AutoForge/prompts/Step-08-Report.md

303 lines
26 KiB
Markdown

```
# Step-08 - Recorded telemetry replay
## Status
COMPLETE. Implementation, controlled/rendered verification, the complete
nine-project Release build, all regressions and native saved-data replay
checks passed. Step-08 remains pending manual orchestrator review; this report
does not grant acceptance. Accepted Step-07/07A/07B is preserved. Step-09 was
not started. No agents, automatic handoff, staging, commit or push occurred.
## Implemented functions, files and interfaces
Application:
- [TelemetryReplayModels.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Replay/TelemetryReplayModels.cs):
separate RecordedReplay instance/capture context, cursor/count/state,
immutable original archive, typed command failures, recorded sample
observation/time context and a persistent actual fault reason.
- [TelemetryReplaySession.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Replay/TelemetryReplaySession.cs):
fresh-session factory, strict explicit load/list, atomic replacement,
per-load partial acknowledgement, definitions/rate validation, time/ordinal/
Next/Previous/First navigation, Play/Pause/Stop/Restart, shared cleanup and
asynchronous close. Its inputs are only ITelemetryArchiveStore,
TimeProvider and IReplayWaiter. It has no diagnostic/profile/model/save port.
- [ReplayPrefixProjection.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Replay/ReplayPrefixProjection.cs):
metadata-only positive-snapshot binding index and bounded exact-prefix
rebuilding with per-module current/last-Good/status caches and original
sample positions. It does not infer a catalogue, fill future values, sort
model time or modify archive objects.
- [DashboardGaugeProjection.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/LiveDashboard/DashboardGaugeProjection.cs):
pure shared definition/binding validation and numeric/quality/range/
last-Good projection extracted from the accepted live feature.
[LiveDashboardAcquisition.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/LiveDashboard/LiveDashboardAcquisition.cs)
consumes those helpers while retaining its actual session/acquisition owner.
UI:
- [TelemetryReplayPage.razor](../src/AutoForge.UI/Features/Vehicles/Telemetry/Replay/TelemetryReplayPage.razor)
owns /telemetry-replay, a fresh replay session, list/load, recorded provenance,
original completion, acknowledgement, all playback/navigation/rate controls,
current raw/decoded inspection and close/unmount drainage.
- [LiveDashboardPanel.razor](../src/AutoForge.UI/Features/Vehicles/Telemetry/LiveDashboard/LiveDashboardPanel.razor)
shares the existing editor, pinned-unit draft handling, canonical bounds,
display conversion and generic numeric/bar rendering through a distinct
Replay input. It creates no live feed for replay and does not own its clock.
Recorded labels include Good, Paused, Completed, unavailable and Stale
values, plus actual observation/capture/diagnostic/model context.
- [AddAutoForgeUi.cs](../src/AutoForge.UI/Features/Foundation/AddAutoForgeUi.cs),
[MainLayout.razor](../src/AutoForge.UI/Layout/MainLayout.razor) and
[Home.razor](../src/AutoForge.UI/Features/Foundation/Home.razor) wire factory
composition and shared navigation.
New tests are the TelemetryReplayTests partial class in
[TelemetryReplayTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryReplayTests.cs),
[TelemetryReplayLifecycleTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryReplayLifecycleTests.cs),
[TelemetryReplayPresentationTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryReplayPresentationTests.cs)
and [TelemetryReplaySourceCoexistenceTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryReplaySourceCoexistenceTests.cs).
The initial fixture, timing, rejection and rendered contracts were written
before production implementation; additional interleavings were added during
the post-green source-path review.
No new project, dependency, schema, transport, protocol, physics clock,
DiagnosticSessionIdentity, global vehicle registry or generic timeline framework
was introduced.
