Your LLM checks, watched.

A dead man's switch for LLM output quality: anything that can run curl — an eval script, a CI job, the canary runner — reports pass/fail to a ping URL, and we alert on failures and on silence: a check that stops reporting is an incident too.

1. Create a check. One name, one expected period — you get a private ping URL.
2. Ping it from anywhere. curl on green, curl …/fail on red. Copy-paste snippets for cron, GitHub Actions, Python and the canary runner are waiting on the other side.
3. We keep watch. Transition-only alerts to ntfy, Slack, Telegram or signed webhooks — on failure, and on silence.

example: 90 days of one check's verdicts — green days pass, red days fail, gray days are silence

Privacy is structural: the ping API accepts verdicts, timestamps, mode, latency and score — never prompts, outputs, assertions, or API keys. Judged checks run on your side; we only referee.