34 KiB
34 KiB
STEP IDENTIFIER: Step-13
TITLE: Verified synthetic calibration comparison and demonstration
STATUS: COMPLETE - worker implementation and exit verification passed.
MANUAL GATE: Step-13 acceptance remains pending. STOP; Step-14 remains blocked.
DATE: 2026-11-01
BASE: 245c464 ("After Step 12"); initially clean worktree; no commit/staging/push.
OUTCOME
An explicit current preview can enter the existing unit-pinned canonical draft,
but Save, backup, confirmation/approval, Apply, independent full readback and
comparison remain separate actions. The actual owning rehearsal admits copied
historical verified evidence, not caller-constructed success records.
Only warm-idle target and torque cells project into a NEW immutable configuration.
Two fresh actual Simulation/v2 runs provide bounded real signals/captures,
canonical measured metrics, explicit no-overwrite save/readback and existing
RecordedReplay evidence. Original configurations/runs and ordinary recordings
are not replaced. No physical operation, general executor or runtime manager.
FINAL VERIFICATION
Command 1:
dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --no-restore --filter "FullyQualifiedName~Comparison_|FullyQualifiedName~PreviewHandoff_|FullyQualifiedName~TorqueDraft_" --verbosity minimal
PASSED; exit 0; 50 passed, 0 failed, 0 skipped; approximately 8 s.
This is 49 new comparison/handoff cases plus one retained pinned-draft case.
Required sequential commands, executed after that focused check:
dotnet build .\src\AutoForge.slnx --configuration Release --no-restore
PASSED; exit 0; all nine projects; 0 warnings, 0 errors; 2.29 s.
dotnet test .\src\AutoForge.slnx --configuration Release --no-build
PASSED; exit 0; 179 unit + 466 integration = 645; 0 failed, 0 skipped.
Unit duration approximately 198 ms; integration approximately 9 s.
All 596 accepted cases retained; 49 integration cases added.
| Project | Final build | Final tests |
|---|---|---|
| AutoForge.Core | PASSED | N/A, non-test project |
| AutoForge.Diagnostics | PASSED | N/A, non-test project |
| AutoForge.Simulation | PASSED | N/A, non-test project |
| AutoForge.Application | PASSED | N/A, non-test project |
| AutoForge.Components | PASSED | N/A, non-test project |
| AutoForge.UI | PASSED | N/A, non-test project |
| AutoForge.Client | PASSED, actual Windows MAUI Hybrid executable | N/A |
| AutoForge.UnitTests | PASSED | 179 passed, 0 failed/skipped |
| AutoForge.IntegrationTests | PASSED | 466 passed, 0 failed/skipped |
Earlier focused checks passed 25, then 49 cases as coverage expanded.
Earlier complete builds/suites passed 621 and 644 cases respectively.
One intermediate full 644-case run FAILED with TaskCanceledException in the
unmodified ScenarioOwnerCannotBeReplacedAndResetOrCloseSuppressesOldRunPublication
regression during Reset/disposal. Its actual control/page/session/test files
have no diff. The isolated selector below passed all 7 cases, and subsequent
full 644- and final 645-case runs passed. This intermittent result is disclosed,
not erased or repaired by an unrelated dynamics change:
dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --no-build --filter "FullyQualifiedName~ScenarioOwnerCannotBeReplacedAndResetOrCloseSuppressesOldRunPublication|FullyQualifiedName~Comparison_Held" --verbosity minimal
FUNCTIONS, FILES AND INTERFACES
| File | Implemented responsibility |
|---|---|
| [SyntheticCalibrationProjection.cs](../src/AutoForge.Core/Calibration/SyntheticCalibrationProjection.cs) | Pure canonical Read/Create; exact supported definition/value validation and all-field immutable constructor projection |
| [DynamicsConfigurationService.cs](../src/AutoForge.Application/Features/Vehicles/Simulation/DynamicsConfigurationService.cs) | IDynamicsConfigurationService.RegisterProjection reuses profile validation and existing canonical catalogue collision handling |
