AutoForge/prompts/Step-03A-Report.md

266 lines
15 KiB
Markdown

```
# Step-03A Worker Report
## Status
**COMPLETE** for the assigned Step-03A corrective implementation, regression
testing, complete Release build/test verification and native baseline smoke.
Step-03 remains **NOT ACCEPTED** pending manual orchestrator review. Step-04
was not started.
## Scope and implementation
Corrected the three Step-03 integrity findings without changing the nine-project
architecture, introducing a package/project/host, or enabling physical or replay
operations.
### R-01 — Incomplete and ambiguous requests
**Root cause:** `SyntheticDiagnosticRequest.Version` had an initializer of `1`;
`TryDecodeRequest` relied on object deserialization alone, so missing version or
logical tick could be supplied by defaults. Duplicate JSON object names were
left to last-value deserialization.
**Fix:** The v1 decoder now requires all ten exact camelCase wire properties,
checks object-property uniqueness recursively before deserialization, and still
uses `JsonUnmappedMemberHandling.Disallow`. Missing, unknown, case-variant,
duplicate, invalid-time or invalid-timeout requests are malformed. The
simulation transport returns its explicit no-response outcome before session
binding/address/service dispatch, so no trustworthy response correlation is
fabricated and no virtual service runs. Explicit zero logical time remains valid.
A structurally complete version-2 request still receives the pre-existing
`UNSUPPORTED_VERSION` negative response.
Unit tests remove each of the ten required request fields, test invalid timing,
explicit zero, unknown/case-variant properties and a real duplicate service
property. The real Simulation transport test sends five malformed clear frames:
missing version, missing logical time, duplicate conflicting service,
duplicate conflicting correlation and duplicate conflicting version. The
IntegrationTests-only oracle confirms `SYN-P0123` remains unchanged after each
rejected frame. Duplicate test inputs are parsed and counted as containing two
actual properties before dispatch; the test does not reconstruct them through a
dictionary that would erase duplicates.
### R-02 — Complete reply validation and truthful trace
**Root cause:** Interpreted service payloads had incomplete semantic checks, and
discovery compared a decoded probe module only after `ExchangeAsync` had already
recorded a `PositiveResponse`.
**Fix:** The protocol client now rejects duplicate object properties before
interpretation; requires exact known envelope/service fields; checks response
success/error/data consistency; and validates nonblank trimmed identifiers/text,
module/profile identity, unique DTC/service/signal identifiers, service and
quality tokens, and static signal source/timing. Signal snapshot ticks must equal
the request logical start tick. Valid empty DTC collections and explicit JSON
`null` signal values remain valid.
Probe identity and advertised-service validation now run in the same decoder
path as other services. An invalid probe produces one `ProtocolError` result and
one `ProtocolError` trace entry with the actual immutable raw request/reply;
there is no positive trace followed by a contradictory discovery result.
Invalid/inconsistent responses return no partial value. No-response, timeout,
cancel and malformed transport behavior remains unchanged.
Actual DI/transport integration tests mutate actual encoded Simulation replies
to cover negative signal tick; blank signal identifier/unit; wrong module
source; invalid quality; duplicate signal identifier property; duplicate signal
IDs; blank DTC code/description; unknown/duplicate advertised services; wrong
probe module; duplicate response correlation; missing/unknown response fields;
positive response with error text; and negative response with data. Each asserts
a typed failure, null result value and matching trace outcome/raw response.
Existing unsupported-version/address/service, valid empty/nonempty DTC,
identity, timeout/no-response/cancellation and trace tests remain in the full
suite.
### R-03 — Clear acknowledgement versus verified state
**Root cause:** UI treated any successful post-clear DTC read as proof the
virtual clear succeeded, including a nonempty readback.
**Fix:** Added `DiagnosticSession.ClearTroubleCodesAndVerifyAsync` and the
Application result `DiagnosticClearVerificationResult`. The use case preserves
the low-level acknowledgement separately, then performs a fresh read. It marks
the operation verified only if that read succeeds with an empty DTC set.
Nonempty readback is retained as observed data with verification failure;
negative/protocol/no-response/timeout/cancelled or closed-session readback is
explicitly uncertain/failed and never verified. A zero acknowledgement count
does not prevent verification when the fresh read is valid and empty.
The UI now calls the Application use case rather than duplicating verification
logic. It renders acknowledgement and verification as separate outcomes,
preserves valid remaining codes, reports failure/uncertainty, never emits the
old “was cleared and re-read” success for a nonempty readback, and explicitly
states that no physical write occurred.
A deterministic test transport decorator sits below the real protocol client
and Application session: it returns a positive clear acknowledgement without
dispatching clear to the actual Simulation transport. The actual subsequent
readback and independent oracle show `SYN-P0123` remains. Rendered UI asserts
error tone, visible remaining code, verification failure, no false success and
unchanged oracle state. Additional Application integration tests cover genuine
clear/empty verification, a second already-empty clear with acknowledgement
count zero, no-response, timeout, malformed protocol response, request
cancellation and session close.
