Hackathon E2E test results — 2026-05-11

[slavazeph-coder]

Hackathon E2E results — 2026-05-11

Production target: https://www.brainsnn.com (Chromium resolves https://brainsnn.com to the lowercase www host).
Final main commit tested: caa85e6.
Railway deployment: fe571686-c006-4fbe-ab59-c0dab24694db — SUCCESS.

#	Test	Status	Actual output
1	Build + boot	PASS	npm run build transformed 914 modules; production browser booted; brain tick advanced 3 -> 5; in-browser MCP catalog has 28 tools.
2	Best Gemini text analysis	PASS	Gemini status: gemini-2.5-flash online; golden prompt result source gemini:gemini-2.5-flash; scores: emotional activation 95%, cognitive suppression 90%, manipulation pressure 98%, trust erosion 92%.
3	Best Gemini multimodal	PASS	Uploaded warning-sign.png; media result rendered with evidence traces.
4	Veea prompt injection blocked	PASS	Tester returned BLOCK with prompt_injection and secret_leak: anthropic_key.
5	Veea PII redaction	PASS	Redaction path returned [email] and [phone]; toggle off returned ALLOW; toggle back on returned REDACT.
6	MCP tool gating	PASS	reset_brain returned Lobster Trap destructive-tool block, then returned { reset: true } after allowToolDestructive=true.
7	Red Team → Brain Evolve	PASS	Baseline F1 0.82; best evolved candidate F1 0.851; post-promote F1 0.85.
8	Multimodal RAG + Vector-Graph Fusion	PASS	Indexed two-doc bulletin/reminder corpus; query retrieved Phishing Bulletin; fused search returned results after graph-weight adjustment.
9	MCP from outside	PASS	node mcp-server/server.js stdio smoke: tools/list returned 28 tools; lobster_trap_inspect returned block; test_hypothesis returned Supported.
10	Connected Systems / Session Rooms	PASS	Created room and submitted score as codex-a; second browser opened ?room=<code> and saw the leaderboard entry.
11	Onboarding copy + README	PASS	Fresh onboarding mentions Google Gemini 2.5 (with Gemma 4 as a fallback) and Veea Lobster Trap; README contains Gemini + Lobster architecture/env rows.
12	Snapshot report engine attribution	PASS	generateReport(..., geminiResult) emits ## Gemini (gemini-2.5-flash) Deep Analysis, not Gemma.

Fixes merged during E2E

• Fix E2E test 9 MCP stdio local tools #33 fixed Test 9 (mcp-server invalid node:ws import + local pure tool handling).
• Fix E2E test 2 Gemini env in Railway Docker builds #36 fixed Test 2/3 Railway Docker build-time VITE_* env injection.
• Fix E2E test 10 session room share links #37 fixed Test 10 ?room=<code> shared-room loading.
• Fix E2E test 2 Gemini JSON truncation #39 fixed Test 2 Gemini JSON truncation by raising default output tokens to 2048.

Remaining manual deliverables

• Demo video recording/upload.
• lablab.ai submission form.
• Submission receipt screenshot and release tag after form submission.