Initial review-ready release of the Verity LLM evaluation framework. Includes hermetic Tier 1 replay, semantic and adversarial evaluation tiers, seeded-defect evidence, judge-calibration methodology, cost tracking, and GitHub Pages report publishing.\n\nAudit: XR-02