Methodology

In category formats (9-cat, and 8-cat which drops turnovers), player value is standard z-scoring over the draftable pool for your league size, with volume-weighted percentages (a 92% free-throw shooter on ten attempts moves your team; on two attempts he doesn't). In points formats, value is per-game fantasy points under the platform's published default weights — ESPN (PTS 1, 3PM 1, FGM 2, FGA −1, FTM 1, FTA −1, REB 1, AST 2, STL 4, BLK 4, TOV −2) or Yahoo (PTS 1, REB 1.2, AST 1.5, STL 3, BLK 3, TOV −1). Custom weights aren't supported yet; presets keep the math server-side and honest.

Keeper verdicts compare a player's value against what picks in that round typically return, so a keeper is only ever as good as the pick it eats. Auction dollars re-denominate the same values: the draftable pool splits teams × budget exactly, a dollar per roster spot plus value above replacement, so every league dollar is priced and none are invented.

The decision tools (keeper, compare, auction, picks) run on last season's per-game numbers — verified, legible, and the right anchor for "what did I actually watch." The projected rankings run on our model: three seasons blended with recency weighting, small samples shrunk toward league averages, age adjustments fitted from the data, and a manual minutes layer for news. Before it shipped, the model was graded once against a pre-registered bar on a held-out season — it beat last-season carryover on rank correlation (0.777 vs 0.753) and per-category error (0.418 vs 0.426). The full backtest write-up, including where it loses, is coming; the next model version adds explicit uncertainty — bust probabilities, floor/ceiling ranges — and 2026-27 itself will be graded publicly in spring, whatever it says.

We publish derived values only, never raw NBA statistics. Estimates carry uncertainty; anyone selling you certainty about next season is selling.