10 KiB
10 KiB
# Step-05B Worker Report
Identifier: Step-05B
Title: Independent range status and decisive dashboard tests
Status: COMPLETE — assigned worker implementation and verification passed; pending manual review. Step-05 and Step-05A remain NOT ACCEPTED pending orchestrator review.
## Summary and files
Corrected the Uncertain-value projection so range status is computed independently of signal quality. Added decisive real-provider/rendered assertions for low/high, in-range and exact-bound Uncertain values; expanded invalid-definition and finite-percent boundary coverage; and made controlled Stop-drain cleanup failure-safe. The dashboard feature, session ownership, providers, hosts, public contracts, schemas, protocols and lifecycle model were otherwise left unchanged.
Files changed:
- `src/AutoForge.Application/Features/Vehicles/Telemetry/LiveDashboard/LiveDashboardAcquisition.cs` — Uncertain finite readings now pass the existing shared `GetRangeState` result into the gauge projection.
- `src/tests/AutoForge.IntegrationTests/LiveDashboardAcquisitionTests.cs` — real-provider range/render assertions, full independent invalid-definition vectors, unchanged snapshot checks, and owned-session cleanup.
- `src/tests/AutoForge.UnitTests/LiveDashboardGaugeRenderingTests.cs` — isolated invalid-percent geometry inputs with a Good finite numeric presentation and a measurable visible 85% control.
- `docs/build-and-test.md` — appended the Step-05B pre-fix, focused/full test, and final native recheck evidence without rewriting Step-05A history.
- `prompts/Step-05B-Report.md` — this report; the only new prompt file.
## Actual pre-fix reproduction
Before changing the projection, ran:
`dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --no-restore --filter FullyQualifiedName~RealStaticReadsPreserveRangeQualityFailureAndRecoveryStates`
Exit code 1; the real static-provider/rewrite-transport regression failed on the actual finite Uncertain -10 rpm response: expected `OutOfRangeLow`, actual `null`. This was a behavioral assertion failure, not a compilation failure. After the fix, the same test passed with rendered numeric and bar states.
## Corrective matrix and evidence
| ID | Input / precondition | Executed assertion and result |
|---|---|---|
| C-01 | Real finite Uncertain -10 and 1100 rpm against 0..1000 | `RealStaticReadsPreserveRangeQualityFailureAndRecoveryStates` asserts exact numeric values, `Uncertain` quality, independent low/high `RangeState`, and 0%/100% bar projection. Its rendered panel asserts the same range and quality separately on both numeric and bar gauge cards; bar accessibility reports 0/100. Passed after fix. |
| C-02 | Uncertain in-range and exact minimum/maximum; existing Good/Stale/Stopped behavior | The same real-provider test asserts Uncertain 100 has Good range status and Uncertain values exactly 0 and 1000 have no low/high status (`RangeState.Good`) with 0%/100% geometry. It also checks stale and stopped last-known 850 rpm remain range-Good, while existing Good/out-of-range behavior remains intact. Passed. |
| C-03 | Start with one valid definition and a Running feed; reject each invalid vector | `InvalidDefinitionReplacementIsAtomicAndEnforcesZeroToEightLimit` tests NaN and positive/negative infinity in bounds; equal, reversed and finite-span-overflow ranges; precision -1/4; undefined kind; blank/padded/case-duplicate gauge IDs; blank/padded labels; blank/padded/unknown module and signal IDs; blank and nonmatching exact unit. Every rejection is typed and asserts the exact prior snapshot object, definitions, source identity and Running feed remain unchanged. Empty and eight definitions are accepted; the ninth is rejected with the same prior snapshot, source and feed. Passed. |
| C-04 | Good state and real numeric text `850 rpm`, changing only Percent | `BarGauge_GoodNumericValueWithInvalidPercentDoesNotRenderGeometry` separately tests null, NaN, positive infinity and negative infinity; each has no meter, no zero-width fallback, no invalid CSS value, and still displays 850 rpm/Good. The valid control `BarGauge_RendersFiniteProjectionStateAndSafeText` asserts visible `width: 85%` and accessible `aria-valuenow=85`. Passed. |
| C-05 | Assertion/command failure while controlled Stop cancellation is held | `StopDrainRejectsStartWithoutMutationAndAllowsRetryAfterCleanup` now releases the controlled cancellation barrier and disposes the session in `finally`, while preserving the rejection during drain and successful retry assertions. The other new session tests use async session disposal for failure paths; the pending-response barrier test already releases its barrier and disposes both sessions in `finally`. Passed. |
## Existing criteria mapping
| Criterion | Evidence |
|---|---|
| AC-01 | Pre-fix real-provider failure recorded above. Post-fix C-01/C-02 prove numeric, range, quality, geometry, and rendered accessibility are independent and correct. |
| AC-02 | C-03 executes the assigned invalid boundaries one at a time and proves exact atomic preservation; C-04 isolates the finite-number guard from Unknown/missing values and verifies the valid visible 85% control. |
| AC-03 | C-05 `finally` cleanup releases the scheduler barrier, drains the owner and disposes the source even if assertions fail; existing controlled reject/retry behavior remains asserted. |
| AC-04 | Complete nine-project Release build and 225-case Release regression suite passed. The exact rebuilt native Client configured and read gauges through actual static-v1 and dynamics-v2 sessions; native evidence and its limits follow. |
| AC-05 | Only the assigned Application projection, focused tests, directly related build/test documentation, and this report changed. Assignments, State/Roadmap, shared gate, prior reports and protected files are unchanged. No future feature was implemented. |
## Commands and outcomes
- Pre-fix regression command shown above: exit 1, expected behavior failure (`OutOfRangeLow` vs null).
