Latent Utility Model of Attribution

LUMA

An open data, community-driven initiative to quantify NBA player impact.

01 — attribution

Latest research

November 5, 2026 · LUMA Basketball Research

Analyzing SGA and the Value of Soaking Up UsagePoints Above Cost

On a night where injuries forced OKC to play rookies for 44 minutes, and with only Ajay as another shot creator, SGA scored 43 points on 35 shots (−4.9 rTS).

(Efficiency cannot be looked at in a vacuum. It has to be viewed through the lens of opportunity cost.)

Read the post →SGA on ARC →

OKC half-court offense with and without SGA

Season	With SGA	Without	Rank of 30	On − off
2026–27*	118.3	94.1	30th	+24.2
2025–26	119.0	106.7	26th	+12.2
2024–25	119.9	105.7	25th	+14.2

At Cleveland, Nov 2: 43 points, 35 FGA, 12 FTA. 53.4% TS against a 52.0% PAC break-even.

Points per 100 half-court possessions. Without: games he played. *2026–27 through Nov 4.

Every model on this site is written up before it is shown.All research →All games →

ARC v3.1 · LOKI v3.3 · Changelog · Downloads

Research

Every model on this site is written up before it is shown.

Research note · 2026-11-03

Analyzing SGA and the Value of Soaking Up Usage

On a night where injuries forced OKC to play rookies for 44 minutes, and with only Ajay as another shot creator, SGA scored 43 points on 35 shots (−4.9 rTS).

Research note · 2026-10-10

Possession by Possession Fundamentalism: Offensive Conversion and the Distribution of Player Value

Possessions are a scarce, finite resource, and thus player value must be derived from the cumulative margin added within and across each possession played.

Article · 2026-10-09

Introducing ARC v3: An Expected Box Value Prior RAPM (EBV RAPM)

Contemporary all-in-one impact metrics often regress box score actions onto RAPM to find their weights.