| [CalibrationAuthoringService.cs](../src/AutoForge.Application/Features/Vehicles/Calibration/CalibrationAuthoringService.cs) | Baseline extraction reuses the pure projection Read; existing Save/head/history contracts remain |
| [RunbookPreviewService.cs](../src/AutoForge.Application/Features/Vehicles/Runbooks/RunbookPreviewService.cs) | CopyCurrentCandidateToDraft validates exact expected current preview/workspace/head and returns complete copied canonical values, without Save/target action |
| [CalibrationRehearsalSession.cs](../src/AutoForge.Application/Features/Vehicles/Calibration/CalibrationRehearsalSession.cs) | CaptureComparisonSourceAsync plus current owning proof guard; fresh full readonly admission and source/lifetime checks |
| [CalibrationComparisonSource.cs](../src/AutoForge.Application/Features/Vehicles/Calibration/CalibrationComparisonSource.cs) | Internal-constructor copied proof bound to actual owning service/generation, original backup and verified Apply lineage |
| [CalibrationComparisonModels.cs](../src/AutoForge.Application/Features/Vehicles/Calibration/CalibrationComparisonModels.cs) | Typed state/failures, immutable bounded samples/metrics/captures/save outcomes, canonical readonly report and explicit result retirement |
| [CalibrationComparisonSession.cs](../src/AutoForge.Application/Features/Vehicles/Calibration/CalibrationComparisonSession.cs) | Factory, EvaluateAsync, real bounded RunAsync, complete canonical signal validation, SaveCapturesAsync, Cancel/Close/drain |
| [CalibrationPage.razor](../src/AutoForge.UI/Features/Vehicles/Calibration/CalibrationPage.razor) | Explicit preview handoff; prior pinned units and exact typed unedited candidate retained until separate Save |
| [CalibrationRehearsalPanel.razor](../src/AutoForge.UI/Features/Vehicles/Calibration/CalibrationRehearsalPanel.razor) | Explicit comparison opening binds THIS actual owning rehearsal, without automatic evaluation |
| [CalibrationComparisonPanel.razor](../src/AutoForge.UI/Features/Vehicles/Calibration/CalibrationComparisonPanel.razor) | Explicit operation/provider/Evaluate/Cancel/Close/Save controls, readonly report, stale/partial/per-ID errors and async unmount |
| [AddAutoForgeUi.cs](../src/AutoForge.UI/Features/Foundation/AddAutoForgeUi.cs) | Existing UI composition registers comparison factory |
| [CalibrationComparisonTests.cs](../src/tests/AutoForge.IntegrationTests/CalibrationComparisonTests.cs) | Actual proof/projection/model/capture/store/replay, independent invariants, faults/boundaries/controlled owners |
| [CalibrationAuthoringUiTests.cs](../src/tests/AutoForge.IntegrationTests/CalibrationAuthoringUiTests.cs) | Actual rendered pinned preview handoff and exact tiny/nonuniform canonical Save |
| [CalibrationRehearsalTests.cs](../src/tests/AutoForge.IntegrationTests/CalibrationRehearsalTests.cs) | Existing fixture optionally supplies real dynamics or a below-boundary faulting dynamics transport; old default unchanged |
| [DynamicsEngineTests.cs](../src/tests/AutoForge.UnitTests/DynamicsEngineTests.cs) | Strengthens the retained independent warm regression: exact 3361/336.1 and every one of 3000 cruise model steps within 45 MPH +/-1 MPH |
Nine-project graph unchanged: Core owns pure mapping; Application owns trusted
admission/workflow; UI owns explicit commands. Application/UI do not reference
Simulation implementations or access a private engine sample/oracle.
No new project/package/host/schema, protocol change, importer duplication,
physical port, runbook executor, alternate runtime or dependency installation.
EXECUTED FROZEN CONTRACT MATRIX
Named Comparison_* cases below are in the comparison test file linked above.
The preview UI case is in the authoring UI test file linked above.
