Singular

Method

How the ratings are built, how they are tested, and how the data are checked. For a plain-language walk-through with examples, see How it works.

The rating: SiPM

SiPM rates a player's effect on his team's scoring margin, in points per 100 possessions. Zero is the league's average player that season, weighted by minutes played. Offense plus defense equals the net rating.

It has two parts. The first is a lineup rating: a regularized plus-minus over every lineup of the season, adjusted for shooting luck and pulled toward his box-score rating. The second is the box-score rating itself, adjusted for the strength of the opponents. A final layer combines the two, with weights fitted on how well they predict the next season's lineup results, and adds an age curve that differs by player quality.

As played and forecast. From 1996-97, a season's rating on player and season pages is his rating as played: that season's lineup rating, the first part above, with no step forward in age. SiPM at the end of a season is a forecast for the next season, so it includes the expected change with age; season tables show it beside the as-played figure. The rating at the top of a current player's page is the forecast for the rest of this season.

Projections

This season. The rating for this season is the season model's forecast for 2026-27, updated by his games so far. After each game, a game-level rating (a fixed map from one game's box line to SiPM's offense and defense) updates a running estimate, and the forecast moves with it. Rookies start at the level rookies play at in their first season, which is lower than where they settle after it.

Five-season outlook. The mean rating over 2026-27 to 2030-31, with 10th and 90th percentiles, if he plays. It reflects age, current level and how players like him have moved. It averages two forecasts, half each: one follows players with his rating, age and minutes; the other forecasts each part of his game separately (shooting, passing, rebounding, turnovers, rim protection, minutes) and rebuilds the rating from the parts. Players without a rating last season (rookies, returners) keep the first forecast alone.

Best season ahead. His expected rating in his best later season (2027-28 to 2030-31), the age he is likely to reach it, and the chance that some later season beats this one. A typical path can be flat while the best season still lies ahead: careers zigzag around the trend.

Projected stats. Usage and minutes come from a trajectory model that includes a role-change block: who left his team, who arrived, and whether the top option is gone. Shooting and per-75 stats come from a short-term box-score forecast that weights recent games more.

Career and Legacy

Career (prime-based) puts a whole career on the rating's scale. It starts from his prime, his minutes-weighted rating at ages 24-31, with his playoff series counting for a quarter of it, and adds small adjustments for his minutes across the prime and for his years before 24 and from 32 on.

Legacy (with playoffs) is Career plus credit for playoff success, counted in championship odds: what each series did to his team's chance of a title.

Before 1996-97

Lineup data start in 1996-97. Earlier seasons are rated from the box score and, from 1953-54, how his teams did with and without him. The all-time table and player pages also add an estimated lineup part to those seasons; it is cut back on purpose, and figures that include it carry an asterisk and an 80% range. Before 1973-74 the NBA did not record steals, blocks or the split between offensive and defensive rebounds, so each player's are estimated, with ranges.

How it is tested

Each test scores forecasts on seasons or dates the model never saw when it was fitted.

Team test
0.82
correlation with next season's team net rating, given the minutes actually played, 2008-09 to 2025-26 (540 team-seasons); last season's own net rating gets 0.59
Calibration
0.98
slope: a point of forecast is a point of margin (90% interval 0.93 to 1.03)

Against public models. Ratings frozen on January 6, 2026, using only games played before it, were scored on the lineups and margins of the rest of that regular season. Singular's ratings did at least as well as those of DARKO and EPM, two widely used public models. The test has limits: it covers one season, its main scoring fits each model's scale to the games being scored, and the current version of the model was adopted after the test had been run.

Data checks

Every source is audited for errors before use. The audit has traced 76 root causes: 52 in the providers' data and 24 in our own processing. 58 are fixed, 9 excluded, 9 documented, and 0 open. Full audit.

Providers sometimes change data without notice.

Appendix: defense before blocks were recorded

The NBA did not record blocks or steals before 1973-74. Season ratings for that era use estimates of both (see Before 1996-97), so they see rim protection only roughly.

Team results means how his teams did in games with him and without him. Before 1996-97, each season rating adds part of this term to his box-score rating. Single seasons use half of the defensive term, because one season of it is noisy.

A career pools many seasons, so the noise averages out. Career defense uses the full term: box-score defense plus team-results defense, averaged over his career. Its box-score part uses only the stats that were recorded, so before 1973-74 it has no blocks or steals. The table ranks 811 players with 15,000+ minutes, 1953-54 to 2025-26. Higher is better.