## Architecture and interfaces
The nine-project graph and production dependencies are unchanged. No assembly,
executable, package, workload action, external provider or public state oracle
was added. The new Application result and session method depend only on the
existing Diagnostics protocol/session contracts. Normal Application/UI paths
still cannot access Simulation internals; the virtual DTC oracle remains
IntegrationTests-only. Existing exact profile/revision/provider/kind selection,
metadata-only target resolution, state isolation, reset/close lifecycle,
logical-clock/trace behavior and synthetic-only restrictions remain.
The test-only `RewritingDiagnosticTransportFactory` wraps the real simulation
factory only in IntegrationTests. It is not registered by Client and is not a
runtime fault selector.
## Before/after regression evidence
Before production fixes, the focused unit command below failed both new tests:
missing `version`/`logicalTimeTicks` was accepted, and the actual duplicate
`service` property was accepted.
`dotnet test .\src\tests\AutoForge.UnitTests\AutoForge.UnitTests.csproj --configuration Release --filter "FullyQualifiedName~SyntheticProtocolV1_RequiresExplicitWireVersionAndLogicalTime|FullyQualifiedName~SyntheticProtocolV1_RejectsDuplicateDispatchAndBindingProperties"`
Result: **2 failed, 0 passed**. The assertions expected decoder rejection and
observed `Actual: True`.
Before production fixes, the five real integration/rendered regression tests
failed as expected:
`dotnet test .\src\tests\AutoForge.IntegrationTests\AutoForge.IntegrationTests.csproj --configuration Release --filter "FullyQualifiedName~SimulationTransport_RejectsIncompleteAndAmbiguousClearFramesWithoutMutation|FullyQualifiedName~ProbeIdentityMismatch_IsProtocolErrorInTheSingleMatchingTraceEntry|FullyQualifiedName~RealProtocol_RejectsInvalidSignalAndDtcDataWithoutSuccessValues|FullyQualifiedName~ResponseEnvelope_RejectsDuplicateBindingFieldAndContradictoryPositiveError|FullyQualifiedName~DiagnosticRoute_ClearAcknowledgementWithRemainingCodesIsNotReportedVerified"`
Result: **5 failed, 0 passed**. The real transport accepted the missing-field
clear as positive; malformed signal/DTC and inconsistent replies were returned
as `PositiveResponse`; wrong probe identity was discovered as an error while its
trace remained `PositiveResponse`; and the UI reported `Information` for a
nonempty readback rather than verification failure.
After fixes, focused decoder tests passed **3/3** and focused real integration/
rendered correction tests passed **12/12** (the latter includes the existing
normal native-provider rendered workflow).
## Full verification
The final complete commands were run sequentially after source/test changes:
- `dotnet build .\src\AutoForge.slnx --configuration Release --no-restore`
— exit 0; all nine projects built, including the actual Windows Client target
`net10.0-windows10.0.26100.0` / `win-x64`; **0 warnings, 0 errors**.
- `dotnet test .\src\AutoForge.slnx --configuration Release --no-build`
— exit 0; all tests passed, with no failures or skips.
| Test project | Discovered | Passed | Failed | Skipped |
|---|---:|---:|---:|---:|
| AutoForge.UnitTests | 111 | 111 | 0 | 0 |
| AutoForge.IntegrationTests | 64 | 64 | 0 | 0 |
| Total | 175 | 175 | 0 | 0 |
This retains the historical Step-03 baseline of 108 unit + 53 integration
(161 total) and adds 14 test cases. The original Step-03 run did not cover the
Step-03A vectors and is not represented as having done so. No dependency
manifest changed; no explicit restore command or workload operation was
required. Test-runner restore checks reported all projects up-to-date.
## Native Client versus rendered fault verification
Launched the actual Windows Client executable:
`.\src\AutoForge.Client\bin\Release\net10.0-windows10.0.26100.0\win-x64\AutoForge.Client.exe`
For local smoke inspection only, the child process received
`WEBVIEW2_ADDITIONAL_BROWSER_ARGUMENTS=--remote-debugging-port=9222`; the
WebView2 DevTools endpoint was accessed via loopback at `127.0.0.1`. No source,
launch configuration or persistent environment setting was changed. The
native Client process was stopped after the smoke.
Native observations:
- Explicitly loaded the synthetic truck example, explicitly selected exact
profile revision/provider/kind and opened Simulation session
`diagnostic-session-000001`.
- Probed three expected modules: engine and body responded; transmission
`0x20` visibly returned `NoResponse` and was not counted as a responder.
- Read engine identity, `SYN-P0123`, static engine speed 850 rpm, voltage 12.6 V,
quality/unit/origin/source and logical time.
