# Phase 6 Evidence: Demo, Trace UI, Documentation, and Verification

## Scope

Phase 6 completed the run-aware demo/documentation surface and final verification for `session-run-trace-isolation`.

Implemented:

- Demo scripts read Chat `runId`, query exact Trace with `?runId=...`, and submit feedback with `runId`.
- Trace UI accepts `?sessionId=...&runId=...` and calls the exact Trace API when `runId` is present.
- Chat UI remembers the latest run target and links to Trace Workbench with `sessionId + runId` when available.
- MVP table and architecture docs now describe `chat_session`, `diagnosis_run`, `agent_step.run_id`, `tool_invocation.run_id`, and transitional `case_library.diagnosis_id` semantics.

## Static Verification

Commands:

```powershell
node --check src\main\resources\static\app.js
node --check src\main\resources\static\trace.js

$scripts = @(
  'mvp\demo\scripts\run-payment-timeout-demo.ps1',
  'mvp\demo\scripts\run-interview-demo-check.ps1'
)
foreach ($script in $scripts) {
  [scriptblock]::Create((Get-Content -Raw -Encoding UTF8 $script)) | Out-Null
  Write-Host "Parsed $script"
}

openspec validate session-run-trace-isolation --strict
```

Result:

- JavaScript syntax: passed.
- PowerShell script parsing: passed.
- OpenSpec strict validation: passed.

## Focused Tests

Command:

```powershell
mvn -q "-Dtest=ChatControllerTest,DiagnosisTraceServiceTest,FeedbackControllerTest,FeedbackServiceTest,AiOpsServiceTest" test
```

Result: passed.

## Final Same-Session Multi-Turn E2E

Startup command:

```powershell
mvn spring-boot:run -Dspring-boot.run.profiles=mvp-demo
```

Startup log:

- `target/e2e/phase6-mvn-20260710-211831.out.log`
- `target/e2e/phase6-mvn-20260710-211831.err.log`

Application readiness:

- `Started Main in 15.501 seconds`
- `ReadinessState changed to ACCEPTING_TRAFFIC`

E2E session:

- `sessionId`: `e2e-phase6-chat-codex-20260710-2120`
- `run1`: `run-e2a97696-4398-4abc-90e4-28f45c838f92`
- `run2`: `run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172`

Artifacts:

- `target/e2e/phase6-chat1-20260710-2120.json`
- `target/e2e/phase6-chat2-20260710-2120.json`
- `target/e2e/phase6-trace-run1-20260710-2120.json`
- `target/e2e/phase6-trace-run2-20260710-2120.json`
- `target/e2e/phase6-trace-latest-20260710-2120.json`
- `target/e2e/phase6-e2e-summary-20260710-2120.json`

Observed:

```json
{
  "sessionId": "e2e-phase6-chat-codex-20260710-2120",
  "run1": "run-e2a97696-4398-4abc-90e4-28f45c838f92",
  "run2": "run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172",
  "chat1Success": true,
  "chat2Success": true,
  "trace1RunId": "run-e2a97696-4398-4abc-90e4-28f45c838f92",
  "trace2RunId": "run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172",
  "latestRunId": "run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172",
  "trace1Steps": 10,
  "trace2Steps": 9,
  "trace1Tools": 14,
  "trace2Tools": 8
}
```

Interpretation:

- Two Chat requests reused the same `sessionId`.
- Each Chat request returned a distinct `runId`.
- Exact trace for run1 returned run1 only.
- Exact trace for run2 returned run2 only.
- Session-only Trace latest fallback returned run2.

## Database Inspection

Tool: `scripts/query_mysql.py`

`diagnosis_run`:

```text
run_id                                   | session_id                          | status  | agent_flow | step_count | tool_call_count | has_answer
-------------------------------------------------------------------------------------------------------------------------------------------------
run-e2a97696-4398-4abc-90e4-28f45c838f92 | e2e-phase6-chat-codex-20260710-2120 | SUCCESS | CHAT       | 10         | 14              | 1
run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172 | e2e-phase6-chat-codex-20260710-2120 | SUCCESS | CHAT       | 9          | 8               | 1
```

`agent_step` grouped by `run_id`:

```text
run_id                                   | step_rows | min_step | max_step
--------------------------------------------------------------------------
run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172 | 9         | 0        | 5
run-e2a97696-4398-4abc-90e4-28f45c838f92 | 10        | 0        | 6
```

`tool_invocation` grouped by `run_id`:

```text
run_id                                   | tool_rows
----------------------------------------------------
run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172 | 8
run-e2a97696-4398-4abc-90e4-28f45c838f92 | 14
```

Mixed row check:

```text
mixed_rows
----------
0
0
```

`chat_session` metadata:

```text
session_id                          | status | message_pair_count | last_active_at
---------------------------------------------------------------------------------------
e2e-phase6-chat-codex-20260710-2120 | ACTIVE | 2                  | 2026-07-10 21:23:59
```

`GET /api/chat/session/e2e-phase6-chat-codex-20260710-2120` returned `messagePairCount=2`.

