How we got there: What survives an NFL offseason

We asked: of everything you can measure about a WR season, what carries information into the next one?

The sample is every WR consecutive-season pair from 2016 to 2025 with at least 40 targets and 8 games in both years, from the public player_season_pairs view. Every cell below re-executes against current data when you open this notebook.

First ask: does production persist? Prior receiving yards per game against next season. The cell below runs the correlation live; in our published run it explained 46 percent of next-season variance, the best single predictor we found.

Second ask: does role predict? Share of team air yards against next-season production. Published run: 31 percent of variance, with target share at 38. Role is a forward-looking stat.

The twist: stable is not predictive. Average depth of target repeats year over year (first cell below) yet predicts next-season production at almost exactly zero (second cell). A stat can be a fact about a player without being a forecast about him.

The running backs: yards per carry in consecutive seasons, 80+ carries both years. The cloud is the finding.

The article also splits the decade for an out-of-sample check (the r-squared ordering replicates; the breakout base rates do not, because the era owns them) and runs residual analysis beyond prior production. Those need a scripted analysis rather than single cells; the full spec and results live in the article backtest block. Fork any cell here to push further.