CARD K1-16  (pre-registered 2026-09-26 02:36:05 -0700)
Base: k14 (11 hunks; real residual: 0xfc "add esp,0xc" scheduled after the subject_race store + actor-index reload
  in ours, before them in January; 0x1d49 magnitude_squared2d operand order).
Evidence (D, /Od 0x48d724-0x48d787): subject block writes subject_comm_groups_enabled[0] then [1] in both arms
  ([0]=1,[1]=1 / [0]=1,[1]=0), the same order as the cause block. Ours writes [1] before [0] in the subject block
  (an earlier lane's spelling chosen to reproduce January's tail-merged [0]=1 store).
Change: subject block in the /Od order ([0] first), like the cause block.
Prediction: EITHER the 0x144-0x166 region stays January-identical (scheduler sinks [0]) and the IL-order change
  moves the 0xfc tie, OR the region breaks (then [1]-first is January's order and this is recorded as evidence
  that /Od differs here). Weak prior: 0xfc unchanged.
