12 KiB
12 KiB
# Step-07A worker report
## Status
COMPLETE — Step-07A worker execution and verification are complete. Manual
orchestrator review remains pending. Step-07 remains NOT ACCEPTED; Step-08
remains blocked and was not started.
## Contract and implementation
Step-07A corrected the original Step-07 R-01 through R-04 without expanding
the archive format or changing the recorder design:
- R-01: permitted an actual empty `TransportOutcome.Response` while preserving
its trace representation as `ProtocolError` with null response bytes; applied
only the source-approved case-insensitive identity comparisons for response
envelopes and module identities.
- R-02: validates signal association before dereference and reuses the shared
pure v2 signal-shape validator for known IDs, exact units/JSON kinds, finite
values and duplicate IDs. Negative model time is rejected; v1 opaque,
nullable and non-numeric values remain supported.
- R-03: removes stopped, failed, limit-stopped and source-closed captures from
the diagnostic session lifetime list and session ownership map while keeping
current stopped samples available for save/inspection. The controlled
Stop-drain regression releases its observer barrier in `finally`.
- R-04: distinguishes a legitimate missing first-use directory from invalid
roots/listing failures and preserves actual panel save errors by refreshing
only after explicit save success.
Production changes are in [ProviderContracts.cs](../src/AutoForge.Diagnostics/ProviderContracts.cs),
[SyntheticDynamicSignalContract.cs](../src/AutoForge.Diagnostics/SyntheticDynamicSignalContract.cs),
[SyntheticDiagnosticProtocolClient.cs](../src/AutoForge.Diagnostics/SyntheticDiagnosticProtocolClient.cs),
[DiagnosticSessionService.cs](../src/AutoForge.Application/Features/Vehicles/Diagnostics/DiagnosticSessionService.cs),
[SessionTelemetryCapture.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Recording/SessionTelemetryCapture.cs),
[TelemetryArchiveCodec.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Recording/TelemetryArchiveCodec.cs),
[FileTelemetryArchiveStore.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Recording/FileTelemetryArchiveStore.cs),
[TelemetryRecordingWorkspace.cs](../src/AutoForge.Application/Features/Vehicles/Telemetry/Recording/TelemetryRecordingWorkspace.cs),
and [TelemetryRecordingPanel.razor](../src/AutoForge.UI/Features/Vehicles/Telemetry/Recording/TelemetryRecordingPanel.razor).
Regression coverage is in
[TelemetryRecordingIntegrationTests.cs](../src/tests/AutoForge.IntegrationTests/TelemetryRecordingIntegrationTests.cs).
## Frozen corrective matrix and acceptance mapping
| Requirement | Evidence and result |
|---|---|
| R-01 / C-01 | Pre-fix the transport constructor rejected empty `Response` payloads. Post-fix `CaptureRecordsActualPositiveNegativeTimeoutNoResponseCancellationAndMalformedReplyBytes` exercises real observations; the empty reply stays `ProtocolError` with null response bytes and round-trips through save/new-store load without fabricated bytes or successful signals. PASSED. |
| R-01 / C-02 | Pre-fix case-variant source-approved envelope/module identities failed archive association validation. `CaptureArchive_AcceptsCaseInsensitiveProtocolAssociations` exercises the real positive source response and archive round trip, retaining original bytes and spellings; exact request, signal ID, unit, protocol token and dynamics identities remain exact. PASSED. |
| R-02 / C-03 | Pre-fix a positive signal read with null snapshot escaped as `NullReferenceException`. `ArchiveWithPositiveSignalReadAndNullSnapshot_IsRejectedAsTypedInvalidArchive` now verifies typed `InvalidArchive` and that prior workspace inspection remains unchanged. PASSED. |
| R-02 / C-04 | `DynamicArchive_RejectsCoherentEditsThatViolateTheV2SignalContract` updates raw and decoded values coherently and recalculates encoded-byte bookkeeping for wrong unit/JSON kind, unknown/duplicate IDs and negative model time; the source contract is not weakened. PASSED. |
| R-02 / C-05 | Existing v1 nullable/non-numeric/opaque/Unknown handling, valid v1/v2 readbacks, malformed-reply evidence, exact source/context association and workspace inspection preservation remain covered by recording integration tests. PASSED. |
| R-03 / C-06 | `StoppedCapture_IsRemovedFromSessionLifetimeResources`, `CaptureFailure_ReleasesSessionLifetimeRegistration`, the count/byte-limit assertions, discard/replacement, and source-close tests check registration/ownership deterministically while stopped unsaved data remains inspectable/saveable and close draining remains intact. PASSED. |
| R-03 / C-07 | `CaptureOwnership_RejectsCompetingOwnerAndStopDrainsAcceptedCallbackBeforeReturning` releases the controlled observer barrier in `finally` before observer disposal/drain. PASSED. |
| R-04 / C-08 | `FileArchive_CancellationAndFilesystemFailureAreTypedAndLeaveNoPublishedArchive` and storage listing assertions verify an existing-file root/list failures are typed while a missing first-use directory remains empty. No permissions were changed. PASSED. |
| R-04 / C-09 | Pre-fix actual panel event evidence: `ActualPanelSaveFailure_RemainsVisibleAndSuccessfulRetryRefreshesCatalogue` failed because the collision error was replaced by a malformed-archive catalogue error after refresh. Post-fix it verifies the collision error remains rendered, the capture stays unsaved, the existing file is unchanged, and retry succeeds and refreshes. PASSED. |
| C-10 | The 303-test Step-07 baseline (179 unit + 124 integration) remains represented; seven additional focused integration cases yield 179 unit + 131 integration (310 total). Existing limits and source/model behavior pass, along with full Release and native workflow checks. PASSED. |
| AC-01 | Actual pre-fix R-01 reproductions and post-fix real protocol/file round trips for empty response and case-variant identities. PASSED. |
| AC-02 | Typed null-snapshot rejection, coherent v2 semantic negatives, and preserved valid v1/v2 cases. PASSED. |
| AC-03 | Deterministic owner registration checks for normal stop, failure/limit stop, discard/replacement and source close; barrier cleanup in `finally`. PASSED. |
| AC-04 | Typed store/workspace outcomes, invalid-root behavior and actual panel error/retry rendering. PASSED. |
| AC-05 | 303 prior cases plus seven integration cases, exact configured limits, nine-project Release build, full test suite and native baseline. PASSED. |
| AC-06 | Frozen matrix, documentation, post-green self-review, protected-file/history check, this report and stop-before-Step-08. PASSED. |
## Actual pre-fix results
The focused recording selection before correction ran 23 tests: 17 passed and
6 failed (exit code 1). Observed behavioral failures were:
- Empty `TransportOutcome.Response` was rejected by the transport constructor
with `ArgumentException`, preventing the actual empty-reply trace from being
recorded and archived.