## Frozen C-01..C-14 matrix
All rows below are PASSED. Named methods are executed tests, not proposals.
| ID | Named evidence and exact outcome |
|---|---|
| C-01 | OfflineLoad_PreindexesOnlyMetadataAndReconstructsTheExactFourRecordPrefix; DynamicArchive_ReplaysOfflineWithRepeatedModelTimeAndOriginalCanonicalBytes. Real recorder/codec/filesystem input survives source disposal; offline DI has no original profiles/configurations. Explicit load establishes RecordedReplay plus a new instance and capture ID, Stopped, count 0, position 0, no measurements and no autoplay. |
| C-02 | OfflineLoad_PreindexesOnlyMetadataAndReconstructsTheExactFourRecordPrefix. At 1.5 s: count 1, A=850 Good and no B sample. At 2 s: count 3, A=1000 Uncertain, B=12.6 Good. At 4 s: count 4, A=850 Stale from observation 1 / 1 s, B remains 12.6 Good. The Uncertain 1000 is never promoted into last-Good authority. |
| C-03 | The same independent fixture and MidGapPause_DrainsOneOwnerAndResumeWaitsOnlyTheRemainingInterval. Previous from count 4 reconstructs count 3 / Uncertain 1000; ordinal 1 clears B and its later module status. First/Stop clear all values and caches to count 0 / time 0, retaining definitions, valid speed and acknowledgement. Seek time preserves the exact requested between-record playhead. Restart is explicit rewind-plus-Play; Completed does not silently replay. |
| C-04 | EqualAndZeroTimes_ManualNextIsOneButTimeSeekAndPlayApplyCompleteGroups; EmptyAndNonSignalArchives_NeverFabricateMeasurementsOrUseUtcDuration. Next applies one even inside a tie; time seek/Play apply whole due groups in sequence. Four zero-time records complete without a wait. Empty Play/Restart completes without a timer; empty seek zero is Stopped; empty Next/Previous visibly rejects. DTC observations change inspection/count/time but never invent signals/module read status. |
| C-05 | ControlledPacing_ScalesCaptureIntervalsWithoutChangingRecordedResults at 0.25x, 1x and 1000x; MidGapPause_DrainsOneOwnerAndResumeWaitsOnlyTheRemainingInterval. First 1 s interval waits 4 s, 1 s or 1 ms respectively; the next 1 s interval and final 2 s interval scale identically. Final recorded values/prefixes are identical at every rate. At 2x from 1.5 s, the next group waits 0.25 wall seconds; a 0.10-second pause freezes exactly 1.7 s/count 1, and resume waits exactly 0.15 more seconds despite wall time advancing while paused. |
| C-06 | MidGapPause_DrainsOneOwnerAndResumeWaitsOnlyTheRemainingInterval; EndPublication_RejectsReentryBeforeThePriorOwnerFinishesCleanup; CloseWithDelayedOldWake_NeverPublishesIntoClosedOrFreshFactorySession; WaitFault_IsVisibleAndKeepsCommittedPrefixUntilExplicitRecovery; ClockFaults_AreVisibleWithoutChangingTheLastCommittedSampleAndCanBeStopped (start/wake/pause); WaitFailureDuringPauseOrStop_DrainsButCannotMasqueradeAsSuccessfulCleanup (Pause/Stop). Repeated active Play creates one wait; pending drainage rejects competing starts/navigation/replacement before mutation. Seven end-publication reentries reject Busy. Repeated Stop and Close share their completion. Delayed wakes never publish into a closed/fresh context. Faults retain the committed prefix/sample state, remain visible after cleanup and require explicit recovery; a fault during Stop is not rewritten into successful empty Stopped state. |
| C-07 | InvalidRequests_AreTypedVisibleAndAtomicIncludingClosedAndUnloadedPlayers; ActualRouteControls_ReconcileRejectedRateRequirePartialAcknowledgementAndDrainOnUnmount. Undefined/unloaded/closed and invalid time/count/rate commands reject without context/value/definition/position/rate/ownership mutation, except explicit error notification. NaN, both infinities, 0.249x and 1000.001x reject; 0.25x and 1000x are accepted. Invalid exact unit, label, kind, range, precision, duplicate identifier and ninth gauge reject atomically. Eight and zero gauges are accepted. Actual bound speed input returns to committed 1 after rejecting 1000 while Playing. |
| C-08 | PartialAcknowledgement_IsPerLoadAndEndNeverRewritesOriginalCompletion (Incomplete/Error); ActualRouteControls_ReconcileRejectedRateRequirePartialAcknowledgementAndDrainOnUnmount; native partial warning check. Play/Next/seek require that load's explicit acknowledgement. Stop/First/inspection/Close remain allowed. End means Completed replay/end of retained data while original Incomplete or Error and its reason remain unchanged. Reload creates a new instance, clears acknowledgement/definitions/cursor and retains valid speed/global units. |