Interpretation:

- Run table has exactly two successful Chat runs for the E2E session.
- Step/tool counts match the exact Trace API responses.
- No `agent_step` or `tool_invocation` rows for this session have NULL or unexpected `run_id`.
- `chat_session` metadata confirms multi-turn context continuity at two message pairs.

## Log Inspection

Searched:

- `target/e2e/phase6-mvn-20260710-211831.out.log`
- `logs/application.log`
- `logs/chat.log`

Patterns:

- `e2e-phase6-chat-codex-20260710-2120`
- `run-e2a97696-4398-4abc-90e4-28f45c838f92`
- `run-76ce6a6e-92ab-40c9-800a-eca0c1bb5172`

Result:

- Matching startup, Chat execution, run persistence, and trace lookup log lines were present in Maven output and `logs/application.log`.

## Baseline Drift

Focused baseline command:

```powershell
mvn -q "-Dtest=DiagnosisTraceEvaluatorTest,DiagnosisEvalBaselineDiffTest" test
```

Broader baseline regression command from `mvp/eval/README.md`:

```powershell
mvn -q "-Dtest=DiagnosisTraceEvaluatorTest,DiagnosisEvalBaselineDiffTest,ExecutorGatekeeperServiceTest,VerifierInputHookTest,ChatServiceSequentialAgentTest" test
```

Result:

- Both commands passed.
- The baseline harness evaluates saved fixtures and does not depend on live DB/session tables.
- No baseline drift was observed.

## Final Gate Checks

Commands:

```powershell
rg -n "diagnosisSessionRepository\.save|new DiagnosisSession|DiagnosisSession\.builder|setAnswer\(|setSelfEvaluation\(|setFeedback\(|createFromSession|persistFinalReport\(sessionId, finalReport|save\(session\)" src\main\java\com\superbiz\agent -g "*.java"

rg -n "evaluate\(|evaluateRun\(|persistFinalReport\(|submitFeedback\(|createFromSession\(" src\main\java src\test\java -g "*.java"

git diff --check -- . ':!devflow/index.md'
openspec validate session-run-trace-isolation --strict
```

Result:

- New Chat write path calls `evaluationService.evaluateRun(...)` and writes `diagnosis_run`.
- New AIOps controller path calls `persistFinalReport(sessionId, runId, ...)` and writes `diagnosis_run`.
- Remaining `diagnosis_session` writes are legacy compatibility paths:
  - `FeedbackService.submitLegacySessionFeedback(...)`
  - `AiOpsService.persistLegacyFinalReport(...)`
  - legacy `EvaluationService.evaluate(...)`
- `git diff --check`: passed after Markdown whitespace cleanup.
- OpenSpec strict validation: passed.

## Documentation Review Follow-up

After the final documentation review, the remaining demo helper docs were aligned with the run-aware contract:

- `mvp/demo/trace-inspection-checklist.md`
- `mvp/demo/payment-timeout-acceptance.md`
- `mvp/demo/interview-walkthrough.md`
- `mvp/tables/Agent步骤表-agent_step.md`
- `mvp/tables/README.md`
- `mvp/architecture/data-model.md`
- `mvp/architecture/session-trace-lifecycle.md`

The corrections remove session-only wording for Trace/Feedback and clarify that `run_id` is the execution isolation boundary while Trace API response order is the UI display contract.

Follow-up gate after these documentation fixes:

- `node --check src\main\resources\static\app.js`: passed.
- `node --check src\main\resources\static\trace.js`: passed.
- PowerShell demo script parsing: passed.
- `openspec validate session-run-trace-isolation --strict`: passed.
- `git diff --check -- . ':!devflow/index.md'`: passed.

## Notes

- The AGENTS-required `codebase-retrieval` and LSP tools were not available in this session. Fallback verification used OpenSpec context, `rg`, targeted file reads, focused tests, E2E, DB inspection, and log inspection.