Larry Allen was the 46th pick of the 1994 draft, taken by Dallas out of Sonoma State, a Division II school in Sonoma County, California, with a football program that has since been discontinued. He started 197 games for the Cowboys, made eleven Pro Bowls, was named a first-team All-Pro seven times, and was inducted into Canton in 2013. By total residual z-score he is the largest small-school overperformance in our 1980–2025 dataset, at +31.93 z-units against the pooled pick × position × yos baseline.
The original residual audit flagged rz_cgs × college_tier with η² = 0.032, a real-looking signal that picks from non-power conferences were systematically beating their baseline. The pre-print version of this essay was going to lead with "small-school sleepers as evaluative skill." Then we fixed the calibration bias on the y_cgs outcome, re-ran the trend test, and the η² collapsed to 0.00026 with p = 0.42. The aggregate small-school effect is not statistically significant.
What's actually going on
In the post-Phase-1.5 data, the mean career-starts residual is essentially flat across all four college tiers:
| college tier | n picks | mean rz_cgs |
|---|---|---|
| power | 5,781 | +0.006 |
| mid-major | 3,216 | −0.016 |
| small | 1,823 | +0.010 |
| unknown | 9 | +0.080 |
No tier-level effect. What persists is GM-level dispersion. Some GMs cluster their hits in small-school picks; most don't. Three names lead the pack:
| GM | small-school hits | cumulative total_z |
|---|---|---|
| Jerry Jones (DAL, 1989–) | 15 | +78.87 |
| Bill Polian (BUF/CAR/IND/CHI, 1984–2011) | 8 | +19.97 |
| Les Snead (LAR, 2012–) | 7 | +18.47 |
| Jon Robinson (TEN, 2016–2022) | 7 | +14.70 |
Jerry Jones is the runaway leader. Twenty years of Cowboys drafts that pulled marquee starters out of programs nobody else was watching: Larry Allen (Sonoma State), DeMarcus Ware (Troy), Jared Allen (Idaho State, drafted by Kansas City but listed as the prototypical small-school hit), and a long list of mid-tier OL and DB rotational pieces. The pattern is consistent across decades and across the Cowboys' three drafting eras (Aikman-era, mid-tenure rebuild, current era).
What we make of this
The lesson isn't that small-school picks are sleepers. It's that finding talent at small schools is a skill that varies across GMs more than it varies across schools. The school produces a thin pool of NFL prospects; the GM has to evaluate the few that come through. Most GMs don't beat the position-and-pick baseline on those picks. A handful do, repeatedly.
Other notable small-school hits across the dataset, just to anchor what "small-school overperformance" actually looks like at the pick level: Jerry Rice (Mississippi Valley State, R1 #16, 1985 → SF), Brian Urlacher (New Mexico, R1 #9, 2000 → CHI), Antonio Brown (Central Michigan, R6 #195, 2010 → PIT), Tyreek Hill (West Alabama, R5 #165, 2016 → KC), Randy Moss (Marshall, R1 #21, 1998 → MIN), Shannon Sharpe (Savannah State, R7 #192, 1990 → DEN), Matt Birk (Harvard, R6 #173, 1998 → MIN). All Hall-of-Fame-or-borderline careers built on prospects most rooms wrote off.
Methodology note
The Phase 1.5 calibration that flipped this trend from "significant" to "noise" is documented in the residual decisions log. Briefly: a Jensen-correction gap on the Poisson y_cgs outcome was inflating negative residuals across the board, and small-school picks were slightly over-represented in the right tail of that bias. Removing the offset removed the apparent tier-level effect. The lesson is procedural: calibrate before you adjudicate.