| C-09 | RecordedQualityAndUnitChanges_UseSharedRulesWithoutFutureOrCrossModuleFill (Unknown/null/text/nonfinite/unit-change/real-zero); MissingSignalOrNoPriorGood_UsesRecordedStaleOrExplicitErrorOnlyForTheAffectedModule; CorruptReplacementAndChangedOrigin_AreRejectedByTheActualStrictReaderWithoutReplacingReplay. Missing/unit-changed reads retain prior Good only as recorded Stale; without any prior Good, failed A is Error/--. B remains Good. A real Good zero remains numeric and later Stale zero, not a fabricated missing value. Unknown/non-numeric/nonfinite values stay unavailable. Uncertain 1000 over a 0..900 range retains Uncertain plus high-range status and 100% geometry; a later timeout restores 850 Stale, with original time. Unsupported source-origin edits fail strict loading atomically. |
| C-10 | RenderedRecordedRouteAndSharedEditor_ShowRejectionsPinnedUnitsAndNoGaugeOwnedClock; ActivePlayback_SharedGaugeHideUnitsAndUnmountCannotOwnTheClock; InvalidRequests_AreTypedVisibleAndAtomicIncludingClosedAndUnloadedPlayers; native temperature/bar draft check. Shared configuration supports 0..8 gauges. Invalid real editor events retain the committed definition. An Imperial draft stays pinned across a global Metric switch, and an unedited 40 m/s canonical bound remains exactly 40. Hiding/showing, unit changes, definition removal and gauge unmount during active playback neither cancel nor add a wait; the same owner continues with zero gauges. |
| C-11 | FailedCancelledAndLateLoads_PreserveValidatedContextAndCloseDrainsWithoutStalePublication; CorruptReplacementAndChangedOrigin_AreRejectedByTheActualStrictReaderWithoutReplacingReplay; CloseWithDelayedOldWake_NeverPublishesIntoClosedOrFreshFactorySession. Missing/corrupt/cancelled replacements retain the prior archive/context/data/definitions/position/rate. A held store deliberately ignores cancellation until released; close invalidates its generation first, shares an unfinished completion, drains it and prevents late publication. Busy loads reject overlapping commands. No failure is represented as a healthy empty replacement. |
| C-12 | OriginalPhysicalMetadata_IsNotRewrittenIntoCurrentReplayOrVehicleEvidence; DynamicArchive_ReplaysOfflineWithRepeatedModelTimeAndOriginalCanonicalBytes; actual rendered route/native provenance. Current kind is RecordedReplay; original Simulation or PhysicalHardware metadata remains unchanged. ModelResetAndRepeatedModelTime_RemainRecordedFactsNotPlaybackOrdering verifies actual recorded model steps 3,0,0,0 at capture times 1,2,2,4 seconds, including backward reconstruction of step 3. No diagnostic factory/control or catalogue fallback is used by replay. |
| C-13 | ReplayCoexistsWithRealLiveSource_WithoutReadsModelCallsOrRecordingWorkspaceRebind and the byte assertions throughout. A simultaneously open real dynamics source stays at step 7 with unchanged trace/status; transport-call delta=0, model-control-call delta=0 and archive Save calls=0 during replay/seek/unit operations. Its stopped unsaved recording serializes identically and is not rebound. An explicit subsequent live read succeeds and increments only the transport counter by one. Offline rendered DI also has failing live/model ports with both counters zero. Actual source archive bytes remain identical. All 313 accepted baseline cases remain present and pass. |
| C-14 | ActualRouteControls_ReconcileRejectedRateRequirePartialAcknowledgementAndDrainOnUnmount; RenderedRecordedRouteAndSharedEditor_ShowRejectionsPinnedUnitsAndNoGaugeOwnedClock; RenderedClockFault_RemainsVisibleDespitePriorCommandErrorAndRejectedRecoveryNavigation; native checks below. The real store/strict reader/Application/shared editor/native route supports saved static-v1/dynamic-v2 replay after app restart with an empty profile catalogue, normal controls, units, hide/show, end and close/re-entry. Timing/fault/race proof is controlled, not attributed to native human-speed observations. |
## Acceptance criteria
| ID | Result | Evidence |
|---|---|---|
| AC-01 | PASSED | C-01/C-11/C-12 real file loading, offline DI, atomic failed/cancelled replacement, generation guards and distinct current/original provenance. |