- Source-accepted case-variant response/module identities were rejected by
archive validation.
- A positive `read.signals` with null `signalSnapshot` escaped as
`NullReferenceException`.
- A stopped capture remained registered in the open diagnostic session
(observed resource count 1 where 0 was required).
- A storage root that was an existing file returned an empty catalogue rather
than a typed failure.
- The actual panel save collision path refreshed the catalogue after the
workspace had recorded the save error, replacing the visible collision
message with a malformed-archive message.
The actual panel event was separately rerun before correction using:
`dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --filter "FullyQualifiedName~ActualPanelSaveFailure_RemainsVisibleAndSuccessfulRetryRefreshesCatalogue"`
It failed as expected (1 failed, exit code 1): the rendered/workspace error
did not contain the collision message after the real handler refreshed the
catalogue. This was behavioral evidence, not a compilation failure.
## Post-fix commands and results
Commands ran sequentially after focused validation:
| Command | Exit | Result |
|---|---:|---|
| `dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --filter "FullyQualifiedName~TelemetryRecordingIntegrationTests"` | 0 | 26 passed, 0 failed, 0 skipped. |
| `dotnet build .\src\AutoForge.slnx --configuration Release --no-restore` | 0 | All nine projects, including Windows Client; 0 warnings, 0 errors. MAUI XAML source-generation notice was informational. |
| `dotnet test .\src\AutoForge.slnx --configuration Release --no-build` | 0 | Unit: 179 passed. Integration: 131 passed. Total: 310 passed, 0 failed, 0 skipped. |
The focused final run includes actual source observations, coherent archive
mutations, registration-release tests, the actual panel handler/render
collision and retry test, and the 4,096-operation, 16 MiB observation, 1 MiB
header and 32 MiB archive boundaries. JSON depth remains 128. No limits were
raised.
## Native versus controlled evidence
The rebuilt Windows Client was exercised using its local WebView2 DevTools
endpoint:
- Static v1 captured seven actual operations. It was stopped and saved, the
source was closed, and after app restart the archive loaded with its raw
request/response bytes and decoded source signals.
- Dynamics v2 was paused at model step 7 / 0.7 seconds. With gauges hidden, it
captured four actual `read.signals` operations. It was stopped and saved,
the source was closed, and after restart the archive loaded with raw bytes
and decoded signal/model context.
- Metric/Road preferences remained unchanged. Native captures were
`034e287ae82e4e14aa4d55e78581cbf1` (static) and
`69def5834cf742ce9fa729dd4bea1b16` (dynamics). Those exact task-generated
archive files and temporary WebView2/profile directories were removed
after closing the Client and DevTools listener.
Native evidence covers the normal record/stop/save/restart/load/source-close
flow. Filesystem failures, malformed coherent archives and controlled
lifetime/barrier cases were tested through the actual store, codec, session,
workspace and panel integration paths; no native storage-fault injection is
claimed. In this environment the `LOCALAPPDATA` override did not redirect
MAUI `FileSystem.AppDataDirectory`; only exact task-owned captures were
removed, with no reset of app data or modification of preferences.
## Documentation, self-review and preservation
Updated [README.md](../README.md), [docs/README.md](../docs/README.md),
[docs/architecture.md](../docs/architecture.md),
[docs/telemetry-archive-v1.md](../docs/telemetry-archive-v1.md), and
[docs/build-and-test.md](../docs/build-and-test.md). They describe exact
source identity rules, v1/v2 archive validation, empty-response representation,
bounded capture ownership, typed catalogue failures, successful-save-only
refresh, the corrected build/test evidence and the native/controlled boundary.
The historical Step-07 results remain unchanged.
Post-green self-review traced the actual transport-to-trace-to-capture-to-codec
path, the store/workspace/panel save and refresh path, session close and owner
unregistration, failure/limit cleanup, and controlled Stop drain. Validation
occurs before archive publication/inspection; invalid data does not replace
prior inspection; errors remain explicit; stopped in-memory snapshots remain
available when required. Rejected save does not mark the capture saved or
overwrite an existing file. Success refreshes as before. No additional issue
was found in this review.
The 303 baseline cases, all nine projects, public source/profile/configuration
contracts and fixed capture/archive limits were preserved. No package,
workload, system or project was added or changed. Protected assignment/state/
roadmap/gate/history files were unchanged. Only the requested new report was
added under `prompts`; no staging, commit, push, orchestration edit or Step-08
work was performed. Stop here for manual review.