Daily Picks, Every Sport
Picks drop every morning. Game lines, spreads, totals, player props, and DFS lineups across MLB, NBA, NHL, and NFL. No edit queue, no human scrambling — the engine runs at 6 AM, the slate is on the site by breakfast.
Every Pick Tracked, Publicly
Wins and losses live on the site. Hit rates by sport, ROI by pick type, streaks good and bad. Nothing deleted, nothing back-posted. The public track record is the product — everything we're wrong about stays exactly where you can see it.
A Subscription Built for Sports Fans
A membership that ships the full slate, the DFS lineups, and the parlay constructions. No upsells once you're in. The free tier shows enough of the engine's output that you can watch it work before you pay.
The Degenerate of Data Blog
Our in-house writer — a Hunter S. Thompson tribute act with access to the engine's internals — drops long-form pieces on the weirdest correlations the models are chewing on this week. Sports, data, and controlled derangement. Automated where it can be, voiced where it matters.
The ugly math behind the pretty picks.
- ELO + Glicko + Bayesian Rating Blend
Team strength comes from a blend of ELO-style hierarchical Bayesian ratings, Glicko-2 for volatility weighting, and Pythagorean expectation. An ensemble layer learns the optimal mix per sport from ten-plus seasons of historical games. Cold-start teams don't poison the signal.
- Transformer Attention for Player Form
A transformer-based sequence encoder builds per-player embeddings from recent game logs — modeling the shape of a player's last stretch as a learned representation rather than a naïve rolling average. Those embeddings feed both the game-level predictions and the player props pipeline.
- Monte Carlo + Bayesian Logit Fusion
Every matchup simulates 10,000 times with sport-appropriate distributions — Dixon-Coles low-score correction for NHL and MLB, Gaussian for NBA and NFL. Market probabilities, ML probabilities, and situational signals merge via a Bayesian update in logit space. Isotonic regression recalibrates on held-out games before anything ships.
- Kelly Sizing With Conformal Bounds
Bet sizing runs on fractional Kelly — quarter-Kelly, capped at tail thresholds — guarded by 90%-coverage conformal prediction intervals. Model uncertainty is sampled before sizing, not bolted on afterward. The engine bets smaller when it's less sure, which is most of the time.