- `dotnet test .\src\tests\AutoForge.UnitTests\AutoForge.UnitTests.csproj --configuration Release --no-restore --filter FullyQualifiedName~LiveDashboardGaugeRenderingTests` — exit 0; 8 passed, 0 failed, 0 skipped.
- `dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --no-restore --filter "FullyQualifiedName~RealStaticReadsPreserveRangeQualityFailureAndRecoveryStates|FullyQualifiedName~InvalidDefinitionReplacementIsAtomicAndEnforcesZeroToEightLimit|FullyQualifiedName~StopDrainRejectsStartWithoutMutationAndAllowsRetryAfterCleanup"` — exit 0; 3 passed, 0 failed, 0 skipped.
- `dotnet build .\src\AutoForge.slnx --configuration Release --no-restore` — exit 0; Core, Diagnostics, Components, Simulation, Application, UI, UnitTests, IntegrationTests and Client built successfully; 0 warnings, 0 errors. The MAUI XAML SourceGen notice was informational.
- `dotnet test .\src\AutoForge.slnx --configuration Release --no-build` — exit 0; UnitTests 134 passed and IntegrationTests 91 passed; 225 total, 0 failed, 0 skipped. The previous 224 cases were preserved; the new percent-null theory case accounts for the one additional case.
- `git diff --check` — passed; only the existing documentation LF-to-CRLF conversion notice was emitted.
## Native versus controlled evidence
The final native recheck ran the exact Release Client with an isolated WebView2 user-data folder. On `/diagnostics`, the editor bound a horizontal bar to the actual `module.engine.primary / synthetic.engine.speed` static-v1 signal and rendered 850 rpm from `SYNTHETIC_STATIC`. On `/simulation`, it bound a numeric gauge to the actual `module.engine.primary / engine.speed` dynamics-v2 signal and rendered 260 rpm from `SYNTHETIC_DYNAMIC`, with logical request tick, model step 1 / 0.1 seconds, configuration fingerprint and `dynamics-run-000002`. The dynamics model was paused. Both actual sessions were closed and their panels removed.
The native run validates that the existing configured dashboard baseline still works on both providers. It did not inject Uncertain samples or malformed definitions into the native UI; those specific vectors are asserted through the real static provider/rewrite transport and rendered component/integration tests. Native runtime: .NET SDK 10.0.402 and .NET 10.0.13; MAUI workload 10.0.403, `maui-windows` manifest 10.0.110; Windows SDK target 10.0.26100.0; WebView2 Runtime 156.0.4314.50. The exact native Client/WebView2 processes and port 9351 were verified stopped; the isolated task profile was removed. A separate pre-existing default WebView2 profile discovered during an initial smoke attempt was left untouched.
No package restore/install, SDK/workload modification, system update, staging, commit, push or destructive Git action occurred.
## Documentation and source-path self-review
`docs/build-and-test.md` now records the actual red pre-fix assertion, post-fix focused vectors, full suite counts and final native observations. Existing live-dashboard documentation already describes range/quality independence and exact-unit behavior, so no unrelated guide rewrite was needed. Imperial/metric preferences and conversion remain future, unassigned scope.
Reviewed the changed source path from finite-value/quality classification through the shared range helper, gauge projection and numeric/bar panel consumers. The range helper is reached only after finite numeric validation, uses strict below/above comparisons (so exact bounds remain Good range status), and does not change `Uncertain` source quality or raw numeric values. Rejected definitions still validate before snapshot mutation; assertions confirm the exact old snapshot and feed/source context survive. The invalid-percent test holds all presentation parameters constant except Percent, while the 85% control checks actual accessible geometry. Stop cleanup releases its cancellation barrier before awaiting session disposal. Focused tests and the complete build/test suite were rerun after the self-review assertions.
## Preservation review and stop point
The worktree was clean before this assignment. No pre-existing changes were present to overwrite. No assignment, historical report, `prompts/STATE.md`, `prompts/ROADMAP.md`, shared gate or protected instruction was edited. Only the four listed implementation/test/documentation paths and this new report are changed.
Worker implementation and assigned verification are COMPLETE. Step-05 and Step-05A remain NOT ACCEPTED until manual orchestrator review. Stop here; no next step or automatic handoff was performed.