All projects

Private, available to license

Augur

Does a fast first step at the NFL Combine mean a fast first step on game day? A study that fixed its rules before looking at the answer.

Category
Sports analytics
Role
Solo
Stack
Python, Polars, NumPy, matplotlib
Status
Private, available to license, 2026
Abstract illustration of data points rising along a dashed trend line.

What it is

Augur is a sports science study built for the NFL Big Data Bowl 2027. It asks one question about defensive linemen: does a player’s burst out of his stance in the Combine pass-rush drill tell you how quickly he gets off the ball in real games?

Scouts already have the 10-yard split. Augur asks whether the burst adds anything on top of it.

What it does

  • Measures each player’s first-step burst from Combine tracking data, after removing what his 10-yard split already explains.
  • Measures how quickly he gets off the ball on regular-season pass rushes.
  • Tests whether the burst improves predictions of game-day get-off. The baseline it must beat already knows the split, body weight, draft position, whether he went undrafted, and whether he plays on the edge or the interior.
  • Gives a plain verdict: supported, not supported or inconclusive.

How it works

The rules came first. Before any Combine number was compared with any game outcome, I wrote and locked a pre-registration: the question, the data, the players, the test and the rule for the verdict. Every later change is logged with a date, what I had seen at the time and how it could affect the result.

The study uses regular-season tracking data for defensive linemen from the 2023, 2024 and 2025 draft classes. It keeps straight-ahead rushes and sets aside stunts and drops, using a simple rule based on the rusher’s own path.

Each draft class takes a turn as the test group. The model learns from the other two classes and predicts the third, so every prediction is made on players it has not seen. A shuffle test sets the bar the burst has to clear, so the result is judged against chance and not against a threshold I picked.

Results

  • 25,278 regular-season rushes from 98 players in the 2023 to 2025 draft classes. 70 players meet the bar to enter the test.
  • A pilot on made-up outcomes behaved as it should. Pure noise was rejected. A planted effect was found in all three test groups.
  • Built-in checks: repeat builds give identical files, a deliberately meaningless “garbage” measure runs beside the real one, and every data join is counted.
  • The test runs once, under the locked rules. The verdict cannot be tuned after the fact.

Who it is for

  • Scouting and player personnel departments deciding how much weight to give Combine drills.
  • Sports science and analytics teams who want a template for an honest, pre-registered study.
  • Researchers and data journalists covering player evaluation.

What you get with access

  • Read access to the private repository: the analysis code, tests, the pre-registration and the full changelog.
  • The study notebook and the scripts behind the figures and checks.
  • The NFL tracking data is not included. It comes from the Big Data Bowl under its own terms.

Interested in Augur?

The code stays private. Ask for access and I will reply with terms.