Verified Ops Starter

verified-ops-starter

**Your scheduled job says `exit 0`. Prove it did the work.**

[](https://github.com/Palo-Alto-AI-Research-Lab/verified-ops-starter/actions/workflows/selftest.yml) [](LICENSE) [](#requirements) [](#requirements)

Three small checks, no server, no dependencies. They replace a job's *self-report* with *evidence*:

check	question it answers	catches	code
freshness	is the job's real output young enough?	exit 0 while the artifact rots	verified_ops/freshness.py
wrap	did this run actually change anything?	exit 0 with no work done	verified_ops/wrap.py
rollout	is the fix on every box, read back as a fact?	"rolled out" that never landed	verified_ops/rollout.py

Everything speaks one [exit-code contract](#exit-code-contract), implemented once in `verified_ops/contract.py`, in which a dead checker cannot look like a clean result.

Why

An agent, a cron job and a nightly script all report on themselves, and the report is the first thing to break — each failure below has its own mutant in `selftest.py`:

• the model refused, the API returned an empty page, the token expired, the filter matched nothing — the process still exits 0 and stamps its heartbeat (caught by verified_ops/wrap.py);
• a checker written in Python dies on an unhandled exception and exits 1 — the same code many checkers use for "I found a problem", so the dashboard paints the corpse yellow and moves on (caught by verified_ops/contract.py);
• the fix "was rolled out" because it was sent, and an offline box, a lagging box and a healthy box are all equally quiet (cau