Model
The estimate is the league's Yes rate so far. Team-based estimates were tested on last season and none did better.
This season's record
Each day's estimate is saved before the first game and locked at tip-off; that version is the one graded. Log loss scores a chance against what happened (lower is better; always saying 50% scores 0.693).
2025-26 backtest
Can it tell games apart?
How it works
The prop. Yes if both teams score by 11:00 on the first-quarter clock. Any point counts: field goals, free throws (at exactly 11:00 too) and technical free throws before the first possession.
The data. ESPN's play-by-play for every regular-season and NBA Cup game. A team's first point is its first scoring play of the first quarter.
The estimate. The share of this season's games that hit Yes, with last season's rate counting as games, so early in the season it leans on last season. It's the same for every game.
Break-even. A chance p is worth fair odds of +100·(1−p)/p below 50% and −100·p/(1−p) above. 40% → +150 for Yes and −150 for No. A sportsbook's price has to beat that to be a good bet.
What was tested. Each team's own scoring rate, the rate its opponent allows, home/away, last season's team rates, weighting recent games more, and the correlation between the two sides. Fitted on Oct–Jan, judged on Feb–Apr. A piece stays only if it lowers holdout log loss, and the result has to beat the league rate.