On the board path as it runs now (latest_rows -> PM.current -> lines_for), which stage produces each all-zero line: row choice/emptiness, a membership join, p_play, or the share model? And is the object that locks each week the object the board shows?

Owner lane: construction

Question

On the board path as it runs now (latest_rows -> PM.current -> lines_for), which stage produces each all-zero line: row choice/emptiness, a membership join, p_play, or the share model? And is the object that locks each week the object the board shows?

Prediction

Prereg research/scientist/prereg/2026-09-23T2120Z-h114-row-choice.md: control >= 90% agreement; >= 90% of zeros line-stage not p_play; every row from 2025; >= 60% of zero rows empty (falsifier: < 60% -> row-choice story fails); <= 15% team mismatch; >= 50% of zero men who played 2026 carry a non-zero 2026 share; pinning to 2026 gives fewer zeros.

Finding and verdict

PREDICTION WRONG ON THE MECHANISM, MEASURED. 215 of the 222 all-zero lines on the HEAD board replay (97%) are D-95's off-roster stamp (confirmed_active=0: no player_week 2026 on_roster=1 row, matched by resolved name), not an empty row and not p_play. The name key and the player_id key agree on all 215, so there is no crosswalk defect here. 143 of the 215 have a 2026 row and were never on_roster=1 (reserve/PUP/unsigned-to-53), 72 have no 2026 row. Only 7 are model-stage zeros (6 at raw <= MIN_RAW_FOR_SCALE), and 6 of those 7 played in 2026: that small class is H-114's graded 'zero share'. The graded wk2 zero men were NOT D-95 zeros: 17/21 were on_roster in wk1. POST HOC, CONSTRUCTION: pred_2026_w03..w18 were all written once, 2026-09-15 20:40-20:41Z (a312a9e2), and not since. 12 men are locked-to-be zero in the w03 table and non-zero on the 09-20 board, and 10 of them are wk2 graded zero-line men. Week 3 locks from the stale build unless a writer runs first.

Reasoning

Three lanes had asked what the model knew about these men; nobody had asked whether the row was in the frame as a live player at all. The stage census answers that it mostly is not, by design (D-95), and that the object that locks each week is an eight-day-old build.

Next action

Filed a construction row: the week tables w03..w18 are a single 09-15 build that the board has moved past, and w03 locks from it. Test at wk3 grading: graded men locked at 0 in w03 against the board's number at lock. D-95's 18-week reach (an IR man is 0 in the week he returns) is the other half; it needs a week in which a reserve man returns.