Verify it yourself
Every row is a claim this README makes, paired with the exact command or URL
that proves or disproves it. Repository commands run from a clean checkout after
npm ci. The same rule holds for every figure in this README, not just this
table: each one is either generated from the deployment (a constant named in
its source file, a count with the command that produces it) or paired with the
command that reproduces it.
| Claim | How to check it |
|---|---|
| The migration, table, and seeded-table counts are real | Run the three commands stamped at the top of Limitations; each prints the number quoted beside it |
| The test suite is the size claimed | npx vitest list --run | wc -l. Limitations stamps this count for the commit the file ships in — run it against your checkout and it should match exactly |
| The whole gate passes three consecutive isolated times | npm run release:verify:repeat — writes one attempt per pass to artifacts/release-gate/<sha>/manifest.json and only sets "result": "passed" after three |
| CI runs that same gate, not a lighter one | .github/workflows/check.yml and .github/workflows/repeat-release-gate.yml |
| Calendar invites actually land in a real calendar | app/lib/comms/ics.ts:5 records the end-to-end verification against Gmail on 2026-08-08: native RSVP card, event auto-staged, conflict detection ran |
| AI triage never decides anything | grep -nE '\.(insert|update|delete)\(' app/lib/review/ai-triage.server.ts — the only writes target the ai_triage table: no reviews row, no session status change |
| The deployed demo is up, seeded, and signing in | npm run smoke:demo -- https://demo.callboardhq.com |
| The API is real, documented, and machine-readable | https://demo.callboardhq.com/v1/openapi.json and https://demo.callboardhq.com/developers |
| An agent can orient itself without being told how | /llms.txt on any deployment (app/routes/llms.txt.ts) — the judge path in order, each brief feature's primary URL, and the API strip, as plain text |
| The public schedule works with JavaScript disabled | Turn JS off and load https://demo.callboardhq.com/e/frontier-ai-summit-2026/schedule — search, day tabs, and the track/format/room facets are server-rendered |
Measured latency
Measured against the live demo on 2026-08-12 from a single US-Pacific client. One warm-up request per URL is discarded, then seven requests one second apart; the figure is the median of the successful responses:
curl -s -o /dev/null -w "%{time_total}" https://demo.callboardhq.com/<path>
| URL | Median (warm) |
|---|---|
/ |
190 ms |
/demo |
86 ms |
/admin (unauthenticated — a 302 to /login, not a dashboard render) |
94 ms |
/developers |
224 ms |
/e/frontier-ai-summit-2026/schedule |
390 ms |
/e/frontier-ai-summit-2026/schedule/<sessionId> |
305 ms |
/e/frontier-ai-summit-2026/speakers/<personId> |
305 ms |
/v1/openapi.json |
97 ms |
These are measurements from one client on one evening, not guarantees: they include network transit, they move by tens of milliseconds run to run, and a different region will see different numbers.
One thing they surfaced, recorded here rather than left for a judge to hit cold: under repeated requests the two heaviest public pages intermittently returned Cloudflare error 1102 (Worker exceeded resource limits) — 2 of 7 requests to the schedule and 3 of 7 to a session detail page. Those responses are excluded from the medians above. Pacing the requests did not remove them, so this is Worker CPU on the render path, not edge rate limiting.