1. Player value
Every rostered player gets a rating on the box plus-minus scale. For 2025-26 it blends four inputs, each standardised across the league and weighted by minutes: BPM 40%, WS/48 25%, VORP per 2,000 minutes 15%, and efficiency 20%. 2024-25 uses the first three only (50/30/20).
Efficiency is half usage-weighted true shooting vs. the league average, and half individual net rating. In that net rating, offensive rating above average is scaled by usage, so a 125 ORtg on 30% usage counts far more than on 13%. Defensive rating gets half weight because it mostly reflects the team's defence.
2025-26 counts fully and 2024-25 at half weight. The total is pulled toward replacement level (−2.0) with the weight of 800 minutes, so small samples can't produce extreme ratings.
2. Rookies
Rookies are rated from their 2026 preseason (NBA.com): Player Impact Estimate mapped onto the same scale, minus a point because preseason opposition is weaker. That is regressed hard toward a typical rookie (−2.5) with the weight of 300 minutes. Rookies with no preseason minutes start at −2.5.
3. Career stage (fitted aging curve)
Age adjustments come from real data: 703 players who logged 250+ minutes in two straight seasons (2023-24 to 2024-25, and 2024-25 to 2025-26). For each, the change in rating is measured against age, weighted by minutes. The fit removes regression to the mean, because players coming off great seasons tend to dip regardless of age.
- Guards: +0.80 at 21 and under → +0.49 at 22–23 → +0.29 at 24–25 → about 0 at 26–29 → −0.10 to −0.14 at 30–36 → −0.53 at 37+
- Wings: +0.76 → +0.64 at 22 → +0.44 at 24 → +0.28 at 25 → +0.09 at 26–29 → −0.11 to −0.20 at 31–36 → −0.37 at 37+
- Bigs: +1.09 → +0.57 at 22–23 → +0.32 at 24 → about +0.05 at 25–31 → −0.18 to −0.24 at 32–36 → −0.58 at 37+
Each position's curve is pulled toward the all-player curve (positions have small samples) and can't rise with age. Stage labels follow the fitted values. The data can't see players who declined so far they left the league, so declines after 33 may be understated. Rookies get no age adjustment, since their rating already reflects a typical rookie season.
4. Injuries
Current injuries come from the league-wide injury report, refreshed every morning. Each expected return date becomes the share of the season the player should be available: a Jan 2 return is about 57% of the season, and an Apr 2 return about 5%. Day-to-day players count 98%, and season-ending injuries 0%. Earlier reports (Covers, FantasyAlarm) fill gaps, and the more cautious figure is used when sources differ.
Injury history: durability is the share of games played over the last three seasons, weighted 50/30/20. It sets baseline availability from 55% to 95%. Players returning from a major injury (Achilles, ACL, patellar tendon) also lose 0.4 to 1.0 points of rating in their first season back.
4b. Playoff performance
Each player's playoff BPM and true shooting from the last five postseasons (2022–2026) are compared with his own regular season in the same years. The comparison is against what's expected for a player of his level, not against zero. Across all 2022–26 playoff players, a player's BPM changes by about +0.27 − 0.18 × his regular-season BPM, and his TS by +0.136 − 0.272 × his regular-season TS%. Stars are expected to dip more, because of tougher competition, more defensive attention and regression to the mean.
Playoff score = BPM vs. expected + 0.15 × TS points vs. expected, pulled toward neutral with the weight of 600 playoff minutes. Beating expectations adds 0.30 per point (up to +0.8). Falling short subtracts only 0.15 per point (down to −0.5), since it comes against better competition. Labels: riser at +0.75 or more, dip at −0.75 or less (after small-sample shrinkage). Players with no recent playoff minutes aren't adjusted.
4c. Awards and GM survey
Awards carry information the box score misses, so each player gets an award index over the last three seasons (weights 1 / 0.6 / 0.35, latest first). Points: MVP vote share ×4; All-NBA 1st/2nd/3rd 2 / 1.5 / 1; All-Star 0.75; DPOY share ×1.5; All-Defensive 1st/2nd 0.75 / 0.4; Finals MVP 1; NBA Cup MVP 0.3; Clutch Player share ×0.5; Rookie of the Year share ×1; Sixth Man and Most Improved shares ×0.5; scoring title 0.5.
Does it matter? Across 290 players with 500+ minutes in both 2024-25 and 2025-26, each index point predicted +0.29 rating the next season beyond the prior season's stats (standard error 0.09); players with no awards slightly under-performed their stats. The model already blends two seasons, which captures part of that signal, so the applied value is 0.20 per point, capped at +1.5.
The 2026-27 NBA.com GM survey adds a forward-looking scout's view: each vote share is weighted by question (MVP 1.0, best at position and start-a-franchise 0.5, best defender 0.4, breakout 0.6, Rookie of the Year 1.0, smaller weights for skills questions) and converted at 0.5 rating per point, capped at +0.75; for rookies it counts double, since their ratings rest only on preseason games. Awards and GM votes together are capped at +2.0.
Example: Victor Wembanyama (2026 DPOY, MVP runner-up, 77% of GM MVP votes) gets the full +2.0; Cameron Boozer (50% of GM Rookie of the Year votes) gets +0.5. Sources: Basketball-Reference award voting 2022–2026, Wikipedia All-Star rosters 2025–26, NBA.com 2026-27 GM Survey.
