# PIFF 3.0 All-Sports Expansion Plan ## Current State - **PIFF 3.0**: NBA only (points, rebounds) - **PIFF 2.7**: NHL (shots, assists), Soccer (shots, tackles, clearances, etc.) - **No PIFF**: NCAAB, MLB, NFL, MLS ## Data Readiness by League ### Tier 1 — Ready Now (all 3 data pillars: PPL + PGM + DVP) | League | PlayerPropLine | PlayerGameMetric | DVP | Status | |--------|---------------|-----------------|-----|--------| | **NBA** | 1.05M rows, live | 393K rows, 53 stats | 30 teams, 6 stats | ALREADY RUNNING | | **NCAAB** | 3.09M rows, live | 813K rows, 15 stats | 424 teams, 5 stats | READY — highest priority | | **NHL** | 715K rows, live | 145K rows, 21 stats | 32 teams, 4 stats | READY — upgrade 2.7→3.0 | ### Tier 2 — Ready When Season Starts | League | PlayerPropLine | PlayerGameMetric | DVP | Status | |--------|---------------|-----------------|-----|--------| | **MLB** | 375K rows (offseason) | 1.73M rows, 51 stats + Statcast | 39 teams, 10 stats | READY in Apr 2026 | | **NFL** | 214K rows (offseason) | 129K rows, 52 stats + NextGen | 12 teams (sparse) | READY in Sep 2026, needs DVP build | ### Tier 3 — No DVP, use hit-rate-only model | League | PlayerPropLine | PlayerGameMetric | DVP | Status | |--------|---------------|-----------------|-----|--------| | **EPL** | 149K rows, live | 422K rows, 29 stats | NONE | Can run without DVP | | **La Liga** | 106K rows, live | 404K rows, 29 stats | NONE | Can run without DVP | | **Bundesliga** | 86K rows, live | 306K rows, 29 stats | NONE | Can run without DVP | | **Serie A** | 70K rows, live | 347K rows, 29 stats | NONE | Can run without DVP | | **Ligue 1** | 75K rows, live | 304K rows, 29 stats | NONE | Can run without DVP | | **Champions League** | 33K rows, live | 45K rows, 29 stats | NONE | Can run without DVP | --- ## Architecture: Unified PIFF 3.0 Engine Instead of separate scripts per sport, build one parameterized engine. ### File: `scripts/piff_3_0_engine.py` ``` piff_3_0_engine.py --league nba --date 2026-02-27 piff_3_0_engine.py --league ncaab --date 2026-02-27 piff_3_0_engine.py --league nhl --date 2026-02-27 piff_3_0_engine.py --league mlb --date 2026-02-27 piff_3_0_engine.py --league epl --date 2026-02-27 ``` ### League Config (per-sport parameters) ```python LEAGUE_CONFIG = { "nba": { "stats": ["points", "rebounds"], "min_games": 15, "season_window_days": 60, "positions": ["G", "F", "C"], "dvp_enabled": True, "dvp_main_threshold": 0.70, "dvp_border_threshold": 0.60, "line_floors": {"points": 8, "rebounds": 4}, "h2h_min_games": 3, "line_movement_enabled": True, "b2b_detection": True, "max_legs_main": 8, "max_legs_border": 6, "max_per_team": 2, }, "ncaab": { "stats": ["points", "rebounds"], "min_games": 10, # smaller sample sizes in college "season_window_days": 90, # longer window, fewer games "positions": ["G", "F", "C"], "dvp_enabled": True, "dvp_main_threshold": 0.65, # slightly lower — college variance higher "dvp_border_threshold": 0.55, "line_floors": {"points": 6, "rebounds": 3}, "h2h_min_games": 2, # conference opponents may only meet 2x "line_movement_enabled": True, "b2b_detection": False, # college doesn't play B2B "max_legs_main": 8, "max_legs_border": 6, "max_per_team": 2, "ppl_integrity_filter": True, # NCAAB PPL needs integrityStatus='active' }, "nhl": { "stats": ["shots", "assists", "goals"], "min_games": 12, "season_window_days": 60, "positions": ["F", "D"], # no goalie props in PIFF "dvp_enabled": True, "dvp_main_threshold": 0.65, "dvp_border_threshold": 0.55, "line_floors": {"shots": 1.5, "assists": 0.5, "goals": 0.5}, "h2h_min_games": 2, "line_movement_enabled": True, "b2b_detection": True, "schedule_multipliers": { # NHL fatigue is more pronounced "home": 1.05, "away": 0.97, "road_b2b": 0.90, "home_b2b": 0.96, "3_in_4": 0.88, }, "max_legs_main": 8, "max_legs_border": 6, "max_per_team": 2, }, "mlb": { "stats": ["hits", "total_bases", "strikeouts", "home_runs", "pitcher_strikeouts", "earned_runs"], "min_games": 15, "season_window_days": 45, # MLB season is long, recent form matters more "positions": ["OF", "IF", "DH", "C", "SP", "RP"], "dvp_enabled": True, "dvp_main_threshold": 0.65, "dvp_border_threshold": 0.55, "line_floors": {"hits": 0.5, "total_bases": 0.5, "strikeouts": 3, "home_runs": 0.5, "pitcher_strikeouts": 3, "earned_runs": 2}, "h2h_min_games": 3, # pitcher vs batter matchups matter "line_movement_enabled": True, "b2b_detection": False, # MLB plays daily, no B2B concept "platoon_split": True, # L/R pitcher splits are critical in MLB "max_legs_main": 8, "max_legs_border": 6, "max_per_team": 2, }, "epl": { "stats": ["shots", "shots_onTarget", "tackles", "foulsDrawn"], "min_games": 8, "season_window_days": 90, # soccer has fewer