Basketball Projections

How the score works

Both display modes show the same per-game averages in the category columns — never full-season totals. Per-game mode ranks category impact across all 230 players in the model pool without an availability penalty. Season value keeps those averages but ranks by a TOTAL multiplied by projected GP ÷ 82, so missed games directly lower a player's ranking — the games-played factor lives in the order, not in the averages. Turnovers are inverted (fewer is better) and count at half weight by default; every other active category counts at full weight. The Category weights panel above the board can re-weight any category from 0 (a full punt) to 2×.

Projected games come from a five-season durability history with recency weighting — the most recent season counts most — where developmental low-minute seasons (healthy scratches from before a player earned a rotation spot) count at minimal weight while injury-caused low seasons count in full; a healthy-starter prior covers short histories, and current-injury or veteran load-management adjustments apply on top. Where the result diverges materially from independent preseason expectations, it is hedged halfway toward them. The team overlay uses the 2026–27 team assignments in the projection snapshot, treats offense as a fixed 90-FGA pie, and redistributes opportunity within each roster. Young high-upside players receive a modest development curve; manual role changes only alter opportunity, not efficiency.

Projection sets

Two sets — frozen preseason and rest-of-season (ROS). The preseason set is frozen: it is the board exactly as published before opening night, it never changes afterward, and it doubles as the prior for the ROS set. The projection-set toggle above switches everything — rankings, TOTAL, category columns, the CSV export and the league projector — between the two sets. ROS re-blends each player's per-game line as real games accumulate: ROS = (preseason × W + actual × GP) / (W + GP) per category, where GP is the games the player has actually played and the prior strength W is 25 for counting stats and 35 for FG%/FT% — early-season noise moves the line slowly, and real trends take over as the sample grows. Actuals come from official league stat feeds and are folded in at each weekly refresh.

Minutes, roles and injuries in ROS. If a player's actual minutes differ from his projected minutes by more than 4 per game across a sustained 10+ game stretch and there is an identifiable cause — a trade, a teammate's injury, a rotation change — his ROS minutes move toward actual (a 50/50 blend until 20 games played, then actual outright), his counting stats rescale by the minutes ratio, and his team's lines re-normalize inside the same fixed shot pie so team attempts stay constant; the reason is noted on his row. A player who is out counts zero expected games in the weekly projector and a questionable player counts half; a season-ending injury cuts expected remaining GP to the games already played. ROS season value is ROS per-game TOTAL × expected remaining GP ÷ 82, where expected remaining GP is the preseason projection minus games already missed, adjusted for current status.

Where ROS stands today: the season's first games are Tue Oct 20, 2026, so every player has 0 games played and the ROS set is — by construction — identical to the frozen preseason set. The two sets start diverging with the first weekly refresh after opening night.

Ratio categories: FG% and FT% are volume-weighted as (player percentage − pool average) × attempts, then z-scored. Because the base source omits attempts, FGA and FTA start as estimates derived from projected points, threes, efficiency and a position/role free-throw rate — then they are cross-checked against published reference attempt volumes (below). Attempts that are not cross-checked remain directional estimates.

Full 9-category cross-check: Cross-checking every category against independent reference projections… Each matched player's full per-game line — FG%, FT%, 3PM, PTS, REB, AST, STL, BLK and TO, plus FGA/FTA volume — was compared with independent reference projections for 2026–27, captured October 16, 2026. Before any revision, the two sets of numbers tracked closely: correlations of 0.94–0.99 in eight of nine categories (steals lowest at 0.88), mean absolute differences of 9–14% in the counting stats and under a point in each percentage (FG% 0.89 pts, FT% 0.88 pts). Where a category diverged beyond the comparison band — 12% for counting stats and attempts, 1.5 points for FG%, 2 points for FT% — it was revised 65% of the way toward the reference number, the same convergence as the attempts pass. What did not change is the scoring: the reference's own total weights turnovers at 0.25× and applies a 20% GP penalty, and neither is used here. Ours keeps turnovers at 0.5× and our five-season GP model — recency-weighted, developmental low-minute seasons at minimal weight, and a halfway hedge toward the reference only where the two diverge by more than eight games (reference GP appears in each player's tooltip — e.g. Anthony Davis 54 reference vs 46 ours, Kon Knueppel 60 vs 65, Coby White 62 vs 63, Darius Garland 68 vs 60) — and volume-weighted z-score TOTAL in both views. Raw stat lines were revised toward the reference where the reference line looked better grounded; the ranking methodology remains ours.

Attempts cross-check (earlier pass): Cross-checking attempt volumes against published reference volumes… This first pass used only publicly visible numbers; the full cross-check above supersedes it where the two overlap. Where a player's derived FGA or FTA differed from the reference's minutes-scaled published volume by more than 12%, the estimate was revised 65% of the way toward the reference figure (80% for the three players whose full 2026–27 reference lines are publicly visible). Where the derived estimate was already within 12% of the published volume, it was kept unchanged. Player markers show the outcome: cross-checked, or attempts est. where no reference volume was available. Volume-weighted FG%/FT% impacts and TOTAL are recomputed from the revised attempts in both views.

League tools

The Season Projector, Weekly projector and Waiver Wire are league tools, not rankings: they take this board's per-game lines — from whichever projection set the toggle has active — and apply them to one league's rosters, format and current week. Those league facts (identity, settings, a default team, week dates, rosters and matchups) are held as a single structured input, so a different league could be supplied without restructuring the page. A Your team picker in the league header sets whose perspective the tools open from; the Weekly tab's waiver-claim simulator and the Waiver Wire's best-claims ranking test hypothetical add/drop pairs for that team with the same weekly engine — viewer what-ifs, never real league transactions. The Season Projector compares full rosters per game, with round-robin standings and a roto overview; the Weekly projector multiplies ROS lines by expected games from the real schedule for the current week, with a day-by-day view; the Waiver Wire lists every pool player not on a league roster, with his weekly value in the same week window. Each league tab is headed by the league context — league name, the user's team and the current week — so the tools never read as generic rankings.