Gist projects eight stats the box score mostly misses: deflections, steals, contested shots, loose balls, screen assists, charges drawn, box-outs and blocks. Each is a rate per 36 minutes, alongside minutes per game.
How a projection is made
- Start from the player's last three seasons (2023-24 to 2025-26), counting recent seasons more.
- Compare each season with that season's league average. The NBA's tracking counts change from year to year, so a raw count from 2019 and one from 2025 don't mean the same thing.
- Pull players with few minutes toward the average for their position. Stats that repeat reliably from year to year, like contested shots, get pulled less.
- Rookies start from past rookies at the same position. 25 players have no NBA minutes yet, and their ranges are wider.
- During the season, blend the projection with games played. Early games count for less than you might expect: a month in, last season still predicts the rest of the year better than the month does.
How often it's right
We hid 2023-24 to 2025-26 from the model, projected them, and compared. A typical miss is shown as a share of the league's average rate. Each projection also has an 80% range; on those hidden seasons the ranges held 76% to 82% of what happened.
| Stat | Typical miss | Better than repeating last season | 80% range held | Read it as |
|---|---|---|---|---|
| Contested shots | 11% | 8% | 82% | steady |
| Deflections | 16% | 9% | 78% | steady |
| Steals | 17% | 18% | 82% | steady |
| Loose balls | 17% | 23% | 76% | steady |
| Box-outs | 21% | 14% | 81% | moderate |
| Blocks | 24% | 9% | 79% | moderate |
| Screen assists | 26% | 5% | 80% | moderate |
| Charges drawn | 63% | 11% | 77% | rough |
The model beat simply repeating each player's last season on all eight stats, by 5% to 23%. Once the season starts, blending in games through late November beats using those games alone by 19% to 42%.
How it's doing this season
| Stat | Typical miss per 36 | Better than repeating last season | 80% range held |
|---|---|---|---|
| Contested shots | 1.12 | 3% | 82% |
| Deflections | 0.76 | 1% worse | 77% |
| Steals | 0.45 | 8% | 81% |
| Loose balls | 0.27 | 13% | 85% |
| Box-outs | 0.40 | Even | 78% |
| Blocks | 0.33 | 6% | 87% |
| Screen assists | 0.38 | 5% | 87% |
| Charges drawn | 0.06 | Even | 89% |
Through Nov 3: 119 players with 150+ minutes this season, updated daily. Lower misses are better; a well-sized range holds about 80%. Ranges widen for small samples so early weeks aren't judged against a full season.
What it can't tell you
- Minutes are the weakest part. Injuries, trades and role changes move minutes, and box-score history can't see them coming.
- Minutes come with a range too: about ±6 minutes for most players, a bit wider for players on a new team.
- Charges drawn barely repeat from season to season, so read that projection as a rough guide.
- Rookies at the same position share one starting projection until they play.
- A projection describes a likely rate. It isn't a verdict on a player.
Shots and shooting
Shot attempts are projected the same way as the eight stats above: league-relative, weighted toward recent seasons, regressed for players with little history. Shooting percentages lean on a prior worth dozens to hundreds of “imaginary” shots at the league rate, since one real season of makes and misses is mostly noise.
| Shot volume | Typical miss | Better than repeating last season | 80% range held |
|---|---|---|---|
| Rim attempts per 36 | 17% | 5% | 83% |
| Paint (non-rim) attempts per 36 | 23% | 6% | 78% |
| Mid-range attempts per 36 | 30% | 3% | 82% |
| Corner 3 attempts per 36 | 25% | 7% | 81% |
| Above-break 3 attempts per 36 | 21% | 2% | 81% |
| Drives per 36 | 21% | 0% | 76% |
| Catch-and-shoot attempts per 36 | 17% | 7% | 82% |
| Pull-up attempts per 36 | 25% | 2% | 79% |
| Usage rate | 11% | 3% | 80% |
| Shooting | Typical miss | Better than repeating last season | 80% range held |
|---|---|---|---|
| Rim FG% | 3.6 points | 9% | 83% |
| Mid-range FG% | 3.6 points | 21% | 87% |
| Corner 3 FG% | 4.8 points | 18% | 81% |
| Above-break 3 FG% | 2.7 points | 13% | 82% |
| Spot-up points per possession | 0.09 a possession | 14% | 81% |
Mid-range and corner-3 field-goal percentage barely repeat from season to season, so those two projections lean on the league more than any other shot in the tables above.
The fourth-quarter switch works for in-the-key attempts, in-the-key FG%, shots defended and FG% allowed vs. usual: NBA.com splits shot location and shot defense by quarter. Hustle stats like deflections and contests can’t be split: the league doesn’t track them by quarter at all, so a “fourth-quarter deflections” chart would silently show the full season instead.
During the season, for shots
A month of games moves shot volume a lot: blending games in beats sticking with the preseason projection alone by 9% to 29% (worth 100 to 500 minutes of weight, fitted per stat). It barely moves shooting percentage — the gain there is only 0% to 5%, since the shooting prior is worth far more than a month of real shots.
Early changes
“Who’s different this year?” lists a player only when his rate since opening night is further from last season’s than chance would put it (more than 3 standard errors, after 150 minutes). We tested that rule at the Nov 30 mark of 5 past seasons. The direction of a listed change held for the rest of the season 77% to 90% of the time, but only part of its size lasted, so the list shows that part:
| Stat | Direction held | Share of the change that lasted | Changes tested |
|---|---|---|---|
| Contested shots | 90% | 49% | 115 |
| Screen assists | 90% | 62% | 92 |
| Box-outs | 87% | 42% | 60 |
| Deflections | 77% | 32% | 56 |
Loose balls are left out: only 17% of an early change lasted, and 12% reversed.
Breakouts
A breakout is defined as next season +5 min/g or +3 usage points; ≤26 min/g and 20+ games both seasons. Hustle stats don't predict breakouts; age, minutes now and preseason minutes do — adding a hustle composite to a model of age, minutes and usage moves its accuracy (AUC, where 0.5 is a coin flip and 1 is perfect) from 0.700 to only 0.705, while adding preseason minutes moves it further, to 0.726.
| Predicted chance | Actually broke out | Players |
|---|---|---|
| 18% | 13% | 114 |
| 26% | 18% | 114 |
| 33% | 25% | 114 |
| 42% | 40% | 114 |
| 56% | 54% | 111 |
Bigger roles
Give a player more minutes and his per-36 rates dip a little, even once regression to the mean is accounted for. Per extra 10 minutes a game, rates fall by:
- Contested shots: about 6%
- Deflections: about 4%
- Steals: about 3%
- Loose balls: about 3%
- Box-outs: about 13%
- Blocks: about 8%
- Screen assists: about 6%
- Charges drawn: about 16%
Measured across players whose minutes changed from one season to the next; it describes the average player, not any one player.
Aggregate · balldontlie season averages (NBA tracking) for the charts (2025-26) · model history from NBA.com player tracking, 2016-17 to 2025-26 · retrieved October 20, 2026. Projections generated November 3, 2026. Prototype only; public use waits on a data licence.