| AC-02 | PASSED | C-02..C-04/C-08/C-09 independent exact value/count/time/status assertions, before-first/ties/empty/partial handling and affected-module last-Good reduction. |
| AC-03 | PASSED | C-05..C-07 controlled monotonic waits/barriers, exact pause/resume, held cleanup, reentry, clock/wait/drain faults, typed rejections and explicit recovery. |
| AC-04 | PASSED | C-10/C-12 real rendered controls/editor and native numeric/bar/unit/draft/provenance checks; accepted live tests remain green. |
| AC-05 | PASSED | Final nine-project Release build: zero warnings/errors. 179 unit + 171 integration = 350 passes, zero failed/skipped. Native static/dynamic offline playback passed, including a final-binary restart confirmation. Zero live/model/write counters and unchanged native source hashes corroborate isolation. |
| AC-06 | PASSED | Executable frozen matrix, post-green source-path review/fixes, related documentation and truthful native-versus-controlled evidence below. |
| AC-07 | PASSED | main/65eed57 unchanged; no protected/history/orchestration/dependency changes, no staging/commit/push/agents, task-owned native data/helpers cleaned, and this single fenced report saved. Worker stops here. |
## Final commands and outcomes
SDK: dotnet --version -> 10.0.403 under the existing global.json roll-forward.
The framework/platform is existing xUnit v2/VSTest, not MTP. No packages,
SDKs, workloads or tools were installed, updated, repaired or restored.
The following ran sequentially after the final implementation/self-review:
| Exact command | Exit | Result |
|---|---:|---|
| dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --no-restore --filter "FullyQualifiedName~TelemetryReplayTests|FullyQualifiedName~LiveDashboardAcquisitionTests|FullyQualifiedName~DisplayUnitControlTests" | 0 | 49 passed, zero failed/skipped: 37 new replay cases plus 12 accepted live-dashboard/unit-selector cases. |
| dotnet build .\src\AutoForge.slnx --configuration Release --no-restore | 0 | All nine projects, including native Windows Client; 0 warnings, 0 errors. MAUI XAML source-generation output was informational. |
| dotnet test .\src\AutoForge.slnx --configuration Release --no-build | 0 | 179 unit + 171 integration = 350 passed, zero failed/skipped. |
| git --no-pager diff --check | 0 | No whitespace errors. Git reported only its existing LF-to-CRLF conversion notices for README/build guide. |
| git --no-pager diff --exit-code -- prompts Directory.Build.props Directory.Packages.props global.json NuGet.Config src\AutoForge.Core src\AutoForge.Diagnostics src\AutoForge.Simulation | 0 | Existing assignments/history/state/gate, dependency policy and foundational source implementations unchanged before adding this report. |
Per-project build result: AutoForge.Core, AutoForge.Diagnostics,
AutoForge.Simulation, AutoForge.Application, AutoForge.Components,
AutoForge.UI, AutoForge.Client, AutoForge.UnitTests and
AutoForge.IntegrationTests all PASSED. Portable projects remain net10.0;
Client remains net10.0-windows10.0.26100.0 / win-x64.
The 313-case baseline is 179 unit + 134 integration. This step adds 37
integration cases and deletes or weakens no baseline tests.
Existing limits remain 4,096 observations, 16 MiB encoded observations,
1 MiB header, 32 MiB archive and JSON depth 128. Gauge limits remain 0..8
with precision 0..3; valid rates are 0.25x..1000x inclusive. Individual waits
are bounded at 30 wall seconds with cancellable monotonic pacing.
Earlier iterations are not disguised as passing evidence:
- Initial compilation failed on the extracted unused key, Razor binding
quoting/nullability, a test-method insertion and nullable fault-message
flow. Each was corrected and the applicable checks rerun.
- The first executable focused run had 26 passes/4 failures: the shared live
missing-signal reason changed, the dynamic fixture selected a nonexistent
body endpoint, and an InlineData integer could not bind to nullable double.
The original live reason was restored and fixture/type errors corrected.
- Later focused runs exposed a too-exact floating-point text expectation and
an optimized Action event callback in the test renderer. The test now
verifies the independent conversion numerically and dispatches actual
renderer events with field feedback, rather than casting callbacks.
