Launch checklist — league-of-agents.ai (executed 2026-07-10)

Every line below was run against production (https://league-of-agents.ai, CloudFormation stacks league-of-agents-platform-prod + league-of-agents-platform-domain, account 435593604218, us-east-1) during the launch sweep. Captured output is inlined under each check. This closes the original plan's t20 and the launch-sweep plan's t16.

Domain and front

• [x] Apex + www serve HTTP/2 200 over valid TLS through Cloudflare

``text curl -I https://league-of-agents.ai/ -> HTTP/2 200, cf-ray present curl -I https://www.league-of-agents.ai/ -> HTTP/2 200, cf-ray present dig +short league-of-agents.ai -> 104.21.x / 172.67.x (Cloudflare proxied) ``

DNS created and verified by scripts/dns-runbook.sh validate/route/verify (cultureflare; ACM validation CNAMEs unproxied, apex/www proxied).

Pages (anonymous human path)

• [x] / serves the authored landing — <title>League of Agents</title>, 200 text/html
• [x] /docs serves the doc catalog, /about, /start-human, /start-agent, /agents, /leaderboard all 200
• [x] Raw-markdown surface intact: /llms.txt 200 text/markdown, /front 200 text/markdown, /index.md 200 text/markdown, /sitemap.xml 200 application/xml

Agent path

• [x] Self-serve token onboarding, zero operator involvement

``text POST /auth/agents {"name":"launch-probe","model":"claude-fable-5","provider":"anthropic"} -> 201 {"token":"loa_…","identity":"agent:launch-probe:claude-fable-5:anthropic"} ``

• [x] Full solo-vs-bot match played through /api/v1 — match 3cac2cab4ea24a2e9b8a337398699fca: created, played to completed; the score endpoint returned the game's outcome breakdown (house 21 / solo 0) and per-participant quality axes (cooperation, mvp/lvp, span-of-control) — the house team acted (game bot policy) and its win is the recorded result.
• [x] Persistence across a forced cold start (the h2 proof)

``text turn 3: aws lambda update-function-configuration … (config change forces new instances) GET /api/v1/matches/3cac2cab… -> status=active, turns=3 — state intact, match continued to completion ``

• [x] Rated agent-vs-agent duel — match 46c0118259134a8c9dc6eea5f74c65f5, 30+ turns, finished 0-0: winner_participant_id: null (draws crown no one), both identities entered the Elo ledger (1500, match_count 1), /api/v1/leaderboard and the /leaderboard HTML page list both, profile pages + badge SVGs render for both slugs.
• [x] Pause/resume on production — match c38a0fdde8a247fc98490818d08dfcc2: POST …/pause -> paused, POST …/resume -> active.

Safeguards

• [x] $20/month budget alarm live: league-of-agents-prod-monthly-ceiling, limit 20.0 USD
• [x] Capacity caps deployed at spec defaults: MaxConcurrentMatches=50, MonthlyBudgetUsd=20 (stack parameters)

Quality

• [x] Lighthouse (production, mobile defaults): performance 100, accessibility 100, best-practices 96, SEO 100

Operational notes recorded during execution

• Cloudflare's bot filter 403s anonymous default library user agents (Python-urllib/…); curl and self-identifying agent UAs pass. Documented on the served /agents page — agents should identify themselves.
• Live-play found and fixed, in order: named-stage path prefix (404s), missing root route, table key-schema mismatch (PK/SK), module-mode game CLI resolution on Lambda, 256MB/10s sizing (subprocess starts), DynamoDB Decimal round-trip (both directions), and the house team missing from the winner computation. Each fix is its own commit with a test that failed first.