| Contract | Named executed evidence and decisive assertion | Result |
|---|---|---|
| C-01 | Comparison_CurrentPreviewHandoffRejectsStaleOrSupersededWithoutSaving (superseded/stale-head/wrong-workspace); PreviewHandoff_KeepsExistingPinnedUnitsAndExactCandidateUntilSeparateSave; real model preview handoff and final native unit-toggle recheck | PASSED: complete canonical candidate, unchanged head/history on Copy/reject, no target reads/writes; separate Save; existing pinned units preserved |
| C-02 | Comparison_ProofCannotBeConstructedByPublicCallerSuccessRecords; Comparison_TrustedAdmissionRejectsWrongOrSupersededActualProof; Comparison_FreshOwningReadRejectsDivergentOrUnavailableProofBeforeProjection; Comparison_RecoveryRequiredRejectsUntilActualReadOnlyReconciliationVerifiesApply | PASSED: no public proof/result constructor; actual original backup, owning Apply/current Proposed/candidate/identity/revision/full readback agree; wrong operation/head/closed/external revision and malformed/unavailable evidence reject |
| C-03 | Comparison_ActualVerifiedApplyProjectsOnlySupportedFieldsAndRunsEquivalentCapturedScenarios; Comparison_NoChangeProjectionRetainsExactContentFingerprintAndCopiesNonuniformCells; Comparison_ProjectionRejectsInvalidCompleteValuesWithoutChangingOriginal | PASSED: independent whole serialized configuration equality after changing ONLY the two supported properties; no-change truthful fingerprint; nonuniform cells; 11 key/unit/axis/shape/bound/nonfinite negatives |
| C-04 | ActualVerifiedApplyProjectsOnlySupportedFieldsAndRunsEquivalentCapturedScenarios; Comparison_3900And3901StepBoundaryIsMeasuredWithUnchangedCaptureLimits; real native pair and all four actual archive raw-context inspections | PASSED: fresh distinct real run/capture IDs, exact Simulation/v2 provider, StepAsync(1), engine AND transmission every five steps plus final, no duplicate final reads; no replay/oracle/canned telemetry as physics |
| C-05 | ActualVerifiedApplyProjectsOnlySupportedFieldsAndRunsEquivalentCapturedScenarios; original real local-model native path through separate Save/backup/approval/Apply/full readback; final native manual recheck of changed UI/lifetime/report paths | PASSED: exact 650->1250 rpm and 2000-rpm 360->350 N*m, all other cells/axes unchanged; both last TEN warm samples within 1 rpm and stationary <=0.001 m/s |
| C-06 | Comparison_RealFileReadbackAndReplayPreserveExactBytesAndRejectWrongReferences; actual model-pair native Save/readback and both actual RecordedReplay prefix/end/raw trace inspections | PASSED: canonical measured metrics and full lineage, distinct archive identities, exact file/readback bytes, actual RecordedReplay not physics; wrong filename/run corruption retains prior replay context |
| C-07 | Comparison_HeldActualSignalTransportCancelsAndDisposesWithoutLatePublication (Cancel/Close); Comparison_HeldFinalActualReadDrainsOrRejectsSupersessionWithoutLateSuccess (Cancel/Close/head/external); Comparison_PendingSaveRejectsCompetingOwnerAndCloseDrainsWithoutDeletingSuccessfulFile; final native observed-pending Cancel/unmount/fresh owner/Close | PASSED: one Busy-gated owner, actual barriers released in finally, drain outside required gates, no late current pair/rebase, prior success retired Stale/as-of and inspectable |
| C-08 | Comparison_ActualTransportSignalFaultNeverPublishesHealthyMetrics (missing/unit/quality/numeric 1e309/wrong kind/run/timeout); Comparison_FewerThanTenActualWarmSamplesRejectsWithoutShortenedPassingWindow; 3900/3901; partial/readback-corrupt/post-publication-failure save tests | PASSED: typed visible errors and partial evidence, no zero/nonfinite/uncertain healthy metrics; exact per-ID uncertainty and retries without overwrite/deletion; unchanged archive caps |
| C-09 | Independent whole-config test; retained WarmBaseline_MeetsConfiguredScenarioThresholdsAndIsReproducible strengthened for ALL 3000 cruise model steps; real transport/store/replay faults and unchanged complete baseline suites listed below | PASSED: producer/consumer faults below feature boundary, independent test-only numerical assertions, no private oracle exposed or invented firmware/checksum/physical-write mechanisms |
| C-10 | Final sequential nine-project Release/645-case suite; actual native original warm capture -> real model -> preview/draft/Save -> verified target -> equivalent pair -> capture Save/readback/actual replay; final native corrections and sequential existing route checks | PASSED: 596 preserved plus 49 new cases; real original model content recorded below, no mock-as-native or physical/unassigned mutation |
| Acceptance | Executed mapping | Status |
|---|---|---|
| AC-01 | C-01..C-03 actual owning source, independent invariants and atomic negative/current-head guards | PASSED |
| AC-02 | C-04..C-06 actual equivalent scenarios, exact idle/driving gates, real captures/readback/replay | PASSED |
| AC-03 | C-07..C-09 real-boundary barriers, exact limits, partial/corrupt/uncertain saves and preserved isolation | PASSED |
| AC-04 | C-10 final 645-case/nine-project Release and genuine native model/target/model-run/archive path | PASSED |
| AC-05 | Source-path self-review, docs, protected hashes/history, accurate native/control distinction and this sole new prompt report | PASSED; manual acceptance still pending |
Preserved executed baseline evidence includes:
- [SyntheticDiagnosticsContractTests.cs](../src/tests/AutoForge.UnitTests/SyntheticDiagnosticsContractTests.cs): strict wire/version/identity/service/result contracts.