- First green was 30 cases. The post-green review expanded this to the final
49 focused cases and 350 full passes. Compilation failures are not claimed
as behavioral pre-fix reproductions; this was a new feature, not a claim
that replay existed before this assignment.
## Native observations and controlled-evidence distinction
The actual Release AutoForge.Client executable was launched with task-owned
WebView2 user data and local DevTools port 9378. A temporary Node/built-in
WebSocket helper operated the actual native WebView controls; there was no
new web host, dependency or production debug feature.
The real MAUI storage root was verified as:
C:\Users\g3\AppData\Local\User Name\com.companyname.autoforge.client\Data\recordings
LOCALAPPDATA was not assumed to redirect FileSystem.AppDataDirectory.
The archive directory initially contained zero files.
Normal native evidence:
1. Static v1 capture 5ae80280d3454f73bfc1fd001d2d88d2 retained four actual
operations, including engine/body DTC and signal reads. It was stopped,
source-closed and saved. After restart the profile catalogue was empty;
the replay route loaded it explicitly at count/time zero, configured a
numeric gauge before data, applied a non-signal first record without
fabricating a value, then displayed recorded 850 rpm from observation 2.
2. Static Play/Pause exposed a frozen between-record playhead. A speed edit
to 1000 while Playing visibly rejected and reset the actual input to 1.
Quiescent 1000x reached all four retained observations; Previous and Stop
worked, and Stop cleared readings to -- while retaining the definition.
3. Dynamic v2 capture 82af7a5151e74820a9a31eeb12de71a1 retained four actual
read.signals operations at stopped model step 7 / 0.7 seconds, with original
configuration/run. It was stopped, source-closed and saved, then replayed
after restart without reimporting profiles or recreating a model.
4. The dynamic horizontal bar displayed source 88.02090573589594 degC as
88.02 degC or 190.44 degF. An edit opened in Imperial units showed 32..212
degF; changing global units to Metric kept the draft pinned to Imperial/
degF, and an unedited save retained exact canonical 0..100 degC bounds.
Hide/show did not change count/playhead. Seek 40,000,000 ticks retained
exactly two records; Next/Previous, Play/Pause, end, First, active-play
route unmount and fresh Unloaded re-entry were exercised.
5. Warning-only fixture c91b0cd68fef4f18b10d4de81f648b35 was a separately
named derived copy of the real static observations, with explicit
Incomplete/LimitExceeded fixture completion. The actual strict store
validated it. Native Next rejected before acknowledgement at count zero;
acknowledgement and ordinal end retained original Incomplete and the
"end of retained data" warning. This is native warning/acknowledgement
evidence, NOT evidence of native count-limit exhaustion.
6. After the final fault/cleanup refinements and final 350-pass build, new
native confirmation captures were produced:
db553b85c9fd4e55a3c887801fafda7b (static v1, two actual operations) and
5f6516fdd718411f96bdb0e00229ce78 (dynamic v2, four actual operations at
model step 7). The app was restarted again. Its profile catalogue was
explicitly confirmed empty. The final binary loaded both stopped before
first, navigated/played them to counts 2/2 and 4/4, showed correct v1/v2
original provenance, stopped and closed with cleanup pending False.
Static duration was 217,926 ticks; dynamic duration was 92,006 ticks.
Source immutability was checked before and after native playback:
| Native capture | Unchanged SHA-256 |
|---|---|
| 5ae80280d3454f73bfc1fd001d2d88d2 | 07336400F122570ADA157C3A7B9B9EEAB8433D2B68121239BF95C4E03FF47DCA |
| 82af7a5151e74820a9a31eeb12de71a1 | 3B6339FAA09C75BEC8518E4E0345B3FAD984819FF279B5237D58364224798ABC |
| db553b85c9fd4e55a3c887801fafda7b | A31C968EC2A88A48CD5F3BB4BBFA03B52E7804488C71897081E0690CF645E87C |
| 5f6516fdd718411f96bdb0e00229ce78 | E7D4712AE417C4128036EEFBF5A6AD7101A19D5E2AE56D8C4ECDF39123C76C3D |
Native timing is normal UI observation only. Exact rates/mid-gap resume,
large-gap overflow bounds, concurrent ownership, malformed/cancelled/late
loads, source limits/errors and injected clock/wait/drain faults are proved
by controlled headless tests, not native elapsed-time claims.
