← ClaudeAtlas

verify-with-andonlisted

After producing or editing any data analysis artifact (report, Excel file, summary table), write an andon verification spec and run it. Use whenever numbers are derived from data and are about to be presented as results.
gulmezeren2-byte/andon · ★ 0 · Data & Documents · score 72
Install: claude install-skill gulmezeren2-byte/andon
# Verify analysis with andon You have just produced (or edited) an analysis artifact: a report workbook, a summary table, a set of claimed figures. Before presenting it, verify it with andon. The verification is deterministic arithmetic — your own confidence in the numbers is not a substitute for it. ## The discipline 1. **Write the spec while the derivation is fresh.** For every number you put in the artifact, you know which source rows and filters produced it. Encode exactly that as a check. If you cannot write the check, you do not actually know where the number came from — treat that as a defect in the analysis, not in the spec. 2. **Cover four layers, in this order:** - `schema.*` — the assumptions about the source data (unique keys, allowed values, date continuity); - `reconcile.*` — every headline figure against the source, with the same filters you used to derive it; - `internal.*` — totals vs parts, percentages vs 100, derived columns vs their formulas; - `excel.integrity` — if the artifact is a workbook. 3. **Run and read:** ``` andon run andon.yaml --json ``` Exit 0 = all pass. Exit 1 = at least one FAIL: the artifact is wrong or the spec is — find out which before touching anything else. Exit 2 = REVIEW flags: report them to the user verbatim; do not resolve them yourself. Exit 3 = nothing was verified (a check could not run, or every check was skipped): fix the spec or say plainly that the claim i