Motorsports Data Audit — Council Ruling

Date: 2026-04-01 Process: Full 5-phase council (Advisory → Anonymization → Peer Review → Chairman Synthesis → Boss Ruling) Advisors: Opus, Sonnet, Gemini 3.1 Pro, Grok 4.20 Reasoning, gpt-oss-120b Winner: Sonnet (2 of 5 peer review votes — Opus endorsement) Status: PENDING BOSS RULING on open questions

COUNCIL SUMMARY

Where Advisors Agreed

1. Qualifying data is #1 missing input — grid position is 60-70% predictive in F1
2. DNF probability must be tracked per driver/constructor — rolling 20-race window, split mechanical vs crash
3. Safety car probability per circuit — historical base rate × weather adjustment (Poisson model)
4. Track type classification critical — F1: power/downforce/street; NASCAR: superspeedway/short/intermediate/road
5. Practice long-run pace is primary intelligence for race-day prediction
6. Tire degradation rates from practice — compound-specific, determines strategy trees
7. Constructor/manufacturer tiers — F1 car performance rated A/B/C tier
8. Grid penalty tracking — engine penalties, sporting penalties affect start position
9. Series-specific models required — F1, NASCAR, IndyCar, MotoGP each fundamentally different
10. Weather integration — rain creates regime changes in competitive order

Where Advisors Disagreed

1. VSC vs full SC impact: Sonnet correctly identified bookmakers misprice VSC vs SC differently. Others treated SC as monolithic. Council verdict: Model VSC, SC, and red flag separately.
2. Simulation granularity: Some proposed position-at-end, others lap-by-lap. Council verdict: Lap-by-lap for F1 (strategy matters each lap), position-at-end for NASCAR (field too large).
3. Data sources: gpt-oss proposed comprehensive dimensional model (dim/fact tables). Others used simpler schemas. Council verdict: Lean schema with series-specific tables where needed.

Strongest Arguments (from peer review)

Sonnet wins with the most analytically precise design:

• Distinguished what matters for each specific market (not one-size-fits-all)
• VSC vs SC vs Red Flag separation with different strategy implications
• Tire cliff prediction model (nonlinear degradation at compound limits)
• Constructor development trajectory tracking (mid-season upgrades)
• Team orders probability model for teammate H2H
• Practice pace fuel-correction formulas
• Superspeedway explicitly flagged as near-random (reduce sizing)

Biggest Blind Spot

No validation framework — How to backtest race simulations against historical outcomes, verify SC probability calibration, confirm DNF model accuracy. No Brier scores, no CLV tracking.

What Everyone Missed (from peer reviews)

1. Regulation changes create structural breaks — New regulations (ground effect 2022, 2026 engine regs) invalidate all historical data. Pipeline needs regulation-era flag and model reset.
2. Reverse-grid sprint qualifying — Some series use reverse-grid formats that completely change grid prediction methodology.
3. Penalty decisions are discretionary — Steward decisions (time penalties, grid drops) are subjective and unpredictable. Need to model penalty probability per incident type.
4. Driver market/contract dynamics — Contract year drivers may take more risks. Team dynamics change when a driver is leaving.

BUILD PLAN

Phase 1: Core Data Tables

motorsport_drivers: driver_id, name, series, team_id, nationality, car_number, contract_end, rookie, active motorsport_teams: team_id, name, series, manufacturer, performance_tier, reliability_rating, pit_crew_rating motorsport_tracks: track_id, name, country, track_type, length_km, turns, surface, sc_prob_dry, sc_prob_wet, red_flag_prob, overtaking_difficulty motorsport_qualifying: race_id, driver_id, session, position, time, gap_to_pole, grid_penalty motorsport_practice: race_id, session, driver_id, best_time, long_run_pace, tire_compound, fuel_corrected_pace motorsport_race_results: race_id, driver_id, grid_pos, finish_pos, status (finished/DNF_mech/DNF_crash/DSQ), laps, fastest_lap, pit_stops, tire_strategy motorsport_reliability: driver_id, team_id, season, races, mech_dnfs, crash_dnfs, dnf_rate motorsport_safety_car: race_id, type (VSC/SC/red_flag), lap_deployed, lap_ended, cause motorsport_weather: race_id, session, temp, track_temp, wind, rain_prob, conditions motorsport_championships: season, series, driver_id, points, position, title_prob

Phase 2: Custom Metrics

Phase 3: 7 Edge Scanners

Phase 4: Dashboard

• Race board, driver cards, qualifying analysis, weather center, SC/red flag tracker, championship tracker, H2H board, reliability monitor, edge alerts, P&L

OPEN QUESTIONS FOR BOSS RULING

1. Series priority: F1 + NASCAR first, or all 4 at launch?
2. Practice data sourcing: Requires live timing scraping. Worth cost?
3. VSC/SC/Red flag split modeling: Build separate impact models?
4. Tire degradation model: Build from practice long-run data each race weekend?
5. Regulation-era model resets: Auto-flag when new regulations invalidate historical data?
6. Sprint race scanner: Separate scanner worth building?
7. Historical depth: How many seasons to backfill per series?

COUNCIL METADATA