The combine is emptying out

Who still tests

Through 2016 a combine invitation meant testing — all but a handful of every class ran the forty. Today an invitation means weighing in. 64% of the 2020 class did the bench press; 14% of the 2026 class did.

Why every other trend here is suspect

Measured across the whole window, every group looks faster. Measured inside the five classes before the collapse, almost all of it disappears. Missing values are not random: the athletes who skip a drill are the ones who expect to test badly, so the median of whoever is left improves without a single athlete improving.

Holds up

the move survives in the years before the collapse, on a stable population

Mostly selection

a real move, but less than half of the full-window figure

Different population

the move is large enough, but too few of the group were measured, or too different a share of them, to credit it to the athletes

Selection only

nothing before the collapse — the whole gain is who stopped testing

No real change

smaller than this group’s own class-to-class swing

Windows are 2000–04 against 2015–19 and 2022–26, pooled over the athletes in each rather than averaged from yearly medians. We do not print a number we cannot defend.

• The full window says cornerbacks carry 0.08 less mass per inch of frame.

  In the five classes before the collapse, cornerbacks carry 0.01 less mass per inch of frame.

• The full window says quarterbacks run the forty 0.18s faster.

  In the five classes before the collapse, quarterbacks run the forty 0.02s faster.

• The full window says interior offensive linemen run the forty 0.12s faster.

  In the five classes before the collapse, interior offensive linemen run the forty 0.01s faster.

• The full window says safeties run the forty 0.10s faster.

  In the five classes before the collapse, safeties run the forty 0.02s faster.

The same trap on the Atlas Score

In 2011 this site could rate 93% of the class, and their median Atlas Score was 55.9. In 2026 it can rate 56%, and their median is 77.9. Read alone, that says 2026 is the most athletic class ever measured. It is not.

2011

55.9

rated 307 of 329

93% of the class

2020

46.1

rated 226 of 337

67% of the class

2023

60.2

rated 210 of 319

66% of the class

2026

77.9

rated 180 of 319

56% of the class

The Atlas Score is a percentile against an athlete’s own position group across all 27 classes, and it needs size plus two performance domains before it will rate anyone at all. As the drills empty out, the athletes it can still rate are the ones who chose to be measured. Since 2011 the rating pool has shrunk to its own best fraction, and the median climbs with it — it measures who showed up, not who is better. Before 2011 the pool was still growing, and the rise over those classes has a different and less suspicious cause: more athletes completing enough drills to be rated at all.

What actually survives

Almost nothing does — and what does is concentrated in the measurements nobody can opt out of. height, weight and PPI are recorded on effectively every athlete in every class, so their record is the only one clean enough to read straight. Every claim below is measured before the collapse, is larger than that group’s own class-to-class swing, and rests on a share of the group that did not move between the two windows.

• Weight

  4 of 11 groups lighter

  between 2 lb and 7 lb, in TE, EDGE, OT, S

• CBcornerbacks broad jump 3" fartherBroad jump measured on 81% then 81%
• TEtight ends weigh 5 lb lessWeight measured on 100% then 100%
• TEtight ends run the 3-cone 0.18s faster3-cone measured on 69% then 67%
• EDGEedge rushers carry 0.10 less mass per inch of framePPI measured on 100% then 100%
• EDGEedge rushers weigh 7 lb lessWeight measured on 100% then 100%
• CBcornerbacks rate 1.2 lower on Power IndexPower Index measured on 78% then 81%

The broad jump is the clearest example of why this page distrusts its own charts. It looks like a league-wide gain, but the share of each group jumping moved by as much as 19 points between those two windows — so more of the group jumped, rather than the group jumping farther. Only cornerbacks, whose coverage barely moved, clear the test.

About this data

• 2021 is pro-day data — no combine was held. It is plotted for context and excluded from every trend line and every stated change on this page.
• Some position labels collapsed upstream: no tackle-versus-interior split exists for 2020, and no safeties were labelled as such in 2023. Those lines break. We never draw across a hole.
• Class sizes are small — a typical group has 20 to 45 athletes in a year, quarterbacks 14 to 23. The heavy line is a five-class rolling median; the dots behind it are the single classes.
• Populations are combine participants only. Drafted players who never attended are in the atlas but never in a rate, a median, or a trend.
• Methodology has the full mapping rules, metric definitions and provenance.