Releases: tonydzi/verified-ops-starter
Release list
v0.1.0 — exit 0 is not evidence
Your scheduled job says exit 0. Prove it did the work. First tagged release.
Three small checks, no server, no dependencies, no account. They replace a job's self-report with evidence.
What is in v0.1.0
| check | question it answers | catches |
|---|---|---|
freshness |
is the job's real output young enough? | exit 0 while the artifact rots |
wrap |
did this run actually change anything? | exit 0 with no work done |
rollout |
is the fix on every box, read back as a fact? | "rolled out" that never landed |
All three speak one exit-code contract, implemented once in verified_ops/contract.py, designed so that a dead checker cannot look like a clean result — the specific trap being that exit 1 means "I found a problem" in most checkers and is what Python prints when the checker itself dies, so a crash gets painted the colour of a routine finding.
Why these three
Each is a failure that happened to us first, on a live agent fleet:
- a nightly artifact cold for two weeks behind a green heartbeat — the job fired every night and produced nothing;
- a checker crashing every night while the board showed a routine finding;
- a keyboard fix that sat undelivered on one box for twelve days after being "rolled out", because an offline box, a lagging box and a healthy box are all equally quiet.
The rule it lives by
A check with no mutant that makes it fail measures nothing. Every state in selftest.py is produced by planting the actual failure — a backdated file, an empty rewrite, a verify that only exits 0, a job that no-ops, a checker that crashes — and demanding both the right state and the right exit code. python3 selftest.py prints 44 checks, 0 failed.
Seven holes in this release were found by two external review rails before publication, not after.
Requirements
Python 3.8+, standard library only. CI: Ubuntu / macOS / Windows × 3.9 / 3.11 / 3.13.
git clone https://github.com/Palo-Alto-AI-Research-Lab/verified-ops-starter
cd verified-ops-starter && python3 selftest.pyScope, stated plainly
This sits next to heartbeat monitoring (healthchecks.io, cronitor, dead-man switches), which answers did it fire? — a different and genuinely useful question. This kit answers did the work land?
What's next
A parity check that catches a box drifting silently over days, read-back adapters beyond the shell (HTTP, SQL), and a human-readable report. Not planned, deliberately: a server, a daemon, a dashboard, an account.
Every noticeable change ships as its own release, so this feed is the record of what the kit can actually prove.
Full Changelog: https://github.com/Palo-Alto-AI-Research-Lab/verified-ops-starter/commits/v0.1.0