LEGACY Ratings

LEGACY by season

Add up to five players.

Show by Seasons with at least

Seasons for the selected players

One row per season. LEGACY is net points per 100 possessions above league average.

											id

How to read this page

The chart

LEGACY is net points per 100 possessions above league average that season: 0 is a league-average player, and replacement level is around −2. Seasons under the minutes threshold are left out, which breaks the line. Rings Added for a season in progress isn't credited until after the Finals. Age is as of February 1 of the season.

Similar careers finds the 4 careers whose LEGACY, by season, season of career or age, is closest to the chosen player's: the smallest typical gap, in points per 100, over his seasons with at least the minutes shown. A season the other player missed or played less than that counts as replacement level (−2).

The table

LEGACY is from every game he played that season (regular season, play-in and playoffs). Minutes are regular-season minutes. Box prior is what the player's box score alone predicts; the rating moves away from it when game results disagree. Exp is the seasons he had played before this one in the leagues covered here (R for a rookie); age is as of February 1 of the season.

Top rated players

Choose All time to rank every season in history. Click a player to add him to the career chart.

All time: · · ·

Show Season Games League At least

										id	season	league

							id

How to read this page

Players

LEGACY is net points per 100 possessions above league average, from every game that season, playoffs included. Wins estimate counts the regular season and playoffs. Exp is the seasons he had played before this one in the leagues covered here (R for a rookie).

Teams

A team's LEGACY is how many points per 100 possessions better than an average team it was over every game that season, playoffs included, adjusted for opponents; its players' LEGACY ratings, weighted by minutes, add up to it. Best player is the one who added the most value (LEGACY over replacement, times minutes), with his LEGACY in parentheses. Title odds are its chances at the start of the playoffs.

Careers

Whole career covers every league. A best stretch of seasons in a row is the one with the most wins estimate, counting only seasons he played, so a year away (like Jordan's 1993-94) is skipped. LEGACY is his average over those seasons, weighted by minutes; wins estimate and Rings Added are summed. Rings Added follows its options on the Rings Added page.

Projected standings

Season Show

Upcoming games

Day

Time (ET)	Away	Record	Proj. rating	Home	Record	Proj. rating	Favorite

How the picks have done

Upsets only

Playoff bracket

Title odds over the season

The teams with the best title odds today, from preseason through each day's games.

How accurate are the projections?

Every season since 1983-84, projected as it would have been at the time and scored against what happened. Lower misses and errors are better.

Season played	Wins miss			Playoff error		Champion's title odds
Season played	Record pace	Results only	LEGACY	Results only	LEGACY	Results only	LEGACY

How the projections work

Reading the tables

Projected rating is the team's projected net points per 100 possessions against an average team. Projected wins are calculated exactly, game by game. Playoffs includes getting in through the play-in. A past season's preseason view uses only what was known then: the previous season's ratings plus a year of aging, the opening-night rotations, and that season's schedule and playoff rules. End of season is what should have happened: title odds for the bracket as it was, from each team's actual playoff rotation (who played, and how much) and full-season ratings, the same odds Rings Added uses; its playoff rating is the team's strength in points per 100 possessions from that rotation.

Upcoming games and picks

Each game's chance comes from both teams' projected ratings, how uncertain they still are, this season's home-court edge, and the NBA's latest official injury report. A team's rating drops by about 0.9 times what its rotation loses with the players listed Out or Doubtful sitting, their minutes going to everyone else. That weight was measured on every game since the report archive begins (December 2018), each picked from that day's report, and it made single games noticeably more accurate in every full season; players listed Questionable were tested too and added nothing, so they count as playing. Picks are made every morning from the previous evening's 5 p.m. report, then redone at 1:15 p.m. Eastern with the 1 p.m. report and at 5:15 with the 5 p.m. report (for games starting at 6 or later). The pick scored is the last one made before tip-off. Records in the picks table are as they stood before each game, and its projected ratings include that day's injuries; picks before today were made from only the games played before each day and the same reports, the same way as the accuracy test.

