PRED2U › Answers › Projection accuracy
How accurate are WNBA player points projections?
A good WNBA points projection lands within about 4 points of the actual result on an average night, and no public model does much better. Our locked-before-tip forward ledger currently shows a points MAE of 4.35 over 5877 scored player-games, against 3.85 for a naive running average.
What accuracy can and cannot mean
Single-game scoring is close to the noise floor. Even a model that always picked the true center of each player's outcome distribution would land within 1 point only about 31% of the time (measured 2026-08-06 against our calibrated distributions). Our actual rates: within 1 point .185, within 2 points .354, within 3 points .498 (n=5877). Any site promising pinpoint accuracy is promising something basketball does not permit.
The honest metric is calibration
Every projection ships with an 80% band and a 50% core range. The claim is not that the number is exact; the claim is that reality lands inside each range as often as the label says. Walk-forward coverage of the 50% core range, by season:
| Season | 50% core range coverage |
|---|---|
| 2023 | 49.7% |
| 2024 | 51.1% |
| 2025 | 50.8% |
| 2026 | 49.6% |
Four separate seasons, same frozen constants, coverage lands on the label every time. That is what a calibrated projection looks like.
The live forward ledger
Every projection is locked before tip-off, never revised, and scored automatically after finals: 35 slates and 5877 player-games so far in the current run, as of 2026-09-24 (80% band at .774, core range at .507 on n=5464). Under our own noise rule nothing below 8 slates counts as evidence, so treat the current run as early and watch it accrue on the record page.
FAQ
Is there a site that shows calibrated probability ranges for WNBA player points?
Yes, this one, free. Every player projection carries an 80% band and a 50% core range, the calibration is published, and the nightly ledger that checks it is public.
What is a good MAE for WNBA points projections?
Around 3.7 to 4.5 points per player-game for public models. The bigger question is the baseline: a running average alone scores about 3.85, so a headline MAE without a baseline comparison is close to meaningless.
How is WNBA player point projection accuracy measured here?
Every player point projection is locked before tip-off, scored against the official box score, and rolled into public accuracy metrics: MAE against a naive baseline, within-1/2/3 rates, and band coverage checked against its stated confidence.
Why do projections miss badly sometimes?
We decompose every big miss: about 86% are rate-driven (hot or cold shooting, blowouts changing usage), not minutes-driven (measured across four seasons, 2026-08-06). Those games are what the ranges exist for.
Can projections beat sportsbook lines?
MAE parity with the market is not an edge. The only honest test is beating break-even when disagreeing with the closing line, over hundreds of tries. Our model-vs-market record is published and is currently far too small a sample to claim anything.
All figures are measured against real box scores by the same pipeline that ingests the games. Information only, not betting advice. Methodology guide.