College to NBA
Every 2026 draft pick and the 2027 draft board, scored from college numbers and draft slot, each with an 80% range. A Ghost Screen Analytics model, not an NBA stat. The tests were written before the model was run.
The scale. Every projection is a LEBRON rating (BBall Index, points per 100 possessions better or worse than an average player), averaged over NBA seasons 2 to 4. 0 is an average NBA player, +2 or more is a clear starter, and about -2 is replacement level. A player who barely plays or never reaches the NBA counts as -2, so most late picks project below zero.
How to read it
- Projected NBA rating
- The expected level in NBA seasons 2 to 4, measured with LEBRON (a public all-in-one rating from BBall Index, in points per 100 possessions). Zero is an average NBA player, about -2 is replacement level (a player who barely plays or never makes the league counts as -2), and +2 or more is a clear starter.
- Slot only
- What the draft position alone would have predicted. The hollow circle on each range.
- Model
- Draft position plus the final college season: age, height, class year, minutes share, usage, shooting splits, rebounding, assists, turnovers, steals and blocks. The dark tick on each range.
- 80% range
- The span the model expects to hold for about four of every five players. It is wide on purpose. Nobody in a class of 60 is a sure thing.
2027 caveat. The 2027 board uses the public Tankathon big board rank as the draft slot and last season's college numbers (the 2026-27 season has not been played). Freshmen, players coming from outside college basketball and anyone without a 2025-26 college season are scored from slot alone and tagged slot only. Treat that class as a rough first look.
The projections
What it is built on
Sample: players drafted from 2008 to 2022 with a Division I season (international and non-college picks are left out), their final college season from Barttorvik, and their NBA rating over seasons 2 to 4. The test predicted each class from 2016 to 2022 using only earlier classes whose three NBA seasons had finished, so nothing in a test was seen in training.
The rules, the code, the data and every result, including the first run, are in the college-to-nba folder on GitHub. A passed bar is not proof the model will keep working: it is seven classes and a few hundred players.