← Model Journal/
New Model

The Rookie Identity. Building a player from signals alone.

4 June 2026v27.1
The Rookie Identity · v27.1

Building a player from signals alone.

Every NBA rookie enters the league as an unknown. No career stats. No track record. Just signals.

The Bourne Identity concept applies directly: a player wakes up with no memory of who they are. We have to build their identity from scratch — before they've played a single NBA minute. The Rookie Identity model does exactly this using three data sources.

The three data sources

What we know before tip-off.

NBA Combine
Wingspan. Lane agility. Max vertical. Standing reach. Body composition. The physical blueprint that predicts defensive range, rebounding, and athleticism.
327 player profiles · 2000–2025
KenPom-Adjusted College Stats
Raw college stats adjusted for strength of schedule. A 20-point scorer in the SEC is not the same as a 20-point scorer in the Sun Belt. 8,314 team-seasons of context.
92.6% match rate to draft class
25 Years of Historical Comps
Find the 10 most similar historical players by physical profile and college production. Their actual NBA careers define the projection range — not a formula, but a distribution of real outcomes.
550 complete training examples · 2000–2022
Model accuracy · within 20% · holdout 2018–2022

The model is honest about what it can and cannot do.

CategoryAccuracyNote
FG%90.5%College efficiency translates well
FT%94%Most stable rookie stat
STL48.8%Moderate — defensive instincts carry over
BLK44%Physical profile helps
PTS41.7%Minutes allocation is the unknown
REB38.1%Role/position dependent
AST35.7%Playmaking role hard to predict
TO28.6%NBA defense creates new turnover patterns
3PM19%Hardest rookie stat in sports analytics

Percentage stats (FG%, FT%) are highly predictable from college efficiency — a player who shoots well in college generally shoots well in the NBA. Counting stats (PTS, REB, AST) are genuinely hard. Minutes allocation in year 1 is the dominant variable we cannot predict before a season starts.

Three-point shooting at 19.0% is the hardest rookie stat in all of sports analytics. The NBA three-point line is longer, the defense is better, and college 3P% has weak predictive validity for NBA 3P%. College shooting mechanics frequently don't survive contact with elite NBA defenders.

The honest output

Ranges, not point estimates.

This is why we show ranges, not point estimates. “This player's historical comps produced between 8 and 22 points per game in year 1. The median was 14.2.” That is more honest than projecting 14.2 with false precision.

The comps method respects what we don't know: NBA coaching decisions, team roster construction, and the randomness of injuries in year 1 are all outside our model. What we can do is show you the full distribution of outcomes for similar players — and let you decide what to do with the uncertainty.

VJ Edgecombe — Example output
Projection range: #52–#118 overall
Comps: Donovan Mitchell · Aaron Gordon · Anthony Edwards
10–22 ppg
Pts range
MEDIUM
Confidence
87%
Comp match
View the Rookie Identity page →
← Back to Model Journal