RID 500/501/503/504 on the machine SID in addition to their English names: since disabled accounts are inventoried (88dfd331), localized Gast/Invité/Administrateur would otherwise flag every non-English device. kioskUser0 (Assigned Access autologon) joins the built-in list; the profile check skips junctions and $<32 hex> folders (cannot be an account name, pure payload noise). ANALYZE-ID-002 1.3.0 explanation mentions the new field.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The path guards, the enumeration's reparse-point skip and the per-file attribute check all judged path strings, while the FileStream feeding the ZIP opened that string later (after earlier files were compressed). Under any recursive source folder a local user can write to, a subdirectory turned into a junction in that window made SYSTEM copy the junction target into the tenant-uploaded package.
PinnedSourceFolder holds each section folder open for the whole section (final path must equal the validated path, 8.3-expanded; FILE_LIST_DIRECTORY without delete share, so NTFS refuses renaming it or any ancestor) and opens each candidate with FILE_FLAG_OPEN_REPARSE_POINT, requiring the handle's final path to be exactly <canonical folder>\<relative path>; the archive is fed from that same handle. Manifest: FOLDER REJECTED (...) / SKIPPED (resolved outside validated folder) / FAILED (open). The race is pinned by a test that swaps the directory from the BeforeSourceFileOpen seam; on the previous code the same test copies the junction target.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
EffectRunner posts ClassifierVerdictIssued from inside DecisionStepProcessor.ApplyStep,
i.e. on the SignalIngress worker thread. With the bounded channel (256) full, Post's
back-pressure slow path blocked the channel's only consumer as a producer: nothing could
ever drain, every collector Post blocked behind it, and the whole decision/monitoring
pipeline froze silently until the max-lifetime watchdog.
Post now detects a call from the worker thread itself and routes the item to a
worker-local follow-up queue that WorkerLoop drains (iteratively, ahead of the channel)
right after the current item. Same processing path — ordinals, SignalLog, emit,
reduce, ApplyStep — just never blocking. Worker posts are accepted during the Stop
drain too, since the verdict belongs to the step being finished. ReentrantPostCount
exposed for observability/tests.
Tests: full channel + worker post releases a blocked foreign producer and processes the
verdict before the next channel item (fails against the previous code by hanging);
follow-up chain drained in order without recursion; verdict posted during Stop drain is
not lost. Docs: agent/decision-engine.md, agent/overview.md, log.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Agent endpoints authenticated per tenant but authorized session writes on a caller-supplied
session id. Sessions rows now carry the creating device identity (cert thumbprint + Intune
device id from the CN, or bootstrap code + serial) in primary-only Owner* columns.
SessionOwnershipPolicy evaluates register / telemetry / error-report requests against it;
SessionOwnerBindingObserver records the outcome on the request row, as a warning and as a
throttled SessionOwnerMismatch ops event. Nothing is rejected in this stage.
Agent: errorCode parsing on 401/403, one-shot session rotation on session_owner_mismatch
without feeding the auth-failure tracker, spool drop, agent_started.sessionRotated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Oliver Kieselbach