What this does: hand an evaluator agent any track or artist, and it returns a structured verdict JSON — scores on five criteria, deal-breaker flags, and a confidence label on every score — in one pass. For $0, no login, no API key. Honest-by-construction: no score ships without a cited source; scores without citable evidence are capped and labeled estimated or unverifiable. It beats dashboards by being composable: a passing verdict feeds the Gear Ledger as a candidate; a failing verdict ships with exactly what evidence flips it, so the pipeline never re-litigates the same track blind.
THE FIVE CRITERIA
0–5 scale per criterion · weighted total × 20 = verdict percent · bands: sign 90–100 · advance 75–89 · develop 50–74 · pass 0–49
WORKED FIRST SPIN — "ZOOTED ZONE"
—weighted total (of 5)—
CONFIDENCE RULES
Caps: estimated caps its criterion at 3 · unverifiable caps at 1 · a criterion with zero citations is unaverifiable by definition · deal-breaker flags cap the verdict band.
FILES
| first-spin.json | the rubric config — five criteria, weights, deal-breakers, confidence rules, the cwi-default taste profile |
| first-spin-schema.json | JSON Schema (draft 2020-12) — validates both the config (cwi-first-spin/v1) and verdict output (first-spin-verdict/v1) |
| verdict-template.json | empty verdict skeleton — fill, cite, ship |
| sample-verdict.json | worked First Spin on "Zooted Zone" from verified facts only |
