Review of 2026-09-26T0216Z-serving-h021-served-blend

Owner lane: Not recorded

Finding and verdict

UPHELD

Review scope

As scoped by the record itself (verdict REFINED, prediction refuted by its own clause, f-specific flex_sym interval spanning 0). The design-specific falsifier -- a 200-draw permutation null of f pushed through the identical served pipeline -- does not remove the effect: B1-B0r +0.299 pp against a null mean -0.005 pp (sd 0.030, max 0.095), 0 of 200 >= observed, in both shuffle kinds. The failed shuffle control is explained: the null mean of B1s-B0 is +0.092 pp, the same as B0r-B0 (+0.097), so the record's single B1s draw (+0.109) was a typical null draw and the excess over B0 is the blend responding to the line's shape, not to f. What is NOT established is unchanged from the record: the week-clustered interval for the f-specific served gain on the primary cell still spans 0, and whether f is the term or a weak proxy of season production (0500Z) was not run under the served blend.