The bracket

Each series shows the chance each team wins it, from the projected team strengths, with home court to the better record, and the more likely winner moves on. Past the first round each series depends on the ones before it, so this is the most likely path, not a forecast: the standings give each team's overall odds. Today's bracket can be seeded by the current standings (winning percentage) or by projected wins; 7 and 8 still have to get through the play-in. Past seasons before 1983-84, and the BAA and ABA, are seeded under that season's playoff rules (who got in, byes and matchups as they were then). The bracket as it was played shows the chance each team had of winning each series from the end-of-season odds (in older seasons the better record on top, with home court).

The accuracy test

Tested on every season since the playoffs went to 16 teams, 1983-84 to 2025-26 (not the 2019-20 bubble), before opening night and after every tenth of the season. At each point, the ratings were refit using only the games played by then and the rest of the season was projected. Wins miss is the typical error in a team's final win total. Playoff error is the Brier score of the playoff odds (0 is perfect; always saying 50% scores 0.25). Record pace extends each team's win rate so far; results only projects from each team's opponent-adjusted margin so far, without player ratings. Champion's title odds is the typical (median) title chance given to the team that went on to win it: higher means the eventual champion was spotted earlier. The seasons before 1983-84, and the BAA and ABA, are scored the same way; nothing in the model was fit on them, so they test it out of sample.

Team strength

Each team's strength is its players' LEGACY ratings weighted by its healthy rotation: each player's share of the minutes in his last 10 games, times how often he's been available, pulled a little toward full health (a player who has missed a few games still counts, a player who has missed most of them counts less). Trades show up as they happen. Tested on every season since 1983-84, this predicted each team's playoff strength better than the last 10 games' minutes alone, at every point in the season. Before opening night it's last season's ratings plus a year of aging, with the opening-night rotation. It's blended with the team's opponent-adjusted results so far: early in the season the roster counts for about two thirds, by midseason the results count for about two thirds. The blend was set by the accuracy test above, not by hand. The rating shown is on the scale of a team's final opponent-adjusted net rating, so preseason, today and past seasons compare directly: the accuracy test measures how many points of final strength each point of the estimate is worth, and the estimate is multiplied by that. Preseason it's about 0.84, because an opening-night rotation plays the best players more than a season does (the same ratings with the minutes actually played read 0.99), and who will miss time can't be told in advance: weighting by past availability or last season's minutes did no better. In the season it's about 1.1.

The rest of the season

Each remaining game's odds come from the two teams' strengths plus home court, fit on the same seasons as the accuracy test (1983-84 on). Home court has shrunk over the years (home teams won about 65% of games in the late 1980s and 56% lately), so each season uses the home-court edge measured on the three seasons before it, known before opening night. That makes single games better calibrated; it barely moves season win totals, since every team plays half its games at home. Three more things are in each game's odds, all measured on the same seasons: a team on the second night of a back-to-back plays about 1.6 points per 100 possessions worse; Denver and Utah get about 1.5 points more at home than other teams (the altitude); and games at neutral sites (abroad, or the NBA Cup rounds in Las Vegas) have no home court. Tested on top of these and left out, because they didn't make the odds any more accurate: more games in the days before (3 in 4, 4 in 5, or simply games in the last 2 to 10 days), rest past one day off, travel (miles and time zones), and a separate home court for every team. A team's true strength isn't known exactly, so it gets a range: its width is set so projected win totals spread exactly as much as actual results did in the accuracy test (about 3.75 points per 100 preseason, 2.5 early in the season, 2.25 by midseason). Projected wins and their range come straight from those odds, game by game, with no simulation. Playoff and play-in odds depend on every team's results at once (tiebreakers, the play-in), which can't be counted out exactly, so those come from simulated seasons: 38,415 of them, the number that keeps every percentage within ±0.5 points, 19 times in 20. Ties in the standings are broken at random, and seeds go by record (before 2016, division winners were guaranteed top seeds). NBA Cup knockout games not yet scheduled are played against an average team.

