grounded-verifiable-gateslisted

# Make AI Output Verifiable: Grounding + Deterministic Gates + Eval Harness Never let a model's self-assessment be the decision. Force every claim to cite source text that literally matches, compute the outcome with a pure tested function over only the grounded claims, and protect that function with an eval harness that runs the real production path and fails CI on regression. ## When to use Reach for this whenever a model's output influences what happens next: a judge that approves or flags, an extractor that fills a record, a classifier that routes, an agent that decides to act. The higher the cost of a wrong, confidently-stated answer, the more this applies. Red-flag thoughts that mean STOP and apply this skill: - "The model rated it 0.9, so we'll pass it." — a self-rating is not a verdict. - "It cited a source, so it must be grounded." — citing is not matching; check the quote against the text. - "We'll eval against a simplified harness / a copy of the prompt." — then you're measuring a thing you don't ship. - "The thresholds feel about right." — feelings are not calibration. - "Precision looks great" (where the same model graded itself) — that number is circular. ## The rule 1. **Demand a verifiable citation per claim.** Require each finding to carry a quote, and verify that quote is a literal substring of its cited source after normalizing both through one shared text pipeline (same Unicode form, whitespace, casing rules). A quote that does not match is marked `u