Loading
Loading the next page.
Loading
MBStatsLoading
Loading
A living notebook by Riel St. Amand
The reproducible trail for the style archetypes. Every feature share ships next to every label in the public archetype views; the assignments are frozen mathematics.
Every cell re-runs in your browser against current data. Covers 1999 through 2025. Data refreshed 1 day ago.
We asked: can you say how a player plays without smuggling in how well?
The shipped features are style only, and every one is z-scored within its own season. Both public views, qb_archetype_season and wr_archetype_season, carry the full feature shares next to every label, so any assignment in the article can be audited down to the route mix. The cell below runs one axis of the quarterback style space live: designed running against scramble rate, the two features that separate the running archetypes.
Rigor check
Designed runs per game against Scramble rate
Pearson correlation, re-run in your browser on current data.
The assignments are frozen mathematics: the centroids are constants in the public view SQL, so a 2016 season and a 2025 season are labeled by the same math forever, and no future data can drift a label. The stickiness receipts (quarterbacks repeat their archetype year over year at 51 percent against a 20 percent chance rate) and the era comparisons need season self-joins, described in the article backtest block.
Built on data from the nflverse project.