#PlayerPlayedCareer defenseper 100Before 1973-74
1Bill Russell1956-57 to 1968-69+3.59Wholly
2Ben Wallace1996-97 to 2011-12+3.46–
3Tree Rollins1977-78 to 1994-95+3.13–
4Mark Eaton1982-83 to 1992-93+3.04–
5Dikembe Mutombo1991-92 to 2008-09+2.85–
6David Robinson1989-90 to 2002-03+2.71–
7Rudy Gobert2013-14 to 2025-26+2.68–
8Nate McMillan1986-87 to 1997-98+2.63–
9George Johnson1972-73 to 1985-86+2.51Partly
10Dennis Rodman1986-87 to 1999-00+2.49–
11Jon Koncak1985-86 to 1995-96+2.40–
12Jack Sikma1977-78 to 1990-91+2.30–
13Wes Unseld1968-69 to 1980-81+2.28Partly
14Hakeem Olajuwon1984-85 to 2001-02+2.26–
15Nate Thurmond1963-64 to 1976-77+2.24Partly

Flagged players (Bill Russell, George Johnson, Wes Unseld, Nate Thurmond) played all or part of their careers before 1973-74.

Before 1973-74: Wholly = every season he played came before it; Partly = some did. Career defense is in points per 100 possessions.

What zero means

The model's own zero sits near the average of everyone who played, each counted once. Good players play more, so the average minute played rates higher than that. Each season's minutes-weighted average is therefore subtracted from every published rating: 0.87 for 2025-26, and this season uses the same figure until it ends. Ranks within a season do not move.

With this zero, an average team's minutes-weighted sum of player ratings comes to about 0, like its net rating. EPM, BPM, DARKO and LEBRON use the same zero.

The lineup rating

Each season's lineups are fitted on their own. The pull is toward his box-score rating plus three quarters of the gap between his lineup and box-score ratings in his last rated season (less if that season was longer ago), so a gap that repeats builds up. The pull counts as much as 8,000 possessions of lineup data.

Threes and free throws count at the shooter's expected percentage rather than the result, so a hot or cold run of shots does not move the rating.

Rookies

We estimate a rookie's first-season level from college stats, draft slot and age. Players with no college or international stats on file get draft slot and age only. From the second season on, the outlook uses the level expected after the first year.

How the outlook was tested

Tested on 11,729 season forecasts one to four years ahead, made in 2013-14 to 2021-22 and fitted only on earlier seasons, the average's typical error is 1.41 points per 100, against 1.44 for the first forecast alone and 1.43 for the second; its 80% range held 83% of outcomes. For the five-season mean the error is 0.91 (first forecast alone 0.94). It mainly helps young stars (25 or younger, rated +3.1 or better): the first forecast alone came in 0.4 points per 100 too low for them on average, the average 0.2.

Best season ahead

All three are fitted on what happened next to players with the same rating, age and minutes since 2007-08. The chance of a better season also uses a term for star quality. All three are averaged, half each, with the part-by-part forecast described above. The averaged chance of a better season ran high for players rated +6.1 or better: in the test seasons it said 44% on average, and 37% of them did have a better later season. It is recalibrated against those outcomes; tested on seasons left out of the fit, it says 39%.

Career, prime-based

Career starts from the prime: his minutes-weighted rating at ages 24-31. Three small adjustments sit on top, each a share of the prime. Load: minutes per season across the prime, measured against the star minutes of his era (from 10% off to 10% on). Early years (before 24) and late years (32 on): how he did against players with a similar prime and height at those ages. Ages before he was allowed into the league do not count. Seasons not yet played fill the prime at a cautious forecast (16th percentile); they can lower the early and late adjustments, never raise them. Playoffs count for a quarter of the prime: his playoff series at ages 24-31, each measured against the average player's playoff drop in that round, shrunk toward his regular-season prime by how many playoff minutes he has. Deep runs can lift him more than early ones; a poor late-round series never pulls him down harder than a poor first-round series would.

Legacy, with playoffs

Legacy (with playoffs) is Career plus credit for playoff success: each series counts for what it did to his team's title odds, through his own play and his share of the result against the odds. Deep runs count more, beating a stronger team earns a lot, losing as the favorite costs, and one championship's worth of title odds adds 0.51 points.

Ratings before 1996-97

Seasons 1951-52 and 1952-53 use the box-score rating alone. From 1953-54 the box-score rating is adjusted for how his teams did in games with him and without him.

The estimated lineup part is what lineup data show for later players with a similar box-score profile and a similar record of how their teams did with and without them.

Steals, blocks and the rebound split before 1973-74 are estimated from each player's box-score profile and, for players who played on past 1973-74, from his later recorded seasons. Turnovers were first recorded in 1977-78.

What each era can measure: the glossary on the all-time page.

A silent data change

After our September pull, NBA.com re-issued its 2024-25 player-tracking data on its newer Hawk-Eye definitions. A routine check caught the jump; the 24 affected files were re-pulled and the results built on them re-run. No headline result flipped: the shot-making rating's correlation with the NBA's Hawk-Eye shot-making moved from 0.71 to 0.70.