Playoffs and title

For each simulated playoff field, title odds are calculated exactly, series by series, with the same playoff game model and home-court rule used for Rings Added, then averaged over the simulated seasons.

Past seasons

From 1983-84 on, every past season is projected the same way, as it would have been at the time: before opening night from the previous season's ratings plus a year of aging and the opening-night rotations (the same projections the accuracy test scores), and again after every tenth of the season, each with that season's rules: division winners guaranteed top seeds before 2015-16, best-of-5 first rounds through 2001-02, no play-in before 2020-21 and no home court in the 2020 bubble. Before 1983-84, and in the BAA and ABA, seasons are projected with the same model, unchanged, two things aside. Each season's playoff rules are rebuilt from its bracket as it was played: who got in, byes and matchups, with ties in the standings broken at random. Home court is measured as it is from 1983-84 on, from the three seasons before (home teams won about 72% of games in the 1950s and 66% in the 1960s); playoff home court differs little by era, so it stays the playoff game model's. Players arriving from the NBL start from their NBL rating less the NBL's measured gap with the BAA and NBA. The first BAA season, 1946-47, has no preseason projection, since no one had a rating yet. These seasons are scored separately in the accuracy test above; none of them helped fit the model. End of season, for every season back to 1937-38, is what should have happened: title odds for the bracket as it was, from each team's actual playoff rotation and full-season ratings, the same odds Rings Added uses.

What if

Put a player from the same season on a team, by replacing someone, trading for him or adding him, and the whole season is replayed with the move: records, seeds, who makes the playoffs and the title odds can all change. The title odds before the move come from that season's real standings and playoff bracket, with every team rated the same way: its regular-season rotation (each player's share of the minutes) times its players' LEGACY ratings.

Season Team

Move

Player	Share of minutes	With the move	LEGACY

The playoffs with the move

Every team with the move

Team	Wins before	Wins after	Makes playoffs	Rating before	Rating after	Title odds before	Title odds after	Change

How the what-if works

The move

The player coming in is someone from the same season. Replace takes a player off the team, trade sends that player to the new player's team, and add brings him in with nobody leaving. Every team is rated from its regular-season rotation: each player's share of the team's minutes times his LEGACY rating that season. The player coming in plays his own regular-season share and leaves the team he played the most minutes for, and when a team gains or loses minutes the rest of its roster gives up or takes the difference evenly. That's how teams really covered for a missing player: in games from 1985 on without a player who usually plays 29 or more minutes, an even split matched the team's strength better than a replacement-level player or everyone scaled up in proportion.

Replaying the season

Each team that gains or loses a player gets a new strength, and every game it played is replayed with it. A game keeps its luck: it only changes if the move would have changed it, so a win the move makes less likely can turn into a loss, a loss it makes more likely can turn into a win, and every other game ends as it did. With no move, every record is the real one. The new records then seed the playoffs under that season's own rules (division winners, byes, the play-in), with ties broken the way the real standings broke them, and home court goes to the better record. Every team plays the playoffs at the same strength it had in the regular season, so teams that made the playoffs and teams that didn't are rated the same way. The season is replayed 38,415 times (enough that every percentage is within half a point, as on Season projections), and each playoff field gets exact title odds, series by series, averaged over the replays. Before the move, that's simply the real bracket. Every BAA, NBA and ABA season is included; the NBL seasons aren't, since there are no game-by-game results for them.

The bracket

The bracket as it was played (with a move, seeded by each team's average record over the replays): each series is between the favorites of the series feeding it, with the chance each team wins it, and the favorite moves on. Past the first round each series depends on the ones before it, so this is the most likely path, not a forecast: the title odds table gives each team's overall odds.

Wins estimate by season

Each team's actual wins, split among its players. Click a player in the table below to show him here.