- [DiagnosticSessionCompositionTests.cs](../src/tests/AutoForge.IntegrationTests/DiagnosticSessionCompositionTests.cs): actual provider/session composition, supported timeout/unsupported/invalid reply behavior and clear/readback semantics.
- [DynamicsWorkflowTests.cs](../src/tests/AutoForge.IntegrationTests/DynamicsWorkflowTests.cs): actual v2 identity/config/telemetry, interruption/reset/close and owner behavior.
- [CalibrationRehearsalLifecycleTests.cs](../src/tests/AutoForge.IntegrationTests/CalibrationRehearsalLifecycleTests.cs), [CalibrationRehearsalReviewTests.cs](../src/tests/AutoForge.IntegrationTests/CalibrationRehearsalReviewTests.cs), [CalibrationRehearsalReopenTests.cs](../src/tests/AutoForge.IntegrationTests/CalibrationRehearsalReopenTests.cs): real CAS/readback/recovery/reconciliation/Restore and held close/reopen cleanup.
- [TelemetryRecordingIntegrationTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryRecordingIntegrationTests.cs) and [TelemetryReplayTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryReplayTests.cs): actual recording/archive corruption/limits and exact replay boundaries.
All were executed in the final full suite; no waiver or skipped case.
ACTUAL NATIVE DEMONSTRATION - REAL MODEL PATH
Actual Client executable:
[AutoForge.Client.exe](../src/AutoForge.Client/bin/Release/net10.0-windows10.0.26100.0/win-x64/AutoForge.Client.exe)
Task-owned first Client PID 9248; native WebView2/CDP 9386, not an alternate host.
Initial persisted preferences Metric/Road; initial local archive file count 0.
Qualified shared runtime, left unchanged throughout this assignment:
Ollama 0.40.13; app uses http://127.0.0.1:11434.
Existing serve PID 50460, created 2026-10-31T14:39:16.1305236-04:00.
Current startup 2026-10-31T14:39:16.200-04:00:
"Ollama cloud disabled: true"; actual store D:\AI\ollama\models.
Model huihui_ai/qwen3-abliterated:8b-v2
Digest 543f9deff86d8347fd96ea1f3e54e73861ea70d52def71d31ccaeaf013de6ba6
Installed bytes 5027793060; GGUF/qwen3/8.2B/Q4_K_M; Apache2 reported metadata.
Model metadata is not cryptographic proof of local execution.
No pull/install/configuration/startup/restart/unload/runtime setup was performed.
Original configuration:
sha256:v1:e8991ae139b54959ec53a4fe64b08dc9bfc03feb533496cf741f2c9f3a2342da
Profile autoforge.synthetic.test-truck revision 1.
Engine module.engine.primary; transmission module.transmission.primary.
Model autoforge.synthetic.longitudinal v1.
Definition autoforge.synthetic.engine-authoring revision 1.
Metadata retained as metadata: hardware.synthetic.engine-a;
software.synthetic.engine-1; firmware unknown; calibration.synthetic.engine-base.
Workspace c50f9c62fd3a411986640c6020e8bbcf.
Baseline 79b55bb68c8940418855a048887c5017; revision 1; history 1/32.
Warm target 650 rpm; axis [650,1000,2000,3200,4200,5000] rpm;
cells [240,300,360,350,300,180] N*m.
Original native warm run:
diagnostic-session-000001 / dynamics-run-000001 / same original config.
Explicit actual 100-step advance = 10.0 virtual seconds, then Accelerate at 0%.
Independent ordinary capture 4fb599dccbf74b608c4188aa456fe788;
started 2026-11-01T10:01:42.6843819+00:00;
2 actual validated observations, 12562 encoded observation bytes.
Explicit recording Stop and dynamics Close; stopped unsaved capture retained.
Later ordinary /simulation inspection confirmed THIS exact capture remained
Complete and unsaved after the comparison pair; it was not replaced/discarded.
Exact real model goal:
Propose warm-idle target 1250 rpm and set torque at 2000 rpm to 350 N*m; keep every other value unchanged.
Native generation started 2026-11-01T10:02:57.939Z.
Observed click-to-success latency 18854.89999999106 ms.
Original content passed the unchanged validator without extraction/repair.
Runbook 7878da6a1a424dd893714ab1296f27db.
Preview 68ba217e75844e50ab325e1be6a1df3d; Current / ChangesProposed.
Exact two canonical diffs: warm +600 rpm; torque at 2000 rpm -10 N*m.
