Regenerate and verify the auditable dataset release

[komaksym]

Parent

• Implement auditable UFC dataset fidelity pipeline #14 — Implement auditable UFC dataset fidelity pipeline

What to build

Produce one reproducible release that proves the complete reconciliation contract works together, including trusted outputs, evidence artifacts, documentation, and accounting checks.

Acceptance criteria

• Regeneration produces trusted all-bouts totals, decisive-only output derived from validated all-bouts, certified per-round data, merged accepted scorecards, raw observations, quarantine records, quality summary, and OCR benchmark artifacts.
• Warning fight totals remain visible; quarantined fight totals and uncertified round sets do not enter trusted analytical outputs; raw evidence remains available.
• The decisive-only output applies no second validation policy and preserves stable fight/corner identifiers.
• The quality summary reconciles raw, trusted, warning, and quarantined counts and reports check-code, ID, round, scorecard, and OCR benchmark coverage; any accounting mismatch fails release validation.
• README/data dictionary and regeneration instructions document identity semantics, fight-total versus per-round meaning, quality states, quarantine behavior, benchmark methodology, coverage, and the data-quality snapshot.
• A fixture-only smoke test runs through the public regeneration entry point, produces every expected artifact, and is byte-stable or semantically repeatable.
• The final release validation runs the targeted tests plus the repository's available lint/type/test checks and fails on schema, identifier, quarantine-accounting, or documented-row-count mismatches.

Blocked by

• Reconcile round sets against fight totals and certify per-round data #19 — Reconcile round sets against fight totals and certify per-round data.
• Attach scorecards uniquely to trusted fights #21 — Attach scorecards uniquely to trusted fights.
• Benchmark OCR transcription quality #22 — Benchmark OCR transcription quality.