Career wins estimate

Regular season and playoffs, each split from that season's team wins. A very bad player can go below zero.

					id

How the wins estimate works

Each team's actual wins are split among its players: his share of the team's playing time, plus or minus how much better or worse he was than his teammates (LEGACY above the team's average times possessions played, converted to wins at that season's rate). A team's players add up to its record. The regular season and the playoffs are each split from that season's team wins, so a very bad player can go below zero.

Rings Added

How much a player raised his teams' title odds, in championships, summed over his career.

						id

Rings by season

How many Rings Added each season handed out, and why some seasons count more than others.

League

Difficulty by season

How hard each season's title was to win (1.00 = average).

League strength by era

How strong each season's league was compared with the average league in history (0 = average).

Season League

Title odds with the playoff rotations

Each playoff team's title odds from who actually played in the playoffs.

Title odds by team strength

Each line shows how a team's title odds would change if it were stronger or weaker, with the other teams unchanged. The dot is where it actually stood. Hover a line to pick out a team.

Rings Added that season

Player	Team	Minutes	LEGACY	Rings Added	Regular season	Playoffs	id	season	league

How Rings Added works

Rings Added is how much a player raises title odds. Each season hands out rings in proportion to how hard that season's title was to win (the strength a team needed for even title odds, times how much its season length could prove, compared with the average season in history), times how strong its league was, scaled so the average season hands out 5, a ring for each starter: all of Rings Added adds up to five starters' rings for every title ever won. That season's total is split between the regular season and the playoffs by how much of the title each one decides (about half), measured each season and averaged over every season played in the same format, so seasons played the same way split the same way.

Regular season

Put him on a league-average team in place of a replacement-level player and measure the rise in its title odds, using the title curve for that season's format (every season with the same format pooled, so one runaway favorite doesn't tilt a season's credit toward its own stars). An average team only contends with a star, so peak seasons count far more than good ones.

Playoffs

Every playoff series a team wins raises its title odds (calculated exactly before and after each round). That gain is shared among its players by their playoff minutes times rating above replacement. Deep runs and upsets count most; losing a series costs nothing.

Eras and leagues

A title counts by how strong that season's league was compared with the average league in history (1.00): late-1940s titles about 0.4, the 1960s about 1.4, the early 1970s (expansion and the ABA) about 0.65, the early 2000s about 1.0, today about 1.4. ABA and NBL titles also count less for the measured gap with the NBA (How it's built explains both). Every league's title is a title: in the two-league years (BAA and NBL 1946-47 to 1948-49, NBA and ABA 1967-68 to 1975-76) each league hands out its own pot, weighed by its own strength.

Season length

A record over a short season is more luck than one over a long season (over G games, about G / (G + 11) of it is true strength, measured from 82-game seasons), so a season's difficulty is multiplied by that share, compared with the longest season (1.00). An 82-game season counts 1.00, the 50-game 1998-99 season 0.93, a 13-game NBL season 0.60.

Rings by season

How many Rings Added each season handed out: Rings = difficulty × league strength, scaled so the average season hands out 5 (a ring for each starter of every champion). Difficulty is how good a team had to be to have even title odds that season (league size, playoff format, how strong the top teams were), times its season length (how much of a record that long is true strength rather than luck, compared with the longest season), compared with the average season in history (1.00). League strength is how much a title there counts given how strong the league itself was compared with the average league in history (1.00; stronger leagues count more, weaker ones less; How it's built explains it). Both average 1.00 across history. Playoff share is how much of the title is decided in the playoffs rather than the regular season, measured each season and averaged over every season with the same format (league size, playoff teams and series lengths); it splits each season's titles between the two.

Difficulty

How good a team had to be to have even title odds each season, times how much its season length could prove, compared with the average season in history (1.00 = average). Bigger leagues, longer playoffs, longer seasons and stronger top teams make a title harder to win.

League strength

How deep each season's competition was compared with the average league in history, in points per 100 possessions for a team (0 = average; today is about +3), measured from how returning players held up against the league around them, with the overall level set by league free-throw shooting. The band is the plausible range. It compares leagues to each other, not head to head across eras. ABA and NBL seasons include their measured gap with the NBA.

