NBA Board
How the projections work · experiments and next tests
Interactive forecast & comparisons · October 7 archive
Team forecasts updated October 17. Player tables and research below retain their stated dates.
No sections yet. Run python scripts/nba_team_outlook.py, then predmarkets nba-page.
Season outlook
Win totals, offence and defence for all 30 teams, built up from each player's projected games, minutes and impact, then simulated 6,000 times with the October 17 impact adjustment. Shown against the Kalshi win-total market.
No. 1 is best. Each list orders teams by their expected rating; the 80% range shows where they could finish across simulations. These are league-wide ranks. Exact rating ties share a rank.
Offence
Points scored per 100 possessions above league average. +2 means two more points scored.
Defence
Points prevented per 100 possessions versus league average. +2 means two fewer points allowed.
Top-five chances include uncertainty in player availability and team performance. Wide rank ranges mean the precise order is uncertain.
Points scored & allowed per 100 possessions
Offensive rating measures points scored; defensive rating measures points allowed. A common possession count makes teams comparable regardless of pace. Scored minus allowed is the projected net rating; higher is better.
Hover, tap or focus a team for its ratings. The chart shows projected means.
View ratings table and calculation
Scored = reference average + projected offence above average. Allowed = reference average − projected points prevented above average. Net = scored − allowed. The reference uses estimated possessions (FGA − offensive rebounds + turnovers + 0.44 × FTA), so its level may differ from NBA.com or other providers. This display leaves wins, ranks and probabilities unchanged.
Labels show NBA rank and conference rank by expected wins. Top 6 Outside: 11–15. Select East or West to see each conference in order; use the table for exact wins and probabilities.
All teams
Offence and defence ranks order expected ratings (1 = best); net mean rank averages ranks across simulations. Both rating columns are points per 100 above average, with positive better. Seed odds are within the conference: top 6 avoids the play-in, 7–10 is the play-in.
Where model and market disagree
October 17 forecast versus the captured market 50% crossing. Expected wins and a market crossing are different quantities: these gaps are descriptive. See the interactive forecast for like-for-like threshold probabilities and quote details.
What history says about the usual causes: well-paid players returning from a lost season have played about 30 games; the very top draft picks have been close to average players as rookies (about one extra win); and young rosters have not beaten the sum of their players. Gaps can also come from roster moves after the roster snapshot. Treat them as questions to check, not edges.
Who plays more or less — October 7 estimates
Availability costs — October 7 estimates
Retained from the October 7 forecast; not refreshed for the October 17 team model. Expected games (80% range) and how much the team's net rating per 100 possessions would gain if the player played all 82.
Top rookies — October 7 estimates
Retained from October 7. Projected by draft slot from past classes: games, minutes per game and net impact per 100 possessions (replacement level is −2).
How lucky were they?
Players
Historical WAR and October 7 player projections, retained from the original Board. These individual projections have not been refreshed with the October 17 team forecast.
How many wins each player added over a replacement-level player (−2 points per 100 possessions), split into offence and defence. Full-health WAR is the same rate over every game he could have played.
Leaderboard
Select a name to add him to the comparison below.
Compare careers
Projected next season
Model fair salary vs actual salary
What each player's expected contribution is worth under a shared salary-cap budget, compared with his listed pay. A positive gap means the model values his contribution above his salary.
Player salary comparison
Chart shows the first ten players in the filtered, sorted table. Lines span the model's 10th–90th percentiles from availability uncertainty only.
How to interpret fair salary
In each simulation, we allocate 30 teams' salary caps across players in proportion to positive projected wins above replacement. Team minutes are capped at 240 per game; injuries and depth-chart competition affect the contribution estimate. We average these allocations to obtain model fair salary. Filtering this table never changes those league-wide values.
This is an unconstrained performance-value convention, not a uniquely correct wage or a prediction of the next contract. It does not impose minimum salaries, maximum contracts or rookie scales, and excludes commercial value, playoff fit and bargaining power. Zero means no positive contribution above replacement in this accounting; it does not mean a player can legally be paid nothing. Actual payroll can exceed the salary cap, so aggregate gaps are not a verdict on league-wide overpayment.
The range includes simulated availability and rotation capacity, but not uncertainty in player impact. Listed salary is the current-season ESPN amount when available; missing, zero or unavailable records remain unknown. It is not verified cash paid, guaranteed money, cap hit or total contract value.
The NBA's 2026–27 salary cap is $164.961 million per team. These player valuations retain their original projection dates; the team forecast above is updated to October 17.
Research & market comparison
Historical research — separate from the October 17 team forecast. The studies below test recovery, defensive context and referee effects. These recovery, defensive-context and referee extensions are not part of the October 17 team forecast.
Exact win thresholds
Each row compares P(wins ≥ threshold) with a two-sided Kalshi quote. Depth is the number of contracts at the quoted bid / ask; midpoints are not executable prices. The all-team view shows the 12 largest gaps; select a team to see all its accepted thresholds.
What the experiments found
Recovery evidence
Participation, minutes and effectiveness are separate. A preseason return is not proof of full form; an estimated return date is not a confirmed deadline.