ORIGINAL MODEL CONTENT, VERBATIM (not a second controlled response):
{
"schemaVersion":1,
"runbookIdentifier":"7878da6a1a424dd893714ab1296f27db",
"targetKind":"simulation",
"rationale":"Propose warm-idle target 1250 rpm and set torque at 2000 rpm to 350 N*m",
"steps":[
{
"stepIdentifier":"calibration-preview",
"operation":"proposeCalibration",
"source":{
"workspaceIdentifier":"c50f9c62fd3a411986640c6020e8bbcf",
"expectedHeadIdentifier":"79b55bb68c8940418855a048887c5017",
"profileIdentifier":"autoforge.synthetic.test-truck",
"profileRevision":1,
"configurationIdentifier":"sha256:v1:e8991ae139b54959ec53a4fe64b08dc9bfc03feb533496cf741f2c9f3a2342da",
"moduleIdentifier":"module.engine.primary",
"modelIdentifier":"autoforge.synthetic.longitudinal",
"modelVersion":1,
"definitionIdentifier":"autoforge.synthetic.engine-authoring",
"definitionRevision":1
},
"values":{
"scalars":[
{
"logicalIdentifier":"engine.warm-idle-target",
"unit":"rpm",
"value":1250
}
],
"tables":[
{
"logicalIdentifier":"engine.torque-curve",
"unit":"N*m",
"axisUnit":"rpm",
"axis":[650,1000,2000,3200,4200,5000],
"cells":[240,300,350,350,300,180]
}
]
}
}
]
}
Explicit native Copy produced scalar 1250/cells [240,300,350,350,300,180] while
saved head/history stayed Baseline revision 1. Only a separate Save created
Proposed cc7311bb8c374de4bc69e74418496044, revision 2, parent original Baseline.
Original source/fingerprint and all unrequested saved values remained unchanged.
Actual isolated target/provider autoforge.simulation.calibration-rehearsal:
Instance 9bbd543f8c8f4824a50ee1b72f0a42e9, original revision 0.
Verified two-read backup 1e5518a9d99e4398916be0ebd8df0608.
Apply preparation bee7561f6f9a4f8991a136568a315224.
Explicit confirmation/single-use approval 080d3ad2aad44a028ecc13a79bc3ff96.
Dispatched Apply e50b5940c2ee4b9b8489ee56e61a7f84;
started 2026-11-01T10:08:31.8770393+00:00;
completed 2026-11-01T10:08:31.8786715+00:00.
Acknowledgement True/revision 1, separately from independent full actual readback.
VerifiedApplied / CandidateMatching / actual revision 1.
Full before image 650/[240,300,360,350,300,180].
Full actual readback 1250/[240,300,350,350,300,180].
All six axis coordinates, units and unchanged cells matched exactly.
Explicit comparison 5d89e81c166d40488f3df4590c4e7524.
Admitted historical proof as of 2026-11-01T10:08:46.5550875+00:00.
Derived NEW configuration:
sha256:v1:3ee252697bfb69f36f4b99dca4c0267f7517a17a459666fb47551fa6ed978450
Baseline: diagnostic-session-000002 / dynamics-run-000002;
capture c731886c10fb4d9ba5c82b5bc387be07.
Derived: diagnostic-session-000003 / dynamics-run-000003;
capture df8ef66c6d7b4acabbea0e63b95b9a0c.
Same explicit actual autoforge.simulation.dynamics provider and equivalent
initial state/five-phase scenario; distinct NEW fresh owned sessions/captures.
Both Complete, 1346 actual v2 observations = 673 engine/transmission sample sets.
ACTUAL CANONICAL NUMERICAL RESULTS
| Metric | Baseline | Derived |
|---|---|---|
| Warm target rpm | 650 | 1250 |
| Actual last-ten maximum absolute idle error rpm | 4.084768079337664E-10 | 7.855760486563668E-10 |
| Actual last-ten maximum absolute stationary speed m/s | 0 | 0 |
| Warm phase model steps / sampled sets | 100 / 20 | 100 / 20 |
| Total model steps / virtual seconds | 3361 / 336.1 | 3361 / 336.1 |
| Warm/Accelerate/Cruise/Brake/ReturnIdle model step counts | 100/103/3000/58/100 | 100/103/3000/58/100 |
| Acceleration seconds | 10.3 | 10.3 |
| Braking seconds (round-trip representation) | 5.800000000000001 | 5.800000000000001 |
| Sampled cruise sets / min..max m/s | 600 / 20.1168..20.1168 | 600 / 20.1168..20.1168 |
| Fuel consumed from initial kg | 0.518385971679477 | 0.5184897031483956 |
| Distance from initial m | 6193.507764002321 | 6193.389583605566 |
| Final actual transmission shift count | 4 | 4 |
Derived-minus-baseline measured fuel +0.00010373146891862195 kg;
distance -0.11818039675472392 m; shift count delta 0.
