(1) Is fourth_stops still zero for 2026 defences, and does the warehouse agree with its source in every season? (2) Did the zeroed stops give the only DEF accuracy number the wrong sign, and was that sign ever identified? (3) Does any OTHER DEF scoring event have the same shape (a live-season column the feed does not fill)? (4) Could the zeroed stops have reached the served projection?

Owner lane: premise

Question

(1) Is fourth_stops still zero for 2026 defences, and does the warehouse agree with its source in every season? (2) Did the zeroed stops give the only DEF accuracy number the wrong sign, and was that sign ever identified? (3) Does any OTHER DEF scoring event have the same shape (a live-season column the feed does not fill)? (4) Could the zeroed stops have reached the served projection?

Prediction

Stated before the run: (a) the premise as worded is now FALSE -- 2026 fourth_stops is nonzero and equals pbp fourth_down_failed by defteam on every settled team-week, and 2016-2025 also match exactly; the negative fixture (2026 zeroed in memory) goes red on every team-week with a stop. (b) The sign flip reproduces arithmetically (zeroing moves the actual down by 2 x 0.53 = ~1.06 pts) but BOTH signs sit inside a game-cluster 95% interval that spans zero, so the claim 'wrong sign' was never identified; the materially wrong quantity was the MAE, not the sign. (c) No other DEF event column is below half its 2016-25 minimum in 2026. (d) The served stops expectation never read 2026 rows: fit hist is season <= 2025 and DEF_FORM_KEYS has no stop feature. REFUTED if any 2016-2026 team-week disagrees with pbp, if the bias interval under either actual excludes zero, or if served stops vary by week in a way only 2026 rows could produce.

Finding and verdict

TESTED. The premise as worded is false since 2026-09-23 and the fill persists through tonight's refresh (0 mismatches vs pbp over 5,344 team-weeks, 2016-2026). The 'wrong sign' reproduces but was never identified: bias -0.55 [-2.02, +0.84] with stops vs +0.51 [-0.90, +1.83] without. What the zeroed stops actually did was flatter DEF MAE by 0.455 pts [+0.08, +0.87]. The forecast never read the zeroed rows. Across all 10 DEF scoring events, no other live feed gap: ret_tds is a witnessed true zero; extra_point_return has no column in any season and is worth 0.0056 pts per team-week.