Method · late_one_score_defending@v2

Late, one-score, defense closing

Fourth quarter or overtime with the offense tied or trailing by one score, and neither team above a 95% chance to win. The defense is trying to close the game out and the result is still genuinely in doubt. No time cutoff inside the quarter, because a defense that breaks with nine minutes left broke just the same.

The filter, exactly as it runs

qtr IN (4, 5) AND score_differential <= 0 AND score_differential >= -8 AND season >= 2022 AND vegas_wp IS NOT NULL AND vegas_wp <= 0.95 AND vegas_wp >= 0.05

Parameters

`max_deficit`	8
`first_season`	2022
`season_type`	ALL
`wp_ceiling`	0.95
`wp_floor`	0.05

Where this is arguable

The 95% band removes plays where the game was already decided, which is right, but it also removes the desperate late drives some readers most want counted.
Includes playoff games. Late-game situations in January matter most, but they are also the highest-leverage and least typical.
Requires FTN charting for the pass-rusher count, which is CC-BY-SA and starts at 2022. That license choice, not the football, is why this window cannot reach back further.
Differs deliberately from `clutch`: this is the defense's window and runs the whole quarter, where `clutch` is the QB's window and cuts at five minutes.

These are the objections I already know about. If you have a better one, or you think a different threshold tells a truer story, say so and I will re-run it with your numbers and publish what comes back.

Version 2. Older versions are never deleted, so a piece that citeslate_one_score_defending@v2 stays reproducible after this definition is revised. Source: Data: nflverse (nflfastR), CC-BY 4.0