Methodology

How to read the numbers

Every metric on this site — what goes in, how it is weighted, and where it stops being useful.

A scouting number is only useful if you know what it compares. Everything here is relative to a peer group — a 19-year-old Power 5 guard is measured against other Power 5 guards in that same season, not against the NBA and not against a 23-year-old in a mid-major conference. That is the single most important thing to know before reading any board on this site.

Star% — probability of becoming a top NBA contributor

Star%

Range 0–100 · shown in whole percent · same value on every page

A blend of two models that fail in different ways: Star% = 0.6 × comps + 0.4 × regression

Comps (KNN) — the share of star careers among the statistically most similar historical draft picks. Close to the individual case, but blind to age and context.

Regression (LR) — logistic regression trained on the 2015–2024 draft classes, AUC ≈ 0.81. It catches what the neighbourhood misses: a 23-year-old senior with a strong season finds flattering comparisons, but historically converts to stardom far less often.

The 60/40 split is not a tuning knob — it is the blend the stored values for the 2021–2025 classes were computed with. Any other weighting would make the years incomparable.

Reading the badge. The label (STAR, ROTATION, FRINGE, BUST) follows the number you see: Star only at 50% or above, otherwise the strongest of the remaining outcome classes. And the number shown is always Star% itself — never its inverse.

DCS v2 — Draft Ceiling Score

DCS v2

Range 0–100 · position-adjusted · one season, weighted toward what projects

A weighted combination of the six components below, with the weights depending on position — playmaking carries a guard, rebounding carries a big, and neither should be judged by the other's yardstick.

The name says ceiling, but the input is a single season. What pushes it toward potential rather than production are two corrections: younger players are credited for producing the same output earlier, and the level of competition is factored in.

PositionPUISTIACPMISEURBI
Guard 0.200.280.150.220.15—
Wing 0.220.200.120.150.200.11
Forward 0.220.200.150.120.180.13
Big 0.250.150.150.080.170.20

Above a raw score of 80 the scale compresses: each additional raw point adds only 0.45 to the final score. Without that, the top of every class bunches at 100 and the best prospects become indistinguishable from each other.

DCS v2 is not the same as Star%. DCS v2 describes how good a season was relative to peers. Star% asks a different question — how often players with this profile became NBA stars. A senior can post an excellent DCS v2 and still carry a modest Star%.

The six components

Each is a percentile within position, so 70 always means "better than 70% of comparable players", never a raw stat line.

PUI — Physical Upside Index

Wingspan 55% · standing reach 45%

Length, not size. Standing reach is the strongest single proxy for how big a player plays at the rim; wingspan carries defensive versatility. Requires measurements — from the NBA Combine, PIT or Basketball Without Borders.

STI — Shooting Talent Index

3P% 38% · FT% 27% · TS% 25% · 3-point attempt rate 10%

Percentages are stabilised before they are ranked: small samples are pulled toward the league mean, so 4-for-6 from three does not outrank a season of volume shooting. Free throw percentage is in there because it survives as a touch indicator when the three-point sample is thin.

A bonus applies to young players carrying an unusually high usage — creating your own shot at 19 is harder than finishing someone else's, and raw percentages punish it.

AC — Athleticism Composite

Steal%, block%, offensive rebound% — weighted by position

Athleticism as it shows up in the box score rather than in a vertical jump test. The weights shift with position: guards are read through steals (55%), bigs through blocks (65%).

PMI — Playmaking Index

Assist% 40% · assist-to-turnover 40% · usage 20%

Creation and care for the ball, weighted equally — a high assist rate paid for in turnovers is not playmaking. Usage enters with a small weight as a measure of how much of the offence actually ran through the player.

SEU — Scoring Efficiency at Usage

Draft boards: efficiency residual 50% · scoring volume 50%  ·  Player profiles: residual only

Efficiency drops as usage rises — that relationship is fitted across the whole pool, and what counts is the residual: how far above or below the expected efficiency for that workload a player scored. A high-volume scorer at average efficiency and a low-usage finisher at elite efficiency are separated correctly this way.

⚠️ One caveat we would rather state than hide: the draft boards add scoring volume to the residual at equal weight, the player profiles use the residual alone. Same name, two numbers. Consolidating them is on the list — until then, compare SEU within a page, not across pages.

RBI — Rebounding Impact

Total rebound percentage, by position

The share of available rebounds a player secured while on the floor. Percentage rather than per-game, so pace and minutes do not distort it. Carries no weight for guards.

Peer groups and tiers

Percentiles are computed inside buckets, never across the whole database:

PoolBucketQualification
NCAAPosition × conference tier × season≥ 10 games and ≥ 12 minutes per game
InternationalPosition × league × season

A bucket with fewer than 15 players produces no percentile at all — with a handful of players a rank carries no information. Those cells stay empty rather than showing a number nobody should trust.

NCAA tiers: Power 5 · High-Major · Mid-Major · Low-Major. International tiers run INT-1 (lowest) to INT-6 (EuroLeague). North America is separate: NBA and NBA-2 for the G League. Cross-tier comparisons are possible but always weaker — the same season in two different leagues is not the same evidence.

Reliability gates

A player who never attempted a three is not a 0% shooter — he is a player without data. Rate statistics therefore require a minimum volume before they are ranked at all:

MetricMinimum per game
3-point percentage1.5 attempts
Free throw percentage1.5 attempts
True shooting percentage5.0 field goal attempts
Block percentage0.5 blocks
Steal percentage0.8 steals
Assist-to-turnover1.5 assists

Points, assist rate, usage and rebound rates have no gate — there a zero is a real result.

Box-score terms

TermMeaning
TS%True shooting — efficiency across twos, threes and free throws
USG%Share of team possessions a player ended while on the floor
TRB% / ORB%Share of available (offensive) rebounds secured
AST%Share of teammate field goals assisted while on the floor
FTrFree throw rate — free throw attempts per field goal attempt
3PArThree-point attempt rate — share of shots taken from three
PERPlayer efficiency rating — per-minute box-score production

What these numbers can't do

Every metric here is built from box scores and measurements. That leaves out most of what decides a career:

  • Nothing here sees defence properly. Steals and blocks are events, not positioning. A great team defender who neither gambles nor blocks shots is invisible to AC.
  • Role is not talent. A player in a bad situation posts bad numbers, and the model cannot tell the difference between a limited player and a misused one.
  • The training data is 2015–2024 NBA draft picks. Undrafted paths, late-blooming international players and shifts in how the league values positions are underrepresented by construction.
  • Small samples stay small. Shrinkage and gates limit the damage — they do not create evidence that a 12-game season does not contain.

These numbers are built to narrow a list and to make the reasons explicit — which is exactly what film study is bad at. They are not built to replace the eye. Used as a filter rather than a verdict, they do the job they were designed for.

Start with the draft big board, the NCAA rankings or the international board — or read about who is behind this.