CARD K1-06  (pre-registered 2026-09-26 02:18:46 -0700)
Base: k04 (held lab) - the change itself is an ordinary admissible /Od-attested statement-shape correction.
Observed: January's animation_weight lives at [ebp-0x7c] (shared with the timer-loop counter tv1146, deep); ours
  (k04) at [ebp-0x60] (shared with tv4884, shallower) => ours ranks animation_weight higher => more IL refs.
Source evidence (D, /Od 0x490214-0x4902dd): the impulse test is a boolean local (byte [ebp-0x13ad]) =
  unit_test_animation_impulse(...); "if (impulse && protagonist_actor_index != NONE) { assert(protagonist_actor);
  if (actor_action_class(..) == 2) impulse = FALSE; else if (actor->state.mode == 1) impulse = FALSE; }
  if (impulse) animation_rating = 2.0f; else animation_type = NONE;" -> ONE store of 2.0, one dead NONE store.
  Ours writes animation_weight = 2.0f twice (merged by cross-jumping in the final code, but counted twice as IL
  references before cross-jumping - the apc tail proves pre-merge counting: 8 source stores, 1 emitted).
  J: January 0x173c-0x17a7 is consistent with the threaded boolean (one 2.0 store at 0x17a0).
Change: /Od form with a descriptively named boolean local "animation_impulse" (name not attested).
Prediction: 0x173c-0x17a7 stays January-identical; animation_weight moves deeper (toward January's [ebp-0x7c]);
  frame/hunk effect: at least one pairing change; hunk count <= 36.
Collateral: none outside the function.
