claim-scorecardlisted

# Claim Scorecard Use this skill to turn an advisor's content stream into an accountability ledger. Predictions are cheap because nobody re-checks them; this skill records every checkable claim with its resolution window, expires it when the window closes, adjudicates it against independent evidence, and renders a hit rate the advisor never published. The skill has two layers: 1. **Extraction discipline** — what enters the ledger, governed by `references/claim-extraction-rubric.md`. 2. **Ledger mechanics** — `scripts/claims_ledger.py` persists, expires, adjudicates, and renders. Claude extracts; the script persists. ## Definition Of Done - Every checkable prediction in the source text is in the ledger with `claim`, `source`, `mechanism` (the advisor's stated reasoning, quoted), `confidence` tier, and a `window` where one exists. - Every unfalsifiable claim encountered is in the ledger as `excluded` with an `exclusion_reason` — nothing silently dropped. - Re-running `record` on the same claims file appends 0 rows (dedup by id). - `check` has been run; every past-window claim is `expired`, no stale `open` rows. - Every adjudication note names the outcome and carries a source URL **independent of the advisor**. - `scorecard` rendered; hit rate stated with numerator/denominator, never a bare percentage. ## Core Principle Only falsifiable claims enter the ledger; a claim with no checkable outcome or window is logged as `excluded` with the reason, not silently dropped. The