Review of 2026-09-26T0624Z-h195-mover-volume

Owner lane: Not recorded

Finding and verdict

UPHELD

Review scope

SURVIVES: exact byte-for-byte reproduction of the analyze step from the frozen learned_baseline and all four bound input artifacts; the production entry point (wf_std.py -> PM.fit -> share.feature_sql -> PM.lines_for) is used correctly with a strict season<S fit window and no ALL_FEATURES/SELECT * defect; the placebo control is a genuine, non-vacuous test and the planted-effect control confirms harness sensitivity; E1 is REFUTED (+0.0129, inside both its own CI-vs-zero test and the placebo band) exactly as the record states, on 1,935 mover / 6,307 stayer rows across 7 held-out seasons; E4/E6/E8 corroborate no room- or serving-path-independent mover effect in the walk-forward line. DOES NOT FULLY SURVIVE, but does not overturn the verdict: the record's own release note ('tempo... not imported by the line path') is contradicted by a later, independently-filed finding (H-291) that wf_std.py's output IS sensitive to tempo-table state, meaning the baseline's release identity is less completely pinned than the record claims -- though the affected rows (13,921/51,863, 2022-2023 only, sub-point-level medians) are a minority concentrated in 2 of the 7 held-out seasons whose individual E1 estimates already span zero with wide intervals, and the pooled E1 sits comfortably (delta 0.0129 vs a placebo half-width of ~0.035) inside the null band, so this gap is unlikely to flip the pooled REFUTED finding but is a real disclosure defect worth carrying forward.

Next action

File the tempo-table provenance gap as support for H-291 rather than reopening H-195: H-291 is already OPEN in the mechanical lane and names the fix (commit or revert playerweek/tempo.py, record the builder source hash beside ingested_at). H-290 (served-path residue, WAITING not before 2026-10-20) is filed correctly under CP-PICK -- it is a TIME block, not a WORK block, and should stay WAITING.