Report a problem
Since Oct 23, 2026 · 26 games · Points league
Projection accuracy
Every projection we publish is frozen at tip-off and scored against what the player actually did, good nights and bad. This page updates after every game, once the box score is in. Includes games through Oct 25, 2026.
Headline numbers
Fantasy-relevant players: those we projected for 20+ points, 296 player-games so far.
Early days
Typical miss
Week by week
| Week | Games | Players | Typical miss | Season avg miss | Closer than avg | Head-to-head |
|---|---|---|---|---|---|---|
| Oct 19–25 | 26 | 296 | 11.86 | 14.29 | 56.9% | 60.7% |
Relevant players
By position
| Position | Players | Typical miss | Season avg miss | Closer than avg | Head-to-head | Start/sit hits | Inside range | Lean |
|---|---|---|---|---|---|---|---|---|
| Guards | 136 | 12.10 | 14.10 | 50.7% | 63.1% | 52.8% | 80.1% | +1.40 |
| Forwards | 126 | 11.51 | 14.03 | 61.1% | 57.8% | 41.7% | 84.1% | +0.15 |
| Centers | 34 | 12.21 | 16.05 | 66.7% | 63.3% | 72.2% | 79.4% | +1.24 |
What was live when
By model version
| Model | Games | Players | Typical miss | Closer than avg | Head-to-head |
|---|---|---|---|---|---|
| crow-nba-v1 (minutes × rates, simulated) | Oct 23, 2026 | 120 | 11.58 | 57.1% | 60.2% |
| crow-nba-v3 (learned minutes and chance to play, simulated) | Oct 24, 2026 – Oct 25, 2026 | 176 | 12.05 | 56.8% | 61.2% |
What the numbers mean
Typical miss
How far a player's real score landed from our median projection, on average, ignoring direction. Nightly fantasy scoring is noisy: a hot shooting night, foul trouble or a blowout that benches the starters moves a score by 15 points or more, so even a perfect read misses by several points most nights.
Closer than the season average
The simplest forecast is "he'll score his average so far this season". This is how often our projection landed closer to the real score than that guess. Above 50% means the model knows something the average doesn't: matchups, injuries, role changes.
Head-to-head calls
The start/sit question. For every pair of players at the same position on the same game day, both projected for 20+ points, how often the one we ranked higher actually scored more. A coin flip is 50%.
Start/sit hits
Of the players we ranked in a game day's top 12 at guard, forward or center, the share who finished in that group. "Ranked right" checks every pair at a position, not only the top.
Inside our range
Every projection has a floor and a ceiling built so that about 8 in 10 real scores land between them. Far fewer would mean we're overconfident; far more, too cautious.
Lean
Our average projection (the mean of the simulations) minus the real score. Positive means we projected too high overall. Medians sit a little below averages because big games pull fantasy scores upward, so the lean uses the mean.
How we score ourselves
- Projections lock at tip-off and are never changed afterwards. What you see here is exactly what was on the site before the game.
- Real scores are points-league fantasy points (ESPN's default points scoring) from the official box score. A game counts once its box score is in, usually within the hour.
- A projected player who didn't get on the floor (a late scratch, or a coach's decision) scores 0, as he would in your lineup.
- Headline numbers cover players we projected for 20+ points, so deep bench players who were always going to score near zero don't flatter us.
- The numbers start when our NBA projections launched in October 2026. Before launch the model was tested the same way on seven past NBA seasons (2019-20 to 2025-26): settings chosen on 2020 to 2023 and checked once on 2024-25 and 2025-26, where it missed by 9.51 points on average against 10.08 for the season average.
- See tonight's numbers on the projections page.