When we gave the home team X%, how often did it win?
Every season · dots on the diagonal = the percentages mean what they say
From the first game of the 2026-27 season we save every estimate before tip-off, next to the sportsbook line of that moment. After the game we check it against the result and the closing line. That record only grows forward and cannot be edited after the fact.
A sportsbook price says how often a side has to win to come out even over many tries. At −110 a $110 stake wins $100, so it has to win 110 times in 210: 52.4%. At +150 a $100 stake wins $150: 100 in 250, 40.0%. That number still holds the book’s fee, which is why the over and the under add up to more than 100%.
Apps like PrizePicks and Underdog pay a multiple of the entry instead of a price on each leg. From the app’s payout table we work out the hit rate every leg needs for the entry to come out even. A 6-leg PrizePicks Flex (25x all hit, 2x one miss, 0.4x two misses) needs 54.2% on each leg, like −118 on every leg. That assumes every leg has the same chance and the legs are independent. Payout tables as the apps publish them, checked Oct 2, 2026.
The line most of the sportsbooks we follow post for a player (FanDuel, DraftKings, BetMGM). Market, next to it on the props board, is their chance for your side at that line, with their fee removed.
A book that posts a different line is left out of that chance. Before any book posts a line, the board uses his average instead and marks it AVG.
The replay uses lines built from each player’s recent average, not historical sportsbook lines, so it measures whether the percentages mean what they say, not whether they beat the market. Past accuracy does not guarantee future results. For information and entertainment only, not betting advice.
The same replay for the game page. Before each game day we projected the total, the margin and who wins with only what was known that morning, then checked the final score.
Every season · dots on the diagonal = the percentages mean what they say
Average miss in points: how far the projection was from the final score
| Season | Total | Margin | Winner | Games |
|---|---|---|---|---|
| 2026-27 | 15.7 | 11.5 | 64% | 107 |
| 2025-26 | 15.0 | 11.3 | 69% | 1,230 |
| 2024-25 | 14.6 | 10.9 | 67% | 1,230 |
| 2023-24 | 14.8 | 11.0 | 67% | 1,230 |
| 2022-23 | 14.5 | 10.1 |
From opening night 2026-27 each game’s projection is saved before tip-off and cannot change after it. After the game we grade it against the final score and against the sportsbooks’ closing lines.
| Market | Our miss | Closing line’s miss | Our Brier | Books’ Brier |
|---|---|---|---|---|
| Spread (margin) · 107 games | 11.5 | 11.0 | 0.256 | 0.248 |
| Total points · 107 games | 15.7 | 15.3 | 0.256 | 0.250 |
| Winner · 107 games | — | — | 0.212 | 0.193 |
Since Oct 20, 2026. Miss: average distance from the final score in points. Brier score: our chance vs the books’ chance with their margin removed, on their own closing lines; lower is better.
| 63% |
| 1,230 |
| 2021-22 | 14.7 | 11.1 | 65% | 1,230 |
| 2020-21 | 14.8 | 11.2 | 63% | 1,080 |
| 2019-20 | 15.3 | 10.2 | 65% | 1,059 |
| 2018-19 | 15.0 | 10.1 | 66% | 1,230 |
| 2017-18 | 14.9 | 9.8 | 65% | 1,230 |
| 2016-17 | 14.1 | 10.0 | 65% | 1,230 |
Sportsbooks’ closing lines are typically more accurate than any model, ours included: they take in injury news and lineups up to tip-off. A projected total or spread is a summary of how the two teams have done, and one that differs from the book’s number is not a recommendation. Past accuracy does not guarantee future results. Not betting advice.
Every season and stat · dots on the diagonal = the percentages mean what they say
The gap. We group the estimates by what we said (5–15%, 15–25% and so on) and check how often the player actually went over. If the 70% group went over 70% of the time, the gap is zero. Across all groups our percentages are off by 0.3 percentage points on average.
The Brier score is the standard grade for probability estimates: the average of the squared miss, the result (1 if he went over, 0 if not) minus our chance. Say we said 70%: if he went over, the miss is (1 − 0.7)² = 0.09; if not, it is 0.7² = 0.49. Lower is better; 0 would mean knowing every result in advance, and saying 50% every time scores 0.250.
Comparing rows. The score also depends on how lopsided the lines are: a blocks line a player clears 1 time in 4 is easier to grade well than a points line close to 50/50. Compare a row with 0.250, or a season with other seasons.
| Stat | Brier score | Estimates |
|---|---|---|
| Threes made | 0.169 | 1.0M |
| Blocks | 0.169 | 972k |
| Minutes | 0.182 | 1.1M |
| Rebounds | 0.188 | 1.1M |
| Rebounds + assists | 0.189 | 1.1M |
| Points + rebounds + assists | 0.190 | 1.1M |
| Assists | 0.190 | 1.0M |
| Points + rebounds | 0.190 |
The first games of a season lean on the season before, so early-season estimates hold up as well as later ones.
| 1.1M |
| Points + assists | 0.191 | 1.1M |
| Turnovers | 0.191 | 1.0M |
| Points | 0.192 | 1.1M |
| Steals + blocks | 0.192 | 1.0M |
| Steals | 0.199 | 937k |
| Season | Brier score | Estimates |
|---|---|---|
| 2026-27 | 0.181 | 117k |
| 2025-26 | 0.187 | 1.3M |
| 2024-25 | 0.187 | 1.2M |
| 2023-24 | 0.187 | 1.2M |
| 2022-23 | 0.187 | 1.2M |
| 2021-22 | 0.187 | 1.3M |
| 2020-21 | 0.187 | 1.1M |
| 2019-20 | 0.187 | 1.1M |
| 2018-19 | 0.187 | 1.3M |
| 2017-18 | 0.187 | 1.3M |
| 2016-17 | 0.186 | 1.3M |
| 2015-16 | 0.189 | 1.2M |