Every test we run — including the ones that failed. Most handicappers show you a 16-0 system and hide the math. We publish the method, the result, and the losers. If an idea doesn't beat the vig on our own data, we say so.
We tested 136 mechanism angles (blowout spots, streaks, road trips, line moves) across 32,000+ games. With an honest train/test split, zero held up out-of-sample. The market already prices them.
A real win-probability model (Negative-Binomial, 30k games) that, at the market line, lands at ~50/50 — so it passes when there's no edge instead of forcing a bet. No fake totals plays.
Home-run "unders" hit ~89% — and still lose money (−1.2%) because the juice is brutal. We flag base-rate markets and grade on real prices only, never a fabricated line.
Does a cold streak make a hitter "due" for a homer? We checked: P(HR) actually falls as the drought grows (14% → 8%). So we post the true probability and never bump for "due."
Real, odds-aware ROI for the angles the betting world sells — computed from every MLB closing line 2010–2025. Spoiler: static angles lose to the vig.
Loading measured results…
We keep a running record of every backtest — hypothesis, method, result, and caveats — failures included. The highlights above are drawn from it; the complete ledger (with the scripts behind each) lives in the repo and grows with every test.