What changed · Last season (2025-26)
The 2026-27 season does not have enough games yet, so this shows 2025-26: each player's last 25 games (or his games since a role change or a trade) against the rest of his season. Once players have about 20 games this season, the page switches to this season by itself.
Cunningham's scoring is down 10.3 points a game over his 8 games since Mar 10 (14.9 vs 25.2). Most of it is playing time: his minutes went from 35.2 to 25.3 a game. Also shot volume: his shots per 36 minutes (counting free throw trips) went from 23.0 to 17.5. Partly real: about half should hold.
Cunningham's playing time is down 9.9 minutes a game over his 8 games since Mar 10 (25.3 vs 35.2). No clear cause shows up in our data; it could be an injury, a coaching choice or the schedule. Partly real: about half should hold.
Reed's playing time is up 6.3 minutes a game over his 39 games since Jan 4 (16.4 vs 10.0). No clear cause shows up in our data; it could be an injury, a coaching choice or the schedule. Probably real: about 75% should hold.
Green's 3-point rate is up 9 points over his last 25 games of the season (66% vs 57%). No clear cause shows up in our data; it could be an injury, a coaching choice or the schedule. Probably real: about 70% should hold.
Harris's midrange share is up 10 points over his last 25 games of the season (24% vs 15%). No clear cause shows up in our data; it could be an injury, a coaching choice or the schedule. Probably real: about 65% should hold.
Thompson's rim share is up 9 points over his 40 games since Jan 10 (62% vs 53%). The likely reason: Kevin Huerter joined the rotation on Feb 5 and played 20.5 MPG. Probably real: about 65% should hold.
Stewart's 3-point rate is down 9 points over his last 25 games of the season (25% vs 33%). The likely reason: Kevin Huerter joined the rotation on Feb 5 and played 20.5 MPG. Probably real: about 65% should hold.
Harris's 3-point rate is down 9 points over his last 25 games of the season (28% vs 37%). No clear cause shows up in our data; it could be an injury, a coaching choice or the schedule. Probably real: about 70% should hold.
Thompson's usage is down 4.3 points over his 40 games since Jan 10 (15.0% vs 19.4%). The likely reason: Kevin Huerter joined the rotation on Feb 5 and played 20.5 MPG. Probably real: about 70% should hold.
Holland II's rim share is up 9 points over his last 25 games of the season (48% vs 40%). No clear cause shows up in our data; it could be an injury, a coaching choice or the schedule. Probably real: about 65% should hold.
Holland II's TS% is up 4.7 points over his last 25 games of the season (55.8% vs 51.1%). Most of it is shot-making: his TS% was 6.5 points below what his shot spots predict (9.8 points below before). Also shot diet: his rim share of shots went from 40% to 48%. Partly real: about 60% should hold.
Stewart's corner-3 share is down 10 points over his last 25 games of the season (7% vs 16%). The likely reason: Kevin Huerter joined the rotation on Feb 5 and played 20.5 MPG. Partly real: about 55% should hold.
Robinson's TS% is up 4.8 points over his last 25 games of the season (66.2% vs 61.4%). Most of it is shot-making: his TS% was 9.5 points above what his shot spots predict (5.5 points above before). Also shot diet: his above-the-break 3 share of shots went from 57% to 51%. Partly real: about 55% should hold.
Every night we split each player's regular season (8+ games on each side) in one of three ways, in this order: the day he joined a new team; the game after which his minutes and usage shifted the most (a simple two-part split); or, if neither moved much, his last 10 games. Until the new season has enough games (about 20 each for most players) we use last season and its last 25 games, which catches the roles that changed at the trade deadline.
A change counts as big when it reaches:
Players need 15+ minutes a game on one side. Shooting needs 40+ threes or 60+ shots on each side.
Both windows are pulled toward the player's whole season by a fixed amount of weight M, the sample size where a stat is half skill and half luck: (window x n + season x M) / (n + M). A change is shown only if the pulled gap keeps at least half of the raw gap: Probably real keeps 60% or more, Partly real keeps 50 to 59%, anything less is Mostly noise and stays off the feed.
M values: 3P% 501 threes, TS% 218 shots and points 9.7 games, from Is it real?; minutes 9.4, usage 17, assists 18.9, rebounds 18.1, 3-point rate 13.9, rim share 18.2, midrange 16.8, corner 3 21.6 and above the break 15.4 games, measured on 1,766 pairs of back-to-back 10-game stretches in 2025-26. That is why 3P% changes almost never make it: threes are mostly luck until a player has taken hundreds.
Rules, each with a number. Points split exactly into minutes, shot volume and efficiency. Assists and rebounds split into minutes and per-minute rate. TS% and 3P% split into shot diet (league FG% from the spots he shot from) and shot-making (the rest). Context: a new team, a move into or out of the starting five (starting lineups from our stint data), key teammates (25+ minutes before) who missed most of the window, a new rotation player (20+ minutes), and opponents' defensive rating. We keep causes that explain 15%+ of the change and show the top three.
Backtest on 2025-26: we ran this as if the season stopped on 3 dates and checked each flag against the rest of the season. Probably real changes kept 71% of their size on average (we predicted 66%) and 88% kept going the same way (301 flags). Partly real kept 60% (predicted 55%). Mostly noise kept 39% (predicted 33%), so the pull is a little cautious, which is the safe side.
Latest run: 393 players checked, 447 probably real and 180 partly real changes, and 273 big swings left out as mostly noise.
Injuries a player played through, coaching decisions, matchups and schemes, and anything off the box score. Missing teammates are read from who played, not the injury report. Causes are rough splits, not proof; when nothing in our data fits we say so.