Native preparation limitation: navigating between the existing recording
hosts while a Save-triggered refresh was pending exposed their pre-existing
uncaught "Another archive command is pending" initialization failure. A
brief enabled control/saved label can precede that follow-up refresh. This
stack was in unchanged TelemetryRecordingWorkspace.RefreshAsync and
TelemetryRecordingPanel.OnInitializedAsync, not the replay route. It was
not silently called a successful smoke or broadened into a recording change.
Source preparation was restarted; final confirmation captures were prepared
in separate app runs, then replayed after another restart. Assigned replay
checks subsequently passed with no native error overlay.
Cleanup PASSED: all seven task-launched Client process sessions completed;
only their specific PIDs were terminated, with replay/source already closed
where applicable. The five exact named task-owned archive files above
(including the warning fixture) were removed after verification; the archive
directory again contains zero files. The task-only WebView profile and helper
were removed, no task WebView processes remained, and port 9378 had no
listener. The existing/default WebView data and unrelated app data were not
removed. Original Metric/Road preference values were restored and observed.
## Post-green source-path self-review
Reviewed actual producers/consumers from factory -> strict store/codec ->
metadata index/prefix reduction -> owner/control snapshots -> routed page ->
shared editor/display converter/generic renderer. Public entry points reviewed
were Create, RefreshSavedCapturesAsync, LoadAsync, ReplaceDefinitions,
AcknowledgePartial, SetRate, SeekTime, SeekOrdinal, Next, Previous, First,
PlayAsync, RestartAsync, PauseAsync, StopAsync and DisposeAsync.
Normal, invalid, partial, equal/zero-time, backward, cancelled, faulted and
overlapping paths were walked against C-01..C-14. Eligibility is checked before
mutation. Resource release and the shared completion occur before ownership
becomes eligible for reuse; cancellation/drain awaits do not hold the state
gate. Data publication requires the current owner/generation; late closed
loads/wakes cannot publish. Metadata pre-indexing never supplies current
values. Faults retain original sample state, not a false successful end.
Findings fixed within Step-08:
- Removed the small cleanup/reuse eligibility window by releasing cancellation
resources before clearing owner/load eligibility and completing the shared
task under the state gate.
- Explicitly rejected data navigation/restart from Faulted until Stop/First
after cleanup; added exact atomic-state assertions.
- Preserved actual fault reason separately so an older UI command error or
later rejection cannot conceal a scheduler fault.
- Kept a non-cancellation wait fault during pending Pause/Stop visible Faulted,
rather than ignoring it as a stale wake or resetting it into success.
- Corrected native protocol-version Razor rendering and retained accepted
live missing-signal wording during shared projection extraction.
- Added real simultaneous live-source/unsaved-recording isolation and active
gauge-unmount/zero-gauge owner tests instead of relying on render counts
or an offline-only proxy.
All affected focused checks and required full build/regressions were rerun
after these fixes; final results are the 49/350 passes recorded above.
The final binary's normal saved-capture path was then checked natively again.
No known unresolved defect in the assigned replay exit contract remains.
## Documentation and preservation
Updated [README.md](../README.md),
[documentation index](../docs/README.md),
[architecture](../docs/architecture.md),
[build/test guide](../docs/build-and-test.md),
[live dashboard contracts](../docs/live-dashboard.md) and
[archive-v1 guide](../docs/telemetry-archive-v1.md); added
[recorded replay usage/contracts](../docs/telemetry-replay.md).
They distinguish accepted corrected recording from Step-08 pending review,
current replay versus original provenance, before-first versus zero-time seek,
exact prefixes/ties/last-Good, independent clocks, rates/drain/replacement,
partial acknowledgement, source immutability and native/control limitations.
The worktree began clean on main at 65eed57 and HEAD is unchanged. Existing
prompts/history/state/roadmap/assignment/shared gate were not edited; this is
the only new prompts file. No protected instruction file was created or
modified. The index remains empty. All nine projects/references and central
dependency/SDK policy remain unchanged; source/tests stay under src and
src/tests. Core/Diagnostics/Simulation implementations and physics/oracle
surfaces are unchanged.
Out of scope remains interpolation/resimulation, persistence of replay layouts/
cursor/rate, archive editing/migration/new durability, calibration/programming,
hardware access, AI and real-vehicle verification. No future step or
automatic orchestration action is authorized by this report.
STOP.
```