Title odds with the playoff rotations

Calculated exactly from that season's actual bracket and each team's playoff roster strength (who actually played in the playoffs, and how much, with full-season ratings), so unlike the playoff projections on the Season projections page, these know who was healthy and how the rotations tightened: each game's win chance from the strength gap and home court, each series from best-of-N odds, and each round summed over every possible opponent.

Predicted vs actual team net rating

Each dot is a team. The prediction uses only ratings from seasons before it, times each player's actual minutes that season. Dots on the diagonal were predicted exactly.

Season R² this season: — vs — for last season's team net

Accuracy by season against other metrics

R² for predicting each team's net rating from the previous season's player ratings. Higher is better.

Every metric against LEGACY

How well each metric predicts next season's team net ratings, against LEGACY over the same seasons. Higher R² is better.

Metric	Seasons	Best scale	R²	RMSE	LEGACY R²	LEGACY RMSE	LEGACY spread in R² (95% range)

Splitting credit among teammates

Whether credit went to the right players within a team, tested on stretches played by the same ten players.

Metric	Seasons	R² (%)	LEGACY R² (%)	LEGACY spread (95% range)

Playoff games

Each playoff game's margin, predicted from regular-season ratings.

Metric	Seasons	R²	LEGACY R²	LEGACY spread (95% range)

How the tests work

Accuracy by season

R² for predicting each team's net rating from the previous season's player ratings. Higher is better; 1.0 would be perfect. Outside metrics are each scaled by the factor that suited them best in hindsight (listed in the Every metric against LEGACY table), which favors them. DARKO is its preseason projection for each season, made with data through the season before.

Every metric against LEGACY

Each row compares one metric with LEGACY over exactly the seasons that metric covers. R² is the share of the variation in next season's team net ratings that each metric predicts, from the previous season's player ratings times actual minutes. Outside metrics are multiplied by the scale factor that suited them best in hindsight, which favors them. DARKO is its own preseason projection for each season (also shown with LEGACY's aging curve instead); the others get LEGACY's aging curve. LEGACY is tested on its ratings before the team-totals step (the step that makes a team's published ratings add up to its results), so that step gives it no head start; applying the same step to the other metrics lowers their scores slightly. The last column is LEGACY's spread, how much higher its R² is than the metric's, with the range the spread falls in 19 times in 20 when whole seasons are resampled; a range that includes 0 means the spread could be chance. RMSE is the typical miss in points per 100 possessions (lower is better). LEGACY's R² and RMSE are over the same seasons as that metric. (For last season's team net, the scale is how much of it carries over: 0.7.)

Splitting credit among teammates

The team test can't tell whether a team's credit went to the right players. This one can. Each season's ratings (from earlier seasons only, as in the team test) predict every stretch of the next season played by the same ten players, about 35,000 a season, measured against the rest of that game, so team strength and opponent cancel out. What's left is which lineups outscored the others in the same game, which only depends on how the ratings split credit among teammates. A stretch lasts a few possessions, so most of its margin is luck and every R² is small; what matters is the comparison on the same stretches. R² is in percent here, and each metric gets its best scale in hindsight.

Playoff games

Each playoff game's margin predicted from that season's regular-season ratings (the playoffs left out of the fit) times who actually played in it. Only metrics whose published values leave out the playoffs can take this test: Basketball-Reference BPM and win shares (RAPTOR's, DARKO's and LEBRON's season values include or may include playoff games). The team rating uses no player information. LEGACY here is the published rating, with the team-totals step. Every range reaches 0: a season's 80 or so playoff games, mostly between close teams, are too few to separate these, so LEGACY holds up in the playoffs but doesn't stand out.

My closing five · Week of Nov 11, 2026

Nobody has ever been this good at 38, and 4 more

Also: Houston's best player is 13th, Giannis isn't Miami's problem, the Wolves without Ant, and OKC at +450.