No improvement/sign requirement invented. Fixed 0.5 virtual-second sampling is
not continuous extrema or a complete shift timeline. Independent test-only
baseline assertions inspect ALL 3000 cruise model steps at 45 MPH +/-1 MPH;
production sampled count 600 is explicitly different.
UNCHANGED CONFIGURATION INVARIANTS
Only Model.WarmIdleTargetRpm and EngineTorqueCurve cells changed; axis unchanged.
Independent whole serialized configuration DeepEquals replaces only those two
expected properties before comparison. Original export remained content exact.
Unchanged model fields: StepSeconds, VehicleMassKilograms, WheelRadiusMeters,
DragAreaSquareMeters, RollingResistanceCoefficient, AirDensityKilogramsPerCubicMeter,
GravityMetersPerSecondSquared, DrivetrainEfficiency, MaximumBrakeForceNewtons,
IdleSpeedRpm, MaximumEngineSpeedRpm, LaunchSlipEndSpeedMetersPerSecond,
IdleFuelRateKilogramsPerSecond, FuelKilogramsPerJoule, AmbientTemperatureCelsius,
InitialEngineTemperatureCelsius, EquilibriumEngineTemperatureCelsius,
ThermalResponsePerSecond, ForwardGearRatios, UpshiftSpeedsRpm, DownshiftSpeedsRpm,
FinalDriveRatio. Also unchanged: format/model/profile/revision/vehicle/module/
calibration references, complete initial conditions and every scenario field.
Original warm/accelerate/cruise/brake/return holds/limits: 10/60/300/60/10 s;
driving target 20.1168 m/s. No fidelity/shift schedule/manufacturer data redesign.
NATIVE SAVE, READBACK AND ACTUAL RECORDED REPLAY
Explicit pair Save showed BOTH persisted True/readback verified True.
Real model-pair archives:
c731886c10fb4d9ba5c82b5bc387be07: 8669778 file bytes;
8667485 encoded observation bytes; SHA256
0B4EFFAF4469A5BD2F60939F9DCC4F74BD5045AF8C2EDCD66301147F15CBD3E9.
df8ef66c6d7b4acabbea0e63b95b9a0c: 8668397 file bytes;
8666104 encoded observation bytes; SHA256
7DC25DE26831F7AB46280C5DEB18DE753E20C3A31C10E11579190E2C33FC0623.
Both inside unchanged archive bounds: 4096 observations, 16 MiB observations,
1 MiB header, 32 MiB archive. No global ordinary recording takeover.
Actual native /telemetry-replay selected/loaded each exact saved ID:
Baseline replay instance e2d3f867d30e4dddb934ef63dd39c763.
Applied count 1/1346 inspected as the exact one-observation prefix.
End 1346/1346, capture elapsed 963557 ticks / 0.0963557 s.
Derived replay instance b810e3aa30d34579955b197f411edfca.
End 1346/1346, capture elapsed 1088190 ticks / 0.108819 s.
Both retained original Simulation source, exact profile/session/config/run/v2,
Complete/User source completion and raw original request/response base64.
Final native raw observation 1346: module.transmission.primary/address 20,
read.signals/PositiveResponse; correlations diagnostic-session-000002-001346
and diagnostic-session-000003-001346; diagnostic ticks 1345->1346.
Independent archive-byte decoding confirmed actual final response model step
3361/time 336.1 and transmission.shift-count = 4 / count / Good with exact
header configuration/run correspondence. Capture elapsed time is not model time.
Navigation closes page owners; archived proof is historical, not current target
authority, and RecordedReplay never re-executes physics.
FINAL REBUILT NATIVE RECHECK, FAULTS AND OWNERS
After source-path corrections, final native Client PID 53432 rechecked changed
UI/lifetime/report paths with explicit MANUAL preview, not another claimed LLM.
Final assembly SHA256 matched the final Release build:
Application 19A02DD34C0643D4C8ACF3CB3CD0BD1477D8935898BFE05BF17DE48D97FAE2A7
UI 3A9443FC98152C3273B6BBC5E0A3B4FF22C3436E963DBC0826553D9F1B209E5F
Existing Metric draft stayed Metric/N*m while global units became ImperialUS.
Copy then separate Save kept exact candidate cells; Metric/Road restored.
