Spaces:
Paused
Paused
Commit History
13 days ago
-
Deploy: week 4 — evening slate, passcode security, live ticker, ESPN photos (part 3)
Verified signature22ab885
29 days ago
-
Leaders: player photos (Tank01 ESPN headshots, nflverse fallback), sort toggles, larger full-width BPL table
AIBRUH +1: Claude Sonnet 5Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>79508e1 -
Add SPY BOY aggressive parlay series on FanDuel alternate lines
AIBRUH +1: Claude Sonnet 5Jimmy now builds a second series next to the value tickets: 3-5 legs, each needing BPL 75%+ (after a 5-pt haircut measured on 2026), FanDuel pricing it 60%+, and 80%+ of this season's games clearing it, stacked to +300 or better. Alternate ladders come from a manual Odds API pull (60 credits, never on a page load). A weekly check script measures the BPL against the latest week. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>49fd57a -
Add Big Proppa Line and put Jimmy on BPL-driven FanDuel parlays
AIBRUH +1: Claude Opus 5.5The BPL is the official line, built only from lake data and calibrated walk-forward on 2024-2025 so each line is a true 50/50 point. Leaders now show SA, next opponent rank, sportsbook line, BPL and the gap. Jimmy's LLM paths are paused (code kept) and the engine builds 2-5 leg FanDuel tickets from BPL edges, each logged once to MY BOO as a SIM order. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>5686c5c -
Add IB 2.0 pass-rush matchup score, calibrated from 2023-2025 play by play
AIBRUH +1: Claude Opus 5.5The deep dive (60.5k dropbacks, weights fit on 2023-24, checked on 2025) found IB 1.0's hand-set multipliers overstated the edge: IB CASCADE could floor a leg at 97% from the defense score alone, while IB 5 pass-yard unders hit ~54% vs the lines. Parlay IB shifts now come from ib2_bucket_rates.csv, the score is a defense-vs-offense matchup, and the contradicted check-down receptions leg is gone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>6fe7db5
30 days ago
-
Add Docker deployment for a Hugging Face Space (no Gradio)
Claude +1: Claude Sonnet 5Unverified signatureDockerfile: node stage builds the React/Vite frontend with VITE_API_BASE="" (same-origin), python stage serves it and the FastAPI backend from one process. backend/main.py mounts frontend/dist/ as static files when present (no-op in local dev where dist/ doesn't exist). requirements.txt pinned to the versions already running (fastapi 0.141.1, uvicorn 0.54.0, pandas 3.0.6) so a Space rebuild can't pull a breaking release later. Replaced the stale root README.md -- it still had frontal-lobe2/sdk:gradio frontmatter from the pre-migration repo, almost certainly the source of the Gradio compile errors. New frontmatter is sdk:docker, app_port:7860. scripts/backup_lake_to_hf.py: reusable lake backup to the private HF dataset AIBRUH/big-proppa-lake (52 CSVs uploaded and verified this session, separate from this commit). Verified locally: single uvicorn process serves / and /api/* same-origin, zero console errors, React Flow canvas and MATCHUPS both render -- same standard as every other cha…2f6d364 -
Add TOTALS! ticket, gated on certified Hot Dog games
Claude +1: Claude Sonnet 5Unverified signatureBets whichever direction (over/under) certified dogs hit more often in the backtest (under, 52.3%), priced from the lake's reference total line. Recorded and graded in the POW ledger like every other ticket -- graded against the combined final score. Verified end to end: /api/chart/ parlays/all returns a totals slip, and a synthetic grade against a real 72-point final (BUF 41, DET 31) scored over/under/push correctly. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG846a85e3 -
Hot Dog backtest: lead rate, alt-spread covers, over/under split
Claude +1: Claude Sonnet 5Unverified signatureCertified Hot Dogs (86 games, 2023-2025): lead at some point 73.3% (informational -- no cash-out price in the lake to grade it against), cover a flat +3 49.4%, cover a flat +7 69.2%, and the game goes under the closing total 52.3% of the time vs 47.7% over -- both stored on every group, not just certified dogs, for comparison. hot_dog.backtest() exposes ledRate/cover3Rate/cover7Rate and a second chosen_total_bet (under). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG83b67abc -
Wire the Hot Dog indicator into HOT DOGS!, the POW ledger and MATCHUPS
Claude +1: Claude Opus 5.5Unverified signature- HOT DOGS! now bets only certified Hot Dogs, with the backtest's chosen bet (spread) and hit rate as the leg probability. The old version read the spread sign backwards and picked favorites. - GET /api/hotdogs: this week's underdogs through the indicator. - POW ledger records HOT DOGS! legs and grades them on the final score. - MATCHUPS cards tag the underdog (CERTIFIED HOT DOG / UNDERDOG, wins N of 5) and show the last-5 comparison against the favorite. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG8c85b34b -
Add HOT DOG indicator and its pond, backtested 2023-2025
Claude +1: Claude Opus 5.5Unverified signaturebackend/hot_dog.py compares the sportsbook underdog's last 5 games with the favorite's (reaching into last season) on win %, points allowed, 3rd-down %, giveaways per drive, and the heat map's defense rank. A certified Hot Dog wins 3+ of the 5 and has a winning record. scripts/backtest_hot_dogs.py writes hot_dog_history.csv (every underdog game, 732) and hot_dog_backtest.csv. Certified dogs (86): 39.5% outright vs 31.1% for all dogs, 50.0% ATS vs 48.3%. Chosen bet: spread (50.0%). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG894d7a86 -
team_game_stats: add 3rd-down attempts/conversions and drive counts
Claude +1: Claude Opus 5.5Unverified signatureNeeded by the Hot Dog indicator (3rd-down % and giveaways per drive). Existing columns unchanged (0 diffs across 1,698 rows); league 3rd-down rate 39.3%, 11.0 drives per team-game. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG83d39122 -
Add POW ledger: Parlay Orders Won, measured every week
Claude +1: Claude Opus 5.5Unverified signaturescripts/pow_ledger.py snapshot records the engine board (every leg with its FanDuel line and odds) and each game pick before kickoff; grade --week N scores legs from the lake's box scores (no box-score row = void) and picks from schedule finals. GET /api/pow reports POW per week and season and flags any graded week under 75% as a failure; the ENGINE tab shows it. Week 3 board is on record: 6 tickets, 16 game picks. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG811e46d7 -
Add BIG PROPPA matchup predictor, modern era only (2023+)
Claude +1: Claude Opus 5.5Unverified signatureteam_game_stats now covers 2023-2026 only. backend/matchup.py scores each game on five unit matchups (scoring, red zone, ground, air, ball security) from shrunk season-to-date TeamRankings stats, and turns them into a predicted margin and win probability. scripts/fit_matchup_model.py trains on 2023, tunes and runs the redundancy test on 2024, and reports 2025 untouched: 64.2% of all games (Vegas favorite 65.5%), 88.4% (38/43) when the model is 75%+ confident. Production weights are refit on 2023-2025. GET /api/matchups (this week's slate) and GET /api/matchups/backtest. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG8e798043 -
Add team_game_stats pond: TeamRankings team stats per game, 2015-2026
Claude +1: Claude Opus 5.5Unverified signatureOne row per team per regular-season game (5,856 rows) built from nflverse play-by-play and schedules: points, margin, TDs, exact red-zone trips/TDs (drives reaching the 20), rushing, passing, sacks taken, INTs, fumbles, and team-level passes defended and tackles. Opponent-allowed stats come from the opponent's row. 2026 values match team_week exactly (512/512). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Az2f2NZQuPaekELAJsfFG88137a3c