ANSWER HUB
RunLedger quickstart
Install, init, record once, promote a baseline, then replay deterministically in CI.
Direct Answer
Install RunLedger, initialize a suite, record once, promote a baseline, then replay in CI for deterministic gates.
Quick Decision
| Use RunLedger when | Consider alternatives when |
|---|---|
| You need a fast path to deterministic CI. | You need custom tooling first. |
| You can run a live record locally. | You cannot record tool outputs yet. |
| You want a default suite layout. | You already have a bespoke harness. |
5-minute path
bash
pipx install runledger
runledger init
runledger run ./evals/demo --mode record
runledger baseline promote --from runledger_out/demo/RUN_ID --to baselines/demo.json
runledger run ./evals/demo --mode replay --baseline baselines/demo.json
What to commit
evals/with suite.yaml and case files.cassettes/after redacting secrets.baselines/promoted from known-good runs.
Reminder
Do not commit secrets. Redact cassettes before pushing.
Tradeoffs
- Initial record runs hit live services.
- Baselines and cassettes must be maintained over time.
- Strict gates require some tuning on first rollout.
When NOT to use RunLedger
Skip the quickstart if you cannot record live tool outputs or if your workflow is still exploratory.