Workspace 93a59ed7655346508c739bae1478ae9f;
Proposed 4c82e4b8246d48a5a488b5d3833bd401, revision 2.
Target 336ab6478e5e4acab281d4d5274049e1, revision 0->1.
Two-read backup f3e492dcc6cc4ce3a9a9a7ca53e16a34.
Preparation bccd73d6000d4a0681f8efa5edb24a39.
Explicit approval a87c89d05e854ea9b5dce5085df1d85a.
Actual verified Apply 79c2e83dea11493eb8df3b64ace100d5.
Final comparison 1e5f51c28c364afc937636d4980cd12e;
proof d0f6fe8979754bd8ad22812bb3fa1f4f as of
2026-11-01T10:22:25.2194362+00:00.
Same original/derived fingerprints and every numerical result above.
Baseline diagnostic-session-000001/dynamics-run-000001;
capture 8805241baade45c4bf4a41347d3d0d95, 8669924 file bytes;
8667631 observation bytes; SHA256
F12CE555AF187836C44781A3CF87BA84A8D7EA9DDE440210599D00B541F50C18.
Derived diagnostic-session-000002/dynamics-run-000002;
capture c2c4aea34aec42d18e881cbb19430fc0, 8668178 file bytes;
8665885 observation bytes; SHA256
FD116FF6845FDA98E460AD4E4A42F892899839F675ED651FB6CC44DAD4695916.
Both 1346 observations, explicit Save/readback verified. Final successful native
Save text truthfully said "exact readback verified", not "not saved" beside True.
All four files independently checked actual final raw run/config/step/time/count
before cleanup. Actual native replay evidence above is for the REAL MODEL pair.
Final native Cancel was triggered only after observing
"Evaluating; operation/drain pending True."
Result: "Cancelled; operation/drain pending False."; visible cancellation;
same earlier successful result retained, explicitly Stale/as-of.
Observed-pending unmount then reached Home without a late comparison report.
Reentry created Idle/fresh owner with no old comparison/target authority.
Explicit comparison Close reached Closed, not a reused prior completion.
Original native export still 650/[240,300,360,350,300,180], original five-phase
scenario unchanged; deliberate saved history remained 2/32 before app cleanup.
Earlier real-model native checks also established report unchanged by display
units, and a separate deliberate 1250->1300 authoring Save marked the retained
comparison Stale without changing original config or programmed image.
Reevaluation against that new unmatched head visibly rejected owning proof.
Controlled tests, not native timing claims:
- Held actual wire signal and final target read prove Busy/Cancel/Close/source
supersession; resources dispose once and no late healthy publication.
- Real model 3900-step completion produces 1560 actual observations; required
step 3901 rejects with retained partial evidence and no complete pair.
- Actual shortened warm scenario rejects fewer than ten samples, not a reduced
passing window; true JSON numeric 1e309 and wrong-kind "NaN" are separate faults.
- Second save failure preserves first ID/file and unsaved second capture.
- Readback corruption preserves published ID; retry reads without overwrite.
- Failure reported AFTER final publication checks existing catalogue on retry,
verifies exact original bytes and does not collide, overwrite or delete.
- Held Save Close drains and retains truthful per-ID publication evidence.
- Real file/replay filename/run corruption rejects while retaining prior context.
SELF-REVIEW AFTER FIRST GREEN
Walked actual producers/consumers: preview/current-head -> pinned draft/Save ->
owning rehearsal backup/Apply/readback -> internal proof -> pure constructor/
catalogue -> real Diagnostics control/signals -> independent capture -> store/
codec/readback/replay -> UI publication/Close. Checked normal, rejected, partial
and concurrent paths, exact validation ordering and authoring->rehearsal gates.
Issues corrected within Step-13:
1. Copy originally used current display selection rather than existing draft
units; preserve prior pinned units and exact typed unedited fields.
2. Cancellation can return a diagnostic Cancelled reply instead of throwing;
check token before signal classification. First native attempt was Failed
with a cancelled transport reply; final native attempt is correctly Cancelled.
3. Save can fail after publication; retry verifies existing ID read-only, and
closing owners retain per-ID publication evidence without healthy pair claims.
4. Move cancellation callbacks/Close drain outside required state gates; ensure
diagnostic disposal still executes if capture cleanup fails.
5. Retire retained current-result status on admitted reevaluation/Close, perform
another fresh owning full read before publication, and show as-of/Stale.
6. Surface owning IO failure as typed comparison proof failure (new test first
failed with bare IOException, then passed); wrong replay error expectation
was corrected to the existing typed archive contract without changing replay.