Numbers through games of Nov 6, 2026.

1. Nobody has ever been this good at 38

Steph is having the best season anyone 38 or older has ever had. He's at 6.59 in LEGACY, 5th in the league, and the old record was John Stockton's 6.37 in 2000-01.

This isn't a great player hanging on. It's his best season since the 2022 title year (7.38), and a real jump from 5.01 last year. LeBron, the guy everyone uses as the standard for aging, topped out at 4.60 at 38.

It's also why the Warriors are 6-3 after starting 0-3. Look at who's around him. Podziemski is next best in the rotation at 1.74. Golden State was projected for 30.7 wins before the season and is at 48.0 now, the biggest jump in the league.

Warriors rotation	Age	Minutes	LEGACY
Yaxel Lendeborg	24	285	1.02
Brandin Podziemski	23	233	1.74
Stephen Curry	38	202	6.59
De'Anthony Melton	28	201	0.94
Draymond Green	36	187	0.13
Georges Niang	33	181	-1.58
Will Richard	24	159	-0.84
Al Horford	40	130	0.27

The pushback is minutes. He's played 202, and every other season on that list is a full year. But he doesn't have to stay at 6.59. He has to stay above 6.37, and he was above that in every full season from 2013-14 to 2021-22.

2. Houston's best player is 13th in the league

Houston is the best team in the league, and it doesn't have a top-12 player. The Rockets are at 9.04 in LEGACY, ahead of the Spurs (8.10) and the Lakers (7.92), and their best player, Şengün, is 13th at 3.77.

It's the best Rockets team ever, ahead of the 65-win team in 2017-18 (8.02). That team had Harden, 2nd in the league.

People are calling them the deepest team in the league, and that's exactly what the numbers say. Seven guys in the rotation are at 1.3 or better, and the only real minus is Bogdanović in 110 minutes.

Rockets rotation	Minutes	LEGACY
Jabari Smith Jr.	311	2.01
Kevin Durant	304	2.55
Amen Thompson	293	2.15
Alperen Şengün	246	3.77
Tari Eason	221	1.32
Fred VanVleet	210	3.02
Marcus Smart	137	0.24
Reed Sheppard	125	2.05
Bogdan Bogdanović	110	-1.69

Great teams almost never look like this. Since 1979-80, 44 teams have been rated 8 or better. Only two had a best player ranked lower than Şengün: the 2011-12 Spurs and the 1993-94 Sonics, who won 63 games and lost in the first round to Denver.

The other side: in the playoffs rotations shrink and stars take over. But teams without a star have won it. The 2004 Pistons' best player was Ben Wallace, 14th in the league.

3. Giannis isn't why Miami is 3-6

Giannis is down this year, but he isn't why the Heat are 3-6. Klay, Wiggins and Hardaway are, and between them they're taking away everything Giannis gives.

Start with Giannis, because the drop is real. He's at 4.67 in LEGACY, down from 6.20 last year and his lowest since 2016-17. He's also still 9th in the league.

Now the other end. Klay is at -3.16 in 210 minutes, Hardaway at -3.17 in 140 and Wiggins at -2.60 in 232. Put each guy's rating against his share of the minutes and those three take 3.9 points per 100 off the team. Giannis and Bam add 3.9 together. That's a wash, and the rest of the bench gets you to -3.08, 19th in the league.

	Minutes	LEGACY	Last season
Giannis Antetokounmpo	295	4.67	6.20 (MIL)
Andrew Wiggins	232	-2.60	-0.47 (MIA)
Klay Thompson	210	-3.16	-2.19 (DAL)
Tim Hardaway Jr.	140	-3.17	-1.18 (DEN)

Plenty of people are already saying it's not Giannis's fault. I agree, and I'd name names. Two of the three were already giving minutes back last year.

I don't see Miami getting to .500 until most of Klay's and Hardaway's minutes go to someone else. I'm watching who closes at Sacramento and Utah this week.

