The Profile lab and “Built like the stars” rest on every QB, RB, WR and TE who entered the NFL from 2000 on: their age, combine numbers, draft slot and every season’s fantasy finish. A star here is a player with at least one top-12 RB/WR (top-6 QB/TE) season. These are the findings, in plain words.
The draft slot already knows most of it
- Where a player is drafted is by far the strongest signal. NFL teams draft on speed, size and athleticism, so those are mostly priced in.
- Stars do stand out on some measurements: speed (40 time and speed score) at RB and WR, nearly every drill at TE (69% of star TEs were in the top third of our athletic score, against 34% of all TEs), and youth everywhere.
- But players who matched a position’s whole star checklist became stars about as often as their draft slots predicted: RBs 34% against 33%, WRs 19% against 21%, QBs 20% against 18%. Late picks who matched did no better than their slots either.
- Body types (“builds”: groups by height, weight, speed and age) tell the same story: young WR builds produced stars two to five times as often as older ones, but no build beat what its draft slots predicted.
Age and the rookie season
- Age: younger players at the draft hit far more often (a top-24 season in the first three: RBs 27% for the youngest third against 8% for the oldest), but the draft slot carries most of that too: among players picked in the same range the edge shrinks to a few points, within chance. It’s most useful before the draft, when the slot isn’t known; the college prospect pages show how a prospect’s draft-day age compares with the position’s stars.
- The rookie season: for RBs and WRs, a top-third rookie year in points per game is the clearest signal after the draft (for QBs and TEs it adds little to the draft slot). A rookie’s share of his team’s targets or touches adds nothing once you know his points.
Breakout age, hand size and “deep learning”
- Breakout age (the first college season with a big share of the team’s receiving) separates hits in published studies, but those mostly don’t account for the draft slot. We can’t test it yet: it needs college stats going back to 2000 (the free CollegeFootballData key would unlock it).
- Hand size and arm length aren’t in the app: the public numbers trace back to NFL.com, whose terms don’t allow compiling them, and the sites that sell them (RAS, MockDraftable, PlayerProfiler) would need a licence. Published work on QB hand size found no reliable link to success.
- A bigger model doesn’t help: boosted trees, random forests and neural networks trained on the same players (to predict a top-24 RB/WR or top-12 QB/TE season in the first three) didn’t beat the Profile lab’s simple model (a few hundredths either way, within noise). With a few thousand players and a few hundred stars, there isn’t more for them to find; better data (college production) is what could change that.