What recruiting data actually says
Every college signee since 2010 — 29,802 of them, from 5,852 high schools to 263 colleges — measured against what their own composite ratings predicted. Not scouting. Not a ranking. What happened next.
Does the high school develop players?
1.03×
No — it explains nothing.
Across 1,064 schools with 8+ college signees, the spread in draft outcomes around what star ratings predict is what coin flips at those probabilities produce. A season with no school effect at all beats this number 31% of the time.
170 pipelines, ranked →Does the college?
4.21×
Yes — a great deal.
Same test, same yardstick, 131 programs with 100+ signees. The spread is 4.2 times what chance produces, and 15 programs separate from their own recruits' ratings after correcting for all 131 tested.
The scorecard →The same recruits, different programs
Take only the 2,774 recruits the services rated in the top 10% of their own signing class — identical evaluations, by the industry's own numbers — and ask what each program turned them into. 28% of them were drafted overall.
Top five and bottom three of the 31 programs that signed 25+ top-decile recruits in the 2010–2020 classes. The spread inside this one band alone is 4.3× what chance produces — so this is not the rating curve being fitted badly, it is two recruits the industry rated the same coming out different.
Every program against what its recruits were rated
131 programs that signed 100+ rated recruits in the 2010–2020 classes. “Expected” is those recruits' own composite ratings and positions, fit within each signing class and summed — the same model the high-school pages are measured against.
131 of 131 programs, ranked by gap vs expected. Only 15 separate from their own recruits' ratings once all 131 tests are corrected for — a positive gap on its own is not a finding.
2010–2020 signing classes, each with the full five drafts behind it. “Gap” is picks minus what those recruits' own composite ratings and positions predicted. A verdict is only given where the gap survives correcting for all 131 programs tested. What this does not say.
What a star rating actually buys
29,806 rated signees in the 2010–2020 classes, 2,045 of them eventually drafted. Two things read straight off the bars. The top is a cliff, not a step — and the middle of the four-star range is close to flat, so “high four-star” and “low four-star” are roughly the same recruit.
Does the ranking inside a tier still tell you anything?
At the very top, a great deal. Split the 0.98+ band at its own median national rank (#20) and the better half is drafted 69% of the time against 45% for the rest — 24 points.
In the 0.89–0.92 band the same split is worth 1.5 points. So the ordering carries real information where the sample is thinnest and almost none through the middle, which is where most of the arguing happens.
What this does not say
It is not a coaching grade. A program's number mixes development with playing time — a four-star behind two other four-stars does not start, while the same recruit at a thinner program plays as a freshman and gets three years of tape — with how often scouts attend, and with a program simply identifying better inside a rating band, which is evaluation rather than development. This data cannot separate them.
Credit goes to whoever signed him. 169 of the 1,727 drafted signees this can be measured on (the 2,045 picks include some matched by name and college, which by construction cannot be transfers) (10%) were drafted out of a program that did not sign them, and that share runs 4% in the 2010 class to 28% by 2020. The attribution gets less honest every year. Crediting only the picks who never moved still leaves the spread far above chance.
Better-recruiting programs still carry bigger residuals (r = 0.44 between a program's mean recruit rating and its gap). Under the model this page used until 2026-09-10 that correlation was 0.79 and the headline was 11×, because rating entered as a straight line and could not bend to the real curve. It is binned and cross-fitted now, and the remaining correlation is a limit on the reading, not a result.
Service academies are marked. Air Force, Army and Navy signed 1,001 recruits in this window and had none drafted — a military commitment, not a football outcome.
The evidence underneath
Descriptive hindsight, not a forecast, and not betting advice. Draft outcomes cover the 2010–2020 signing classes — each has the full five drafts behind it — over 29,806 rated signees, 2,045 of whom were drafted (6.9%). Recruit ratings and signings: CollegeFootballData.