Review of 2026-09-15T1002Z-h099-aar-pass-attempt-metric

Owner lane: Not recorded

Reproduction

REFUSED -- source record binds no design_class, protocol, runner, input or result hashes (legacy, unadmitted; selfcheck 4 findings, 2 BLOCK: fit_window grammar 'season <= S-1', no evidence_contract). Not retrofitted. The headline week-1 numbers also do not reproduce from today's warehouse (below).

Finding and verdict

UPHELD

Review scope

UPHELD for the question asked: the -3.50 measures the metric (perfect forecaster reads -3.6 to -3.8 in weeks 1 and 2, exact sign-flip p<=1e-3). WEAKENED for two secondary statements: the recorded week-1 magnitudes (B-A +3.64, D +0.13, SEL 2.15) do not reproduce from today's warehouse (exactly 20.09 targets of played-flag drift), and 'served QB attempts unbiased' reads +0.38..+0.72 at HEAD. Source record stays UNADMITTED (legacy, no evidence contract); this is a REVIEW record, not a PROMOTION, and nothing here enters CHANGES-PROPOSED.md.

Next action

Any re-issue of the H-099 numbers should bind the warehouse (snapshot hash) and the played-flag definition; the 45 played=1 rows with NULL targets in 2026 week 1 are a definitional question for `played`, owned by the warehouse lane, not this one.