What problem does it solve? Property checks run on small, hand-picked samples often produce misleading pass/fail summaries that hide sampling bias, per-group violations, and unmeasured claims. This Skill turns those spot checks into full-corpus sweeps that report exact, verifiable measurements instead of aggregate verdicts. ## Core Features & Use Cases - Full-population measurement: Re-computes invariants, monotonicity guarantees, and signed-quantity distributions across the entire corpus, always reporting the denominator alongside each figure. - Bias and coverage disclosure: Names biased case selections explicitly, marks each claimed property as measured or unmeasured, and labels any uncovered remainder instead of projecting partial results. - Decision-impact tracing: When a measured invariant invalidates a candidate result, traces whether downstream decisions, classifications, or table rows actually change. - Use Case: After fixing a token-capture bug verified on 12 hand-picked failing requests, run a sweep over the full request corpus to report the exact distribution of delta signs, verify the invariant per request group, and state which claimed properties remain unmeasured. ## Quick Start Re-measure the invariant from the last check across the full corpus and report exact figures with per-group breakdowns and any unmeasured properties.