4. The Wolves can live without Ant

Losing Ant doesn't change things as much as you'd think. Minnesota is at 7.62 in LEGACY, 4th in the league, ahead of the 2023-24 team that made the conference finals (6.79).

Ant is out up to two months after knee surgery, and he was at 4.75 before he went down. Take his 135 minutes out and weight everyone else by their own minutes, and the Wolves are still at 6.66. That's basically the 2023-24 team, and still 4th in the West behind Houston, San Antonio and the Lakers.

The reason is LaMelo. He's at 5.26, 7th in the league and a career high, up from 4.59 last year in Charlotte.

Around him it's the stuff that holds up without a second star: McDaniels at 1.72 in a team-high 340 minutes, Gobert at 2.20, Dosunmu at 1.33.

Wolves rotation	Minutes	LEGACY
Jaden McDaniels	340	1.72
Ayo Dosunmu	315	1.33
Jonathan Kuminga	312	0.74
LaMelo Ball	276	5.26
Rudy Gobert	269	2.20
Trey Lyles	182	-0.90
Joan Beringer	165	0.74
Terrence Shannon Jr.	154	-1.79
Anthony Edwards (out)	135	4.75

LaMelo now has to carry an offense with Kuminga as the next guy up, and two months is a long time for the star to be out. But Ant's 4.75 was well above his normal (3.20 last year), so the team without him isn't as far from the team with him as it looks.

5. What Vegas has wrong: OKC's injuries

Vegas still has OKC second on the board at +450, about 18%. I have them at 4.8%, and I think even that's too high, because Vegas is pricing a team that won't be on the floor.

Jalen Williams fractured his fibula and Cason Wallace tore his labrum, and neither is likely to play again this year. Chet is out 8 to 10 weeks with a small patellar tendon tear. Wallace and Chet were second and third on the team in minutes last year. That's on top of a summer that sent Dort and Aaron Wiggins to Atlanta and Isaiah Joe to Detroit.

Last year's rotation	Minutes	LEGACY	Now
Shai Gilgeous-Alexander	2,224	8.60	Playing, 7.50
Cason Wallace	2,020	2.01	Out, likely for the season
Chet Holmgren	1,967	4.39	Out 8 to 10 weeks
Luguentz Dort	1,828	-0.39	Atlanta
Isaiah Joe	1,506	1.57	Detroit
Ajay Mitchell	1,447	1.67	Playing, 1.78
Aaron Wiggins	1,405	-0.08	Atlanta
Isaiah Hartenstein	1,122	2.85	Playing, 2.32
Alex Caruso	997	3.27	Playing, 2.56
Jalen Williams	904	1.92	Out, likely for the season

The minutes are going to McCain (-2.09) and the rookies Stirtz (-1.94) and Mara (-1.54). Even Shai is down to 7.50 from 8.60 last year, though he's still 4th in the league.

And my 4.8% is generous. It still gives Williams and Chet a good share of the minutes, because they played a few games before they got hurt. Take Williams and Wallace out for the year and it's 4.4%. Take Chet out too and it's 2.8%. With Chet back around New Year's, the real number is somewhere in between.

Vegas's side is Shai, plus a team that won the title two years ago and went 64-18 last year. They're 7-2, too. But they're 6th in the league at 5.75, a long way from last year's 11.14.

How it fits together

Every number on the site comes from one rating model.

How much the box score explains

Share of the variation in modern (1997+) game-based ratings that each era's box score predicts, scored on seasons held out of training. Older eras recorded fewer stats, so their box prior is weaker.

Expected change by age

Average change in rating from one season to the next, by the age the player turns. Dots are the raw averages; the line is the smoothed curve the model uses.

How LEGACY is made

One model for all of history

LEGACY stands for Lineup-Estimated Game-Adjusted Contribution by Year. Each player-season gets one LEGACY rating, from every game he played that season: regular season, play-in and playoffs. From 1996-97 on, the model uses play-by-play: every game is split into stints, stretches with the same ten players on the floor (over 1.1 million of them), and each stint's margin per 100 possessions should equal the five home players' ratings minus the five away players', plus home court. Before 1996-97, when no play-by-play exists, each game's margin should equal the sum of the players' ratings weighted by their share of the minutes. (A version split into offense and defense was tested and predicted slightly worse, so the published rating is one number.)