games "positions": ["FW", "MF", "DF"], "dvp_enabled": False, # no DVP data for soccer "line_floors": {"shots": 0.5, "shots_onTarget": 0.5, "tackles": 0.5, "foulsDrawn": 0.5}, "h2h_min_games": 2, "line_movement_enabled": False, # no LineMovement data for soccer "b2b_detection": False, "congestion_days": 4, # fatigue if played within 4 days "congestion_mult": 0.94, "max_legs_main": 8, "max_legs_border": 6, "max_per_team": 2, "max_per_stat": 3, # diversity cap per stat type }, # la_liga, bundesliga, serie_a, ligue_1, champions_league: same as epl with league name swapped } ``` ### Core Engine Flow (same for all sports) ``` 1. Load today's games from SportsGame WHERE league=X 2. Load DVP cache (if dvp_enabled) 3. Load PlayerPropLine for today's games → book lines 4. Load LineMovement (if line_movement_enabled) 5. Detect B2B / congestion from PlayerGameMetric 6. Bulk query PlayerGameMetric for candidates (min_games, season_window) 7. For each candidate: a. Get book line (from PPL, fallback to calculated line) b. Evaluate OVER direction (szn_hr, l5_hr, gap) c. Evaluate UNDER direction (szn_hr, l5_hr, gap) d. Apply DVP boost/penalty (if enabled) e. Apply H2H boost/penalty f. Apply line movement boost/kill g. Apply schedule adjustment (B2B/congestion) h. Assign tier (T1/T2/T3) i. Compute composite edge score 8. Deduplicate: best direction per (player, stat) 9. Select main card (DVP qualified) + borderline card 10. Compute staking (RR combos) 11. Output JSON to picks/piff30_{league}_{date}.json ``` ### PPL Join for Player Names PIFF 3.0 NBA currently uses `PlayerPropLine.marketName` to extract player names via regex. For the unified engine, use the canonical join: ```sql SELECT DISTINCT pp."canonicalPlayerId", cp."displayName", pp."propType", pp."lineValue" FROM "PlayerPropLine" pp JOIN "CanonicalPlayer" cp ON pp."canonicalPlayerId" = cp.id WHERE pp.league = $1 AND pp."gameStart"::date = $2 AND pp."integrityStatus" = 'active' AND pp."periodType" = 'fullgame' ORDER BY pp."snapshotAt" DESC ``` ### DVP-Less Mode (Soccer) When `dvp_enabled: False`, skip DVP gating entirely. Instead: - All candidates go to "main card" (no borderline split) - Raise HR thresholds slightly to compensate: szn_hr >= 0.80 (vs 0.75 with DVP) - Add stat diversity cap (max_per_stat) to prevent one prop type dominating --- ## Implementation Phases ### Phase 1: NCAAB (1-2 days) - Highest impact — NCAAB has the most PPL data and DVP coverage - Copy NBA PIFF 3.0 logic, swap config params - Key differences: no B2B, longer season window, lower min_games - PPL needs `integrityStatus = 'active'` filter (NCAAB has 96% quarantined dupes) - Output: `picks/piff30_ncaab_{date}.json` ### Phase 2: NHL upgrade 2.7→3.0 (1 day) - Replace fixed lines with real PPL book lines - Add direction-agnostic evaluation (currently overs only) - Add goals as a stat (currently only shots + assists) - Keep NHL-specific schedule multipliers - Output: `picks/piff30_nhl_{date}.json` ### Phase 3: Soccer DVP-less 3.0 (1-2 days) - Replace 2.7 soccer runner with unified engine - DVP disabled, higher HR thresholds - Multi-league: EPL, La Liga, Bundesliga, Serie A, Ligue 1, CL - Congestion detection instead of B2B - Stat diversity cap - Output: `picks/piff30_{soccer_league}_{date}.json` ### Phase 4: MLB (when season starts ~Apr 2026) - Full PIFF 3.0 with DVP, strong Statcast data - Platoon splits (L/R pitcher) as unique feature - Pitcher-vs-batter H2H lookup (beyond standard H2H) - Output: `picks/piff30_mlb_{date}.json` ### Phase 5: NFL (Sep 2026) - Build DVP data pipeline first (currently only 12 teams) - Full PIFF 3.0 once season starts - Position-specific adjustments (QB, RB, WR, TE different profiles) --- ## Orchestrator Update Update `piff_3_0_all.py` to run all active leagues: ```python ACTIVE_LEAGUES = ["nba", "ncaab", "nhl", "epl", "la_liga", "bundesliga", "serie_a", "ligue_1", "champions_league"] # Add "mlb" in April, "nfl" in September for league in ACTIVE_LEAGUES: run_piff_engine(league, today) ``` ## PIFF Loader Update (Rainmaker Backend) Update `piff.ts` to load all league files: ```typescript const LEAGUES = ['nba', 'ncaab', 'nhl', 'epl', 'la_liga', 'bundesliga', 'serie_a', 'ligue_1', 'champions_league']; for (const league of LEAGUES) { const legs = loadPiffFile(`piff30_${league}_${today}.json`); // merge into map by team abbreviation } ``` --- ## Expected Impact - **NCAAB**: 424 teams with DVP data → richest prop analysis in college basketball - **NHL**: Direction-agnostic unders will be a new edge (currently overs only) - **Soccer**: 7 leagues getting data-backed props instead of pure Grok - **All sports on Rainmaker**: Every prop unlock will have PIFF data injected into Grok prompt