- Inspected actual raw JSON request and response bytes in the native trace,
including matching correlation and decoded positive probe entry.
- Virtual clear showed one acknowledged item followed by a fresh read with zero
DTCs and the status “Fresh readback verified an empty virtual DTC state… No
physical write was performed.”
- Reset created session `diagnostic-session-000002`; probing and reading restored
the configured initial `SYN-P0123`. The configured transmission no-response
remained visible. Closed the session and verified the active-session view was
removed.
- Rechecked `/execution-targets` provider metadata without a session,
`/vehicle-profiles` imported revision, `/`, and an unmatched not-found route.
The false-ack/nonempty-readback fault is **rendered IntegrationTests evidence**,
not native Client evidence. Production Client has no supported runtime
fault-injection control; adding one merely to create native evidence would
change the product surface and was not authorized. The native run verifies the
normal baseline clear; the rendered test verifies the failure case.
## Documentation and files
Modified:
- `src/AutoForge.Diagnostics/SyntheticProtocolV1.cs`
- `src/AutoForge.Diagnostics/SyntheticDiagnosticProtocolClient.cs`
- `src/AutoForge.Application/Features/Vehicles/Diagnostics/DiagnosticSessionService.cs`
- `src/AutoForge.UI/Features/Vehicles/Diagnostics/DiagnosticPageViewModel.cs`
- `src/tests/AutoForge.UnitTests/SyntheticDiagnosticsContractTests.cs`
- `src/tests/AutoForge.IntegrationTests/DiagnosticSessionCompositionTests.cs`
- `src/tests/AutoForge.IntegrationTests/DiagnosticUiCompositionTests.cs`
- `README.md`
- `docs/README.md`
- `docs/architecture.md`
- `docs/build-and-test.md`
- `docs/synthetic-diagnostics-v1.md`
- `docs/simulation.md`
- `docs/adr-0004-synthetic-diagnostics.md`
Created:
- `src/tests/AutoForge.IntegrationTests/RewritingDiagnosticTransportFactory.cs`
- `prompts/Step-03A-Report.md`
Documentation now states exact wire naming/presence/unknown/duplicate policy,
no-dispatch semantics, full response validation, trace/result consistency,
acknowledgement versus verified empty readback, test counts and evidence
distinctions. Synthetic/non-vehicle, static-signal, no-physical-write,
no-programming and no-replay limits remain explicit.
## Acceptance criteria
| ID | Status | Evidence |
|---|---|---|
| AC-01 | PASSED | All ten fields, unknown/case variant and duplicate rules tested; five conflicting/missing clear frames sent through the real bound Simulation transport; independent oracle confirms no state mutation. |
| AC-02 | PASSED | Exact envelope/service payload policy, timing, identifiers, DTC metadata, signal semantics, probe metadata and success/error/data consistency tested; invalid replies return typed ProtocolError with no value. |
| AC-03 | PASSED | Wrong probe module and invalid service claims produce discovery failures matching one actual-raw-reply ProtocolError trace entry; no positive/partial probe result. |
| AC-04 | PASSED | Application clear-and-verify tested for genuine empty result, zero acknowledgement count, false acknowledgement/nonempty readback, no-response, timeout, malformed response, cancellation and closed session; rendered UI retains remaining code and displays failure; oracle confirms unchanged state. |
| AC-05 | PASSED | Original 161-case Step-03 baseline remains within the final 175-case suite; session/state/target/profile/clock/trace/project-graph regressions pass; nine-project architecture and internal oracle boundary unchanged. |
| AC-06 | PASSED | Actual pre-fix failures recorded; focused corrected regressions pass; complete nine-project Release build has 0 warnings/errors; 111 unit + 64 integration pass; actual native Client baseline smoke performed and distinguished from rendered injected fault. |
| AC-07 | PASSED | README, documentation index, architecture, build guide, protocol guide, simulation guide and ADR-0004 document the corrected contract and evidence. |
| AC-08 | PASSED | Scope and protected orchestration/history files preserved; no Step-04, agents, handoff automation, staging, commit or push; required report written; stop for manual review. |
## Preservation, limitations and handoff
No assignment, roadmap, orchestration state, prior report, protected Copilot
instructions, dependency manifest or solution/project graph was modified by
Step-03A. At task start, `prompts/ROADMAP.md` and `prompts/STATE.md` were already
modified in the shared worktree; they were read as authority and left untouched.
Pre-existing Step-03 implementation changes were preserved and integrated.
No stage, commit, push or automated handoff was performed.
The protocol remains synthetic UTF-8 JSON, Simulation-only and in-process.
Signals are static fixtures; no vehicle, physical transport, replay, telemetry
capture, dynamics, calibration/firmware/programming, persistence or backend is
available. Worker implementation and verification are complete; manual
orchestrator acceptance remains pending. Stop here; do not begin Step-04.
```