Review of 2026-09-23T1916Z-h114-all-zero-lines

Owner lane: Not recorded

Finding and verdict

WEAKENED

Review scope

SPLIT. UPHELD: every number the runner emits reproduces byte-identically (wk2 21 men / 36.0 pts, wk1 25 / 28.5 under the runner's join; A/B/C/U cells; E3 means; the six post-hoc share>=0.05 men and 29/35 below it) and the record's own verdict text ('mechanism NOT established', 'PARTIAL') is honestly scoped. DOES NOT SURVIVE AS WORDED: (a) the wk1 enumeration is not exhaustive: the name+team join drops 3 graded zero-line men who were served on a stale team, so wk1 is 28 men / 29.1 pts, not 25 / 28.5, and the record's 'A+B+C+U = zero-line graded n' normalization only closes on the join-defined set; (b) E1/E2 pool two different served objects: all 255 wk1 zero rows carry an EMPTY stat_line '{}' (no line emitted), all 242 wk2 zero rows carry an explicit computed all-zero dict, so 'a zero share vs a later stage' cannot be one question across the two weeks; (c) E3 has no interval and cells of n=1 and n=0.

Next action

If the wk1 count matters downstream, re-key wk1 on player_id or accept name-only-any-team with a team-mismatch flag and report the stale-team rows as their own class (3 played men here, 0.6 pts). Nothing here changes the served-board loss estimate (1.25-1.67% of graded points; the 28.5 vs 29.1 difference is 0.03 pp of 2288.9).