integrity-audit

# Integrity Audit Skill This skill codifies the "Brutal Truth" audits to ensure the production database and pipeline stay healthy, particularly focusing on identity mapping and data quality. ## When to Use Activate this skill when: - Running a pre-deployment data check. - Diagnosing "zero value" bugs in waiver or lineup recommendations. - Verifying the success of a major data ingestion cycle (Savant, BDL, Yahoo). - Troubleshooting recurring `TypeError` (timezone) or `AttributeError`. ## Core Workflow ### 1. Identity Layer Audit Verify the completeness of the player identity cross-reference table: ```sql SELECT COUNT(*) as total_players, COUNT(yahoo_id) as mapped_yahoo, COUNT(mlbam_id) as mapped_mlbam, COUNT(bdl_id) as mapped_bdl FROM player_id_mapping; ``` ### 2. Contamination Audit Detect placeholder or "stub" data that poisons the z-score pool: ```sql -- Count default batter rows SELECT COUNT(*) FROM player_projections WHERE hr=15 AND avg=0.25 AND rbi=65; -- Count default pitcher rows SELECT COUNT(*) FROM player_projections WHERE era=4.0 AND whip=1.3 AND w=0; ``` ### 3. Pipeline Freshness Audit Identify stalled or failing background jobs: ```sql SELECT job_name, status, started_at, completed_at, error_message FROM data_ingestion_logs ORDER BY started_at DESC LIMIT 20; ``` ### 4. Integrity Scalars Verification Audit the `cat_scores` table for logical consistency: - Batter `avg` should be between 0.0 and 1.0. - Pitcher `era` should be between 0.0 and 20.0. - `z_score_total` should not be NULL for active players. ## Guidelines - **Brutal Truth:** Report the exact failure rate. Do not sugarcoat "mostly working" data. - **Timezone Safety:** Always check for `offset-naive` vs `offset-aware` mismatches in datetime comparisons. - **Fail-Fast:** If contamination > 5% of the pool, recommend a `HARD GATE` (Kelly multiplier = 0.0x). ## Examples "Run a brutal truth audit on the current production state." -> *Action:* Execute the 4-step audit sequence and update Section 16.5 of `HANDOFF.md`.