Gaming & Esports 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: Opus (tied with Gemini at 1 genuine vote each; Sonnet endorsement of Opus breaks tie — deepest analysis at 38K chars) Status: PENDING BOSS RULING on open questions

COUNCIL SUMMARY

Where Advisors Agreed

1. Map veto history is #1 missing data — feeds Map Veto MC but currently lacks structured training data
2. Player stats need normalization — raw HLTV/VLR/OE data must map to unified schema per game
3. Patch version tracking with meta scores is critical — PSS model needs structured patch DB
4. Roster change history with dates and RCIS — currently no structured timeline
5. LAN vs online split must be tracked per team — OLAF needs structured data source
6. Cross-game Elo normalization is low-value — markets don't cross, player pools barely overlap
7. FIFA and CoD should be deferred — limited data, thin markets, scanner-only approach
8. CS2 and LoL have best data quality — HLTV and Oracle's Elixir are comprehensive
9. Scrim intelligence needs confidence scoring — unreliable by nature, must weight by source
10. Non-competitive markets (Game Awards, Steam) need completely different approach — no sharp book anchor

Where Advisors Disagreed

1. Schema complexity: gpt-oss proposed enterprise star schema (Dim/Fact tables). Opus proposed lean, game-specific extensions. Council verdict: Lean schema with game-specific JSON columns where needed.
2. Data acquisition philosophy: Gemini pushed for commercial feeds (GRID, Bayes). Others relied on public scraping. Council verdict: Start with public scraping, evaluate commercial feeds if latency proves to be a bottleneck.
3. Demo/replay parsing: Gemini proposed automated demo parsing as primary data moat. Others didn't prioritize. Council verdict: Investigate as Phase 2 — high potential but significant engineering overhead.
4. Scrim intelligence: Varying confidence in scrim data value. Council verdict: Track but with composite confidence score (reliability × recency × corroboration); ignore below 0.3 threshold.

Strongest Arguments (from peer review)

Opus wins with the most operationally actionable analysis:

• Explicitly deprioritized low-value ideas (cross-game Elo) with reasoning
• Game-specific analysis identifies exact missing data per source (HLTV shows vetoes on match pages but not structured feeds)
• Custom metric formulas with actual thresholds (scrim confidence composite, adaptation speed with sample-size caveat)
• Match card design with exact field layouts and example values
• Tiered approach: full cards for CS2/LoL/Valorant/Dota, simplified for FIFA/CoD
• Priority ranking by expected ROI with conviction

Biggest Blind Spot

Pipeline fragility on patch days — When publishers update games, data telemetry, replay file structures, and APIs often break. Patch Day 1 is when market inefficiencies peak AND when data pipelines are most likely to be down. No advisor addressed failover strategies or "skeleton models" for the 48-hour period when primary data sources break after a patch.

What Everyone Missed (from peer reviews)

1. Player props / SGP covariance — In-game resources (kills, gold) are finite and negatively correlated between teammates. Books price player props as independent. Covariance matrix for intra-team resource distribution = high-yield alpha.
2. Patch-aware temporal data versioning — Hero matchup matrix from patch 7.33 is poisonous on 7.34. Need "as-of" date queries, patch-boundary training splits, no future data leakage in backtests.
3. Match-fixing detection as upstream data product — Need suspicion_score on every match, not just as a kill switch. Especially in lower-tier CS2, FIFA, CoD.
4. Dead rubber/strat-hiding — Teams in locked positions intentionally lose or test strategies. Match Importance Multiplier needed.
5. Real-time streaming ingestion — Batch ETL (nightly scrapes) means always behind books with real-time feeds. Need streaming layer for post-match and live data.

BUILD PLAN

Phase 1: Core Data Tables

esports_teams: team_id, name, game, region, tier (1-3), roster_stability_days, lan_wr, online_wr, olaf_factor, active, updated_at esports_players: player_id, name, game, team_id, role, nationality, contract_status, faceit_elo, active esports_matches: match_id, tournament_id, game, team_a_id, team_b_id, format (BO1/BO3/BO5), lan_online, server_region, date, patch_version, suspicion_score esports_match_results: result_id, match_id, winner_id, map_score_a, map_score_b, total_rounds, duration_min, match_importance esports_map_results: map_id, match_id, map_number, map_name, team_a_score, team_b_score, winner_id, first_blood_team, first_tower_team, side_scores (JSON), game_specific (JSON) esports_map_vetoes: veto_id, match_id, veto_order, team_id, action (ban/pick/decider), map_name esports_rosters: roster_id, team_id, player_id, role, joined_date, left_date, change_type (permanent/stand-in/loan), rcis_score esports_patches: patch_id, game, version, release_date, depot_detected_date, severity, meta_fluidity_index, chaos_window_end, notes_parsed (JSON) esports_patch_team_impact: impact_id, patch_id, team_id, pss_score, pre_patch_wr, post_patch_wr, affected_entities (JSON) esports_player_stats: stat_id, match_id, map_id, player_id, game, kills, deaths, assists, rating, game_specific (JSON — ADR/KAST for CS2, ACS for Val, CS@15/GD@15 for LoL, GPM/XPM for Dota) esports_tournaments: tournament_id, name, game, tier (S/A/B/C), format, prize_pool, start_date, end_date, lan_online, location esports_brackets: bracket_id, tournament_id, round, match_id, seed_a, seed_b, bo_format esports_odds: odds_id, match_id, market_type, book, selection, odds, timestamp esports_scrim_intel: intel_id, team_a, team_b, source, confidence_score, date_reported, result_claimed, map, notes esports_hero_agent_meta: meta_id, game, patch_version, entity_name, pick_rate, ban_rate, win_rate, tier, updated_at esports_steam_noncomp: record_id, market_type (game_awards/steam_ranking/twitch), entity, metric_value, timestamp, notes

Phase 2: Custom Metrics

Phase 3: 8 Edge Scanners

Phase 4: Dashboard

• Match board with roster status, patch context, edge alerts
• Team drill-down with map pool heatmap, roster timeline, RCIS history
• Map veto simulator with real-time probability
• Patch tracker with chaos window countdown and team impact rankings
• Meta monitor showing hero/agent pick/ban/win rate shifts
• Scrim intelligence feed with confidence scoring
• Tournament bracket with live advancement probabilities
• Non-competitive market tracker (Game Awards, Steam)
• P&L by game, market type, edge bucket

OPEN QUESTIONS FOR BOSS RULING

1. Demo parsing investment: Build automated CS2/Valorant/Dota demo parser for proprietary metrics? High value but significant engineering.
2. Commercial data feeds: Start with GRID/Bayes Esports ($$$) or prove concept with public scraping first?
3. Player props modeling: Build covariance matrix for intra-team resource distribution?
4. Match-fixing detection: Build suspicion_score as upstream data product?
5. FIFA and CoD scope: Scanner-only (market odds comparison) or skip entirely?
6. Scrim intelligence: Worth tracking given inherent unreliability?
7. Patch-day failover: Build skeleton models for when primary data pipelines break?

COUNCIL METADATA