Monday, October 19
Model win probabilities for every game on today's slate, against the market. Tap a game for details. Slate 2026-10-19 (ET).
null -3.2% · market fav -3.3%
Kalshi 2025-26 · close · hindsight
negative = model better · n=1209
close better by 2.9 SE
written before tip, frozen after
How these numbers are made
Model is the composite's P(win); market is the price with its margin removed: Kalshi (the exchange, its taker fee included in the price; no-vig from the bid/ask midpoints), or the sportsbook when Kalshi has no price (tagged with the book). The lean is the side the model gives ≥ 50%. Model − market is the lean's model probability minus its no-vig market probability; model EV is what the model claims the posted price is worth, shown beside its null: the EV if the market is exactly right, about minus the hold, not zero. Both are model estimates, not verified edges; the Ledger and Calibration pages track whether they hold up.
Composite Δ is 100 × the standardised four-factor gap (ridge weights, home − away, in win-% points). early · carryover: a team has fewer than 10 games, so last season's log (×0.25) is carried in with its own early-season logit. injury report: from game 10 the latest NBA injury report is folded in (Out/Doubtful players, weighted by last season's BPM). The games-10+ logits also grow the rating's weight through the season, discount opponents' 3-point shooting (mostly luck), add roster talent (minutes share × last season's BPM) and each team's free-throw percentage; see Model. Prices are refreshed each build until tip and frozen after. A game abstains only while a team has no games yet.