Fantasy Hoops HQ

Not affiliated with or endorsed by the NBA. Team names are used descriptively.

How it works

How the projection works

Rankings, trades, pickups and the streaming plan rank players on one rest-of-season projection by default. This page shows how it is built and how well it would have predicted last season, next to the plain averages it replaces. Every number is calculated by the same code the tools run.

The method

A projection is a per-game line for each player for the rest of the regular season, plus the number of games he is expected to play. It is built in five steps.

  1. Past seasons

    His last three seasons, weighted 5/4/3 with the most recent first and by games played, then blended with 10 games of a league-average line so one short season cannot swing it.

  2. Age

    Each counting stat moves along an aging curve from the seasons it is built on to this one, per year: 21 and under +6%, 22 to 23 +4%, 24 to 25 +2%, 26 to 28 flat, 29 to 30 −1%, 31 to 32 −2.5%, 33 to 34 −4%, 35 and over −6%. Makes and attempts move together, so shooting percentages do not.

  3. This season

    The past seasons count as 15 games. This season's line is blended in at its real number of games, so after 15 games the two weigh the same and after 45 this season is three quarters of the projection. The last 14 days add up to 20% on top, from 3 games and in full at 6.

  4. Minutes and role

    Minutes settle faster than shooting, so they are projected on their own: the past counts as only 5 games and the last 14 days weigh up to 50%. Every counting stat scales with the minutes he is projected to play, between 75% and 130% of the minutes his line was built on.

  5. Games left

    His team's games from today to the end of the regular season, less the games before the injury report's return date, times the share of games he plays (never under 40%). Rest-of-season totals are the per-game line times those games.

The weights were set before any backtest was run and have not been tuned to it, so the results below are not fitted to the season they are scored on.

Before the season

The test: project 2025-26 from 2023-24 and 2024-25 only, exactly as the engine would have before opening night, and compare every player's projection with what he really averaged. 393 players who played at least 20 games in 2025-26 are scored. It is compared with two plain averages: last season's line, and the average of every season before.

  • 10.8%lower error than last season's averageBetter than the baseline.
  • 0.74rank correlation with the real 2025-26 values (0.75 for last season's average)Worse than the baseline.
  • 78 of 100of the real top 100 were in our top 100 (77 for last season's average)Better than the baseline.
Error by category, 2025-26. Lower is better; the best in each row is marked.
CategoryOur projectionLast season's averageAverage of the seasons before
PTS0.40 best0.430.43
REB0.38 best0.420.42
AST0.38 best0.390.40
3PM0.41 best0.430.42
STL0.52 best0.600.55
BLK0.39 best0.420.41
TO0.43 best0.470.46
FG%0.47 best0.550.53
FT%0.47 best0.630.54
All nine categories0.43 best0.480.46
Error by age on 2025-26, all nine categories
AgePlayersOur projectionLast season's averageAverage of the seasons before
24 and under1240.46 best0.530.51
25 to 301610.42 best0.480.45
31 and over530.38 best0.410.42

The projection had the lowest error in every category. Where it did worse is in the headline numbers above, when they are marked.

Month by month

The harder test is in season, when the alternative is the season-so-far average every box score site shows. On the first of each month of 2025-26 the projection is rebuilt from only the games played before that day, and scored on what each player averaged from that day to the end of the season. Players need at least 3 games before the date and 10 after it to be scored.

Error on the rest of 2025-26, by the date the projection was made. Lower is better.
  • Our projection
  • Season-so-far average
  • Last season's average
0.000.200.400.600.801.00Nov 1Dec 1Jan 1Feb 1Mar 1Our projection, Nov 1: 0.43Our projection, Dec 1: 0.39Our projection, Jan 1: 0.40Our projection, Feb 1: 0.41Our projection, Mar 1: 0.41Our projectionSeason-so-far average, Nov 1: 0.73Season-so-far average, Dec 1: 0.49Season-so-far average, Jan 1: 0.45Season-so-far average, Feb 1: 0.45Season-so-far average, Mar 1: 0.44Season-so-far averageLast season's average, Nov 1: 0.48Last season's average, Dec 1: 0.50Last season's average, Jan 1: 0.52Last season's average, Feb 1: 0.55Last season's average, Mar 1: 0.56Last season's average
The same numbers. Lower is better; the best on each date is marked.
Projected onPlayersOur projectionSeason-so-far averageLast season's average
Nov 13050.43 best0.730.48
Dec 13730.39 best0.490.50
Jan 13690.40 best0.450.52
Feb 13420.41 best0.450.55
Mar 13100.41 best0.440.56

The projection had the lowest error on every date.

Run on 2026-10-29.

How the error is measured

For each of the nine categories (PTS, REB, AST, 3PM, STL, BLK, TO, FG% and FT%), the error is the average distance between the forecast and the real per-game number, divided by how spread out the real numbers are. That puts the categories on one scale, so steals count as much as points, and the score is the average over the nine. 0 is a perfect forecast, and lower is better. FG% and FT% are weighted by attempts, so a miss on a high-volume shooter counts for more.

Rank correlation says how well the order matches: 1 means the forecast ranked every player exactly as the season did. The top 100 count is how many of the real top 100 by nine-category value the forecast also had in its top 100.

What it does not know

  • Who starts. A move into or out of the starting lineup shows up through his minutes in the last 14 days, not on the night it is announced.
  • Trades and new roles, until he has played in them.
  • Injuries past the injury report. A player listed out has his games removed until his expected return date, and nothing more.
  • Rookies, beyond their minutes. A player with no NBA seasons starts from the league-average line at his minutes, and his own games take over as he plays them.

Every tool can also rank on the season so far, the last 14 days or past seasons: switch the window at the top of rankings, trades, pickups and the streaming plan.

Try in tradeCompare

Per-game numbers in the named window. Games are this season's schedule under your league's games limit.