Box-score prior

Ratings are pulled toward what the player's box score predicts. That prediction was trained on modern seasons and applied to each era using only the stats that era recorded. A few minutes of box stats are mostly noise, so low-minute seasons, and the first games of a season, lean on the player's own previous season instead (a rookie leans on the typical low-minute player). As the minutes add up, this season's box score takes over.

Neighboring seasons inform each other

A season is also pulled toward the same player's seasons before and after it, adjusted for aging, with less weight the further apart they are. That's how Scottie Pippen's 1993-94 season without Michael Jordan helps separate the two in the years they played together. Because later seasons inform earlier ones, past ratings can shift slightly as new seasons are played. The accuracy test refits the model on data through each season only, so it never sees the season it predicts.

Old box scores

Before about 1984 many box scores did not record minutes. Those minutes were estimated from each game's stats and made to add up to 240 per team-game and to each player's season total. Possessions were estimated where turnovers and offensive rebounds weren't recorded.

Eras and leagues

Ratings are relative to each season's league, so on their own they can't say whether the league itself got better. Two independent measures are combined. Returning players: when the league improves, a returning player looks a little worse than his age alone predicts, because everyone around him improved; comparing every returning player's change with the aging curve, season by season, and adding those changes up gives each season's strength relative to the others, except for one small unknown drift per season that compounds over 80 years. Free throws: league free-throw percentage is pure skill (no defense, the same distance in every era), so its gap from today, in points per 100 possessions at today's free-throw rate, is a direct measure of how much one skill improved. The drift is set so the two measures agree as closely as possible across every season: no year is assumed to be a fixed reference. Fitted, the overall gaps come out a little over twice the free-throw gaps, the same shape over 80 years. The free throws only set that one drift; the season-to-season shape comes from the returning players. How much rides on it: forcing the overall gap to anywhere from 1 to 3 times the free-throw gap leaves the career Rings Added top 10 unchanged and moves no one in the top 25 more than five places. A title then counts by the chance that season's champion would beat an equally dominant champion of the average league in history in a seven-game series (today's league is about 3 points per 100 stronger than that average), averaged over resamples of the players (as many as it takes for every season's factor to be within ±0.01, 19 times in 20; currently about 1,000) so an uncertain gap is neither ignored nor taken at face value. Stronger leagues count more than 1 and weaker ones less, scaled so they average 1 across history, like the difficulty weight.

The ABA (1967-68 to 1975-76) and the NBL (1937-38 to 1948-49, the league that merged with the BAA to form the NBA) also count less for the gap measured from players who switched leagues. NBL ratings rest on very little: the NBL recorded only games, field goals and free throws made, free-throw attempts from 1946-47, and points, with no minutes or game results. Each NBL player's rating is what those stats predict (learned from 1950s-60s NBA players), adjusted so each team's ratings add up to its point margin.

Team totals

After the fit, each team's leftover gap between its players' ratings and its actual opponent-adjusted results is spread across its players by minutes, so a team's ratings add up to how it actually played. (The accuracy test on the Model check tab uses the ratings before this step.) During a season, a few games say little, so a team's results so far are blended with what its roster predicts, and the results take over as the games add up.

Data and credits

Game logs and play-by-play come from NBA.com (1996-97 to 2024-25 play-by-play via the shufinskiy/nba_data archive), and season totals from Basketball-Reference. The accuracy comparisons use FiveThirtyEight RAPTOR, Basketball-Reference BPM and win shares, BBall Index LEBRON and DARKO.

All ratings, wins estimates and Rings Added are this project's own estimates. This is an independent project, not affiliated with or endorsed by the NBA, its teams or any of the sources above.