Prove how every pair of minimal cross-grid criterion repairs differs

Extends the robust-repair consequence audit with exact pairwise distinction certificates. Every unordered pair of returned inclusion-minimal cross-grid repairs receives the first deterministic sampled witness on an evaluable grid where their newly admitted sample sets differ, plus exact left- and right-exclusive observation counts. SHA-256 consequence fingerprints are display and comparison aids only; exact sample-ID sets decide distinctions. It does not rank repairs or apply criterion weights. Results are exact for returned samples, not continuous-time proofs, causal claims, quality scores, recommendations, interpretations, predictions, probabilities, or statistical confidence. No licensed catalog, new dependency, or request-time network source is used. One standard quota unit per admitted request; failed or timed-out admitted attempts are not refunded. Do not automatically retry. Historical mode uses an explicit BCE/CE date object and UT1 conventions. Its response is the versioned historical envelope, not the modern UTC envelope. Read result for calculation fields and metadata for limits and uncertainty.