| name | qa |
| description | Verify laboratory experiment changes beyond "tests passed": simulate runs,
analysis outputs, trace/provenance checks, and publication-readiness evidence.
Use when: "QA this experiment", "verify this harness", "check publishability",
"capture evidence", "smoke test this analysis". Trigger: /qa.
|
| argument-hint | [experiment-path|analysis-path|report-path] |
/qa
QA in Laboratory means verifying the scientific workflow surface that changed.
make ci-smoke is necessary, but it is not always sufficient.
Route By Surface
| Surface | QA Path |
|---|
| Experiment harness | Run deterministic simulate mode, inspect CSV schema, confirm no API keys are required. |
| Live harness path | Check explicit --live, provider preflight, budget cap, model/config logging, and trace output. |
| Analysis script | Run against fixture or committed data; verify output tables/charts match the data source. |
| Publication artifacts | Confirm source data is live, trace-backed, and not mixed with simulated rows. |
| Spellbook harness/docs | Check bridge links, repo brief, skill markers, and make ci-smoke. |
| Website/content | Run content sync/build if the touched files affect the site. |
Procedure
- Read
.spellbook/repo-brief.md, AGENTS.md, and the touched experiment docs.
- Identify the exact behavior the change claims to support.
- Run the smallest direct command that exercises that behavior.
- Inspect the generated artifact, not only the exit code.
- For data/report work, check
mode or equivalent provenance before accepting
any publishable claim.
- Record residual unverified paths explicitly.
Simulation Integrity
Blocking rule: simulated rows are harness QA only. If a report, paper, blog post,
executive summary, social thread, or data card presents simulated rows as live
evidence, QA fails.
Report
Return:
- verdict:
PASS, FAIL, or UNVERIFIED
- commands run
- artifacts inspected
- provenance/live-data status
- remaining publication or runtime risk
