Build Week: generate an auditable judge evidence report

[CodeBuildder]

Why

The one-command platform demo proved correlation but left judges to translate terminal logs into a product outcome. During browser verification, it also exposed a stale-dashboard failure mode: an old Vite listener could pass an HTML homepage check while returning index.html instead of JSON for /api/*, leaving Sentinel visibly empty despite valid backend evidence.

Closes #112.

What changed

• prints a PLATFORM RESILIENCE PROOF: PASS scorecard only after Sentinel verifies the shared incident
• measures Argus evidence publication, Phoenix recovery publication, Sentinel correlation, and total verified lifecycle time
• records recovery, approval policy, OpenAI configuration status, sources, seed, and provenance
• writes versioned JSON plus reviewer-friendly Markdown for every run
• maintains latest-demo.json and latest-demo.md pointers for screenshots and submission work
• validates the report schema before Markdown rendering
• requires Sentinel dashboard /api/health to return orchestrator JSON instead of trusting the HTML homepage
• safely restarts a stale port-5175 Vite listener only when its working directory exactly matches the configured Sentinel checkout
• refuses to terminate any listener owned by another directory
• adds fixture-based report tests and stale-proxy integration guards
• documents the evidence artifacts and readiness behavior

Evidence integrity

The deterministic demo explicitly reports replayed Argus evidence and a Phoenix simulator outcome. It reports recovery as verified but availability as not measured. Timings are labeled as evidence publication and correlation timings, not live Falco detection or production MTTR. OpenAI reports configured/not configured and is not falsely described as invoked. No live Chaos Mesh fault is created.

End-to-end verification

The corrected demo replaced the verified stale Sentinel Vite process, launched the current dashboard, and passed through the browser-facing proxy:

• /api/health: Sentinel orchestrator OK, SOG connected
• 123 entities and 81 relationships
• 79 findings and 5 correlated incidents
• Argus connected with 6 findings
• Phoenix connected with 6 findings
• new verified incident: judge-demo-20260718T231343Z-46753
• verdict: PASS

Regression checks

• make test-platform-demo
• make test-agent — 91 passed
• make test-ui — production build passed
• real make demo-platform run against the argus k3s context
• browser-facing /api/health and /api/overview JSON verification
• git diff --check