7. Include exact torque/source/proof/preparation lineage in the readonly report;
correct contradictory successful-Save text and distinguish sampled/full-step
numerical evidence. Preserve all old behaviors and independent Components.
Known unchanged baseline/tool observations, not hidden Step-13 fixes:
- The intermediate unmodified cancellation-regression failure is disclosed above.
- Rapid smoke-driver navigation between recording-host routes while the shared
archive refresh was pending produced the existing recording-panel initialization
Busy error ("Another archive command is pending; the current capture was not
changed") in TelemetryRecordingWorkspace.RefreshAsync/TelemetryRecordingPanel.
Those files have no diff and unrelated recording-navigation cleanup is expressly
out of scope. Recovered only the task-owned Client (PID 40612); then verified
all SIX actual rendered route headings with the actual archive owner idle:
target metadata, vehicle profiles, static diagnostics, simulation, runbooks,
calibration. All passed, Blazor error display none, Metric/Road preserved.
No URL-only proxy was accepted as route evidence.
- A cold smoke-driver attempt omitted explicit profile revision; source creation
and dependent actions correctly rejected. Corrected the driver selection only.
- The task-owned readiness observer initially omitted characterData events;
corrected instrumentation, not product code, and separated dispatch/read checks.
- Initial compiler/test-helper issues were corrected; none remain in final build.
DOCUMENTATION, LIMITATIONS AND PROTECTED HISTORY
Aligned [README.md](../README.md), [documentation index](../docs/README.md),
[architecture](../docs/architecture.md), [build/test](../docs/build-and-test.md),
[authoring](../docs/calibration-authoring.md), [rehearsal](../docs/calibration-rehearsal.md),
[runbooks](../docs/runbooks-v1.md), [local generation](../docs/local-runbook-generation.md)
and new [comparison usage](../docs/calibration-comparison.md).
Step-12/12A is accepted; Step-13 awaits manual review.
Proof is volatile/historical owning evidence, not perpetual external attestation.
Supported logical image is not full ECU backup/firmware/checksum/authenticity.
No real-world fuel/shift/safety benefit, hardware identity or physical permission
is inferred. Failed physics may have a Complete stopped archive coverage prefix,
but its separate run error/null metrics prevent successful comparison.
Pair save is not atomic or crash-durable; no rollback deletion is performed.
No model/private telemetry/oracle/test data is substituted into production.
Protected current SHA256 hashes rechecked unchanged:
[Step-13.md](Step-13.md)
121A97B67A6604BE53E8C5742A7F2001971A99F3B6E9E861671C39704B6BA535
[STATE.md](STATE.md)
8EBE7AC3FF763B9AE042E1044C88A605A1FAE5FAA897C53AA170CE77893C2327
[ROADMAP.md](ROADMAP.md)
0018A86E12D17360363E4F4D4EF6BEA4CEC3891A92B5516C5DD8D8C1C799F599
[WORKER-QUALITY-GATE.md](WORKER-QUALITY-GATE.md)
F9C0B14516B70345E4CE93AD1EDB8856094BC92E26A37AEB16EDE687AD021804
[Step-12A-Execution-Report.md](Step-12A-Execution-Report.md)
C5056359F33CC95C4D53F2ACAF7DBAB9E59B8B99CE66B0CD99B3475B427621F5
All other tracked prompt/assignment/historical reports remain unchanged.
Only this new report added under prompts. Git whitespace check passed;
Git LF/CRLF informational notices are not build warnings.
No agents, orchestration edits, staging, commits, pushes or Step-14 work.
CLEANUP - EXECUTED AND VERIFIED
Stopped only freshly identity-verified task Client PIDs 9248, 53432 and 40612.
Removed only four exact task-created verified archive IDs listed above; initial
and final user archive file count 0. No unknown/user archive or rollback deletion.
Removed 21 exact task native helper/evidence files, the inspected task-owned
step13-webview profile and three exact owned oversized temporary outputs.
Session root/files folder and all other session material preserved.
Remaining Step13 native targets 0; CDP 9386 listeners 0.
Original stopped unsaved warm capture was preserved during evaluation and
discarded only with final task-owned app process cleanup, not pair acquisition.
Original persisted preferences restored Metric/Road.
Shared Ollama serve PID 50460, original creation time/binary/version/model/digest/
store/cloud-disabled setup remain untouched and responsive.
Persistent deliverables are source/tests/docs and this report with original
model content, actual IDs/numerics/hashes and honest execution boundaries.
STOP FOR MANUAL REVIEW. No automatic acceptance or next-step handoff.