How we got there: Does the workout matter?

We asked: where does a pre-draft workout carry information about an NFL career, given that draft position already prices it?

The sample is every drafted combine athlete from the 2000 to 2020 classes in the public combine_career view: workout scores next to whole-career outcomes, with washouts counted as zeros. Rates condition on 16+ NFL games.

The baseline first: draft position against career value. The market is good at this, which is what makes the question hard.

Where the workout still knows something: edge. Flight Score against career sacks per game. Published run: r 0.33, and it survives controlling for draft slot in both halves of the sample.

Where it knows nothing: receiver speed. Speed percentile against career receiving yards per game. Published run: r 0.018. Flat as the hashmarks, while the same number moves draft boards at r 0.25.

The floor at running back: Flight Score against career scrimmage yards per game. The raw link ties edge for strongest in the study, but it climbs through the third quartile and flattens: bad testing disqualifies, elite testing adds nothing. Among rounds 1 and 2 backs the correlation is 0.002.

The partial correlations beyond log draft pick, the train/holdout split, and the quartile hit-rate tables need scripted analysis; spec and results are in the article backtest block. Everything here forks.