5. Minutes and team score
Each player's expected minutes start from his recent minutes per game, nudged by rating and scaled by availability. Every team then fills exactly 240 minutes, cutting from the end of the bench first. Team projection = Σ rating × minutes ÷ 48, centred so the league average is 0.
6. Continuity and in-season blend
Continuity is the share of projected minutes going to players who ended 2025-26 on the same team. The edge is 4 points per 100% difference from the league average, and it halves every 15 games played.
Once games start, the score blends toward actual net rating: actual weight = games ÷ (games + 20). After 20 games it's half and half.
7. Head coach
Quality: the coach's regular-season record, regressed toward .500 with the weight of 400 games. Only a quarter of it is credited to the coach, because rosters drive most of a record. Up to +0.6 more comes from career playoff wins (full credit at 100).
New-coach discount: up to −1.0 in a coach's first season with a team, reduced by his pedigree. Pedigree is 60% NBA games coached (full at 800) and 40% playoff wins (full at 60). A first-time coach takes the full discount, and a long-tenured winner almost none. First-time head coaches also start at −0.2 quality. The discount halves every 15 games.
8. Coaching staff
Former NBA head coaches on the assistant staff add up to +0.5. Each counts by his head-coaching games (full at 400), scaled by his win rate and playoff wins.
Staff continuity, the share of assistants back from last season, adds or subtracts up to about ±0.5 and fades like roster continuity. Dallas and Phoenix haven't published 2026-27 staffs, so they're scored as league-average.
9. Fit
The rotation's minutes-weighted spacing (3-point rate × shooting accuracy), playmaking (assist rate), rim protection (block rate) and rebounding are each compared with the league. Shortfalls cost more than surpluses help.
There's also −0.5 with no regular creator (an assist rate of 25% or more in 24+ minutes), and a penalty when the top five players' usage adds up to over 125 or under 100. The result is centred on the league average and capped at ±1.0.
10. Positional balance
Minutes are split into guards, wings and bigs from roster positions and heights. Only extremes are penalised: guards outside 25–55% of minutes, wings outside 15–50%, bigs outside 15–40%. The penalty is 0.04 per percentage point outside the range, then centred on the league average.
11. Schedule
Schedule load comes from the published 2026-27 schedule: back-to-backs, three games in four nights, games with less rest than the opponent, and travel miles. It's averaged across the season and is small by design, since one back-to-back costs about 1.5 points in that game only.
Strength of schedule is the average calibrated strength (see 11a) of the opponents each team actually faces. It changes projected wins but not the power score.
11a. From power score to projected wins
Wins = 82 × (0.5 + 0.033 × (expected net rating − schedule strength)), where 0.033 is the standard 2.7 wins per point of net rating. Before games are played, the expected net rating is the power score × 0.60.
Why shrink: projected strength is noisier than real strength, so uncalibrated projections spread too wide (an SD of 16.6 wins, with six teams at 60+, which has never happened; the record is four, in 1997-98). Two independent checks set the factor. Actual win totals in 2024-26 had an SD of about 13.5 and good preseason projections miss by about 8.5, implying a projection SD near 10.5 (factor 0.63). The Oct 15, 2026 betting win totals (SD 10.3) track our raw projection with a slope of 0.56 (correlation 0.90). The model uses 0.60: projected wins now have an SD of 10.3, sum to 1,230, and one team clears 60.
The power score and the ranking order are not shrunk. Once games are played, actual net rating takes over (weight GP / (GP + 20)) without shrinkage. Playoff, play-in and lottery odds simulate each season with ±8.5 wins of outcome noise.
11b. Load management
Measured from NBA.com game logs for 2023-24 to 2025-26. For each team's top three players by minutes, I compared their miss rate in the last 15 games with earlier in the season, and on the second night of back-to-backs with other nights, grouped by final seed.
- Playoff teams (seeds 1–6): late-season misses rise from 7.5% to 15.4%; back-to-backs 10.1% vs 7.0%. About 1.6 extra games per top player.
- Play-in teams (7–10): 9.7% → 10.9%; back-to-backs 11.5% vs 9.3%. About 0.4 extra games, so regulars play the most.
- Lottery teams (11–15): 11.0% → 31.9%; back-to-backs 15.1% vs 10.2%. About 3.7 extra games, mostly late-season shutdowns.
Each player's availability already reflects his own history, typical rest included, so only a tier's difference from the league average is applied. Each team's tier chances come from 4,000 simulated seasons (projected wins ± 8.5). Each extra game a top player sits costs his value over a replacement-level fill-in. The adjustment is zero-sum across the league and changes projected wins only, about ±0.25 at most, not the power score.
12. Leadership
There's no direct statistic for leadership, so this uses measurable proxies for the team's top three players by minutes: career playoff games (45%, full at 80), career championships (35%), and years in the league (20%, full at 12).
It's centred on the league average and capped at ±0.8. It can't see locker-room influence, communication or temperament.
Sources
Player and team statistics, box scores, rosters, injury reports, playoff history and betting lines come from a licensed sports data provider and are refreshed every morning. Public facts such as awards, head coaches and assistant staffs, coach records, championships and birthdates were compiled from public records; the 2026-27 schedule and the 2026 preseason rookie stats come from NBA.com. Player ratings, projections and win probabilities are this site's own calculations.
Not modelled: in-season trades, and the two NBA Cup games not yet scheduled.