2026-27 volume-corrected nine-category board

Up: 2026-27 Fantasy Basketball Encyclopedia

Related: 2026-27 fantasy basketball rankings · Projection engine defects and corrections · Category correlations and roster construction · 2026-27 category rankings · 2026-27 fantasy basketball projection methodology · Punt strategies for nine-category leagues

Why this board exists

2026-27 fantasy basketball rankings carries an honest warning that "many dossiers still lack explicit FGA and FTA, and the current parser has used proxy volume in those cases." That understates the problem. The build script fabricated attempts for 429 of 549 players, including 160 of the top 200, using fga = pts / 2.15 and fta = pts / 4.8.

FG% and FT% are two of the nine categories, and they are ratio categories: a player's impact is not his percentage but his percentage weighted by how many shots he takes.

\[\text{impact} = \text{attempts} \times (\text{pct} - \text{league pct})\]

Fabricating the attempts term therefore corrupts two ninths of the board. The proxy fails hardest exactly where it matters most, because free-throw volume is not a function of points scored — it is a function of how a player draws fouls. Giannis Antetokounmpo takes 11.6 free throws per game; the proxy assigned him 5.8. His .665 shooting was being priced at half its true cost.

How the repair was done

Real 2025-26 per-game FGA, FTA, FG%, FT%, 3PA and MPG were recovered for 262 players from ESPN and StatMuse, then validated two ways: every row had to satisfy the scoring identity \(\text{PTS} = 2\,\text{FGM} + \text{3PM} + \text{FTM}\) (238 of 238 non-rookies reconciled within 0.45 points), and 137 rows were cross-checked against career tables built independently by a separate research pass (134 agreed on both games played and points; the 3 disagreements were mid-season-trade splits where the recovered figure was the more complete one).

Projected attempts were then derived rather than guessed:

\[\text{FTA}_{proj} = \text{FTA}_{25\text{-}26} \times \frac{\text{MPG}_{proj}}{\text{MPG}_{25\text{-}26}}\]

\[\text{FGA}_{proj} = \frac{\text{PTS}_{proj} - \text{3PM}_{proj} - \text{FT\%} \times \text{FTA}_{proj}}{2 \times \text{FG\%}_{proj}}\]

Free-throw rate travels with minutes; field-goal attempts are then solved from the scoring identity, so the projected shot profile reproduces the projected points exactly. Nothing is invented. Of the 262 players in the pool, 54 already had stated volume, 194 were repaired with recovered data, and 14 rookies without an NBA season fall back to the pool-median free-throw rate and are flagged estimated in the export.

The same pass also removes a +0.5 x 3PM term the build script applied to the points model, which contradicts the Yahoo default formula the 2026-27 points-league rankings note documents. See Projection engine defects and corrections.

What moved

The elite tier barely moves, because most top-20 players already had stated volume. The damage was concentrated in rounds three through ten, which is precisely where drafts are won.

Fell — bad percentages on more real volume than modelled

Rose — good percentages, or bad percentages on far less volume than modelled

The direction is systematic, not random. The proxy fta = pts/4.8 understates free-throw volume for foul-drawing bigs and downhill guards, so their FT% damage or benefit was underpriced; it overstates volume for spot-up shooters, so an excellent but tiny free-throw sample was being treated as a real asset. Payton Pritchard shoots .889 from the line on 1.6 attempts — nearly worthless as a category contribution, and the old board paid for four times that volume.

Reading this board

The correction fixes the ratio categories and the totals rank. It does not fix games-played forecasts or minute allocations, which are inherited unchanged from the July 18 model.

The totals rank was a sign error

The original engine computed its totals column as \(z \times \sqrt{GP/82}\). Applied to a signed z-sum this runs backwards for every below-average player: 140 of the 262 ranked players carry a negative z, and multiplying a negative number by a fraction makes it larger. The result was that 52 players ranked better for being unavailable — Chaney Johnson gained 41 places on a 25-game projection, Klay Thompson gained 10 on 58 games.

corrected_total_rank now multiplies every category by projected games played and re-standardises across the pool, which is what a totals rank actually is. The correlation between games played and totals-rank gain is now \(+0.86\), where before it ran the wrong way. The biggest totals losers are exactly who they should be: Jimmy Butler III at 30 projected games falls from 93 to 254, Joel Embiid from 24 to 113, Anthony Davis from 23 to 80.

Use corrected_rank for head-to-head per-game leagues and corrected_total_rank for roto or any format where season totals decide the standings.

Use this as the nine-category draft board. Use 2026-27 fantasy basketball rankings for the editorial reasoning and the per-player scenario trees, and treat any disagreement between them as a flag to open the dossier rather than as a settled verdict.

Full export including every column and the volume-source label for each player: Data/player-projections-corrected.csv, regenerated by Data/recompute_corrected.py.

Corrected top 150

Table refreshed 2026-07-19 (second pass): all 533 dossier projection cards re-parsed as the new base (Data/parse_cards.py), repairing the Orlando rows the original snapshot parser dropped and carrying the Dort/Nembhard/Risacher trade roles; nineteen games-played corrections from the availability audit applied on top.

Delta is old rank minus new rank, so a positive number means the player rose. Arrows mark moves of ten or more slots. FG and FT columns show the projected percentage and the projected attempts behind it.

Open questions

• The 14 rookies without NBA volume use a pool-median free-throw rate; a position- and archetype-specific prior would be better, and college free-throw rate is the obvious input once it can be recovered without hitting anti-bot blocks.
• Free-throw rate is assumed to scale linearly with minutes. For players changing role substantially — a bench scorer moving into a starting creator job — rate per minute should rise, not hold, so their FTA is probably understated here.
• The pool mixes players with real or recovered volume against a small remainder still carrying proxy values, which shifts the mean and standard deviation of both impact terms slightly. Recovering volume for the full 549 would remove the last of the contamination.
• Should the reference rates stay at the engine's .472 and .780, or move to the actual 2025-26 league averages? The choice shifts every ratio impact by a constant and is worth stating explicitly rather than inheriting.