CARD K1-02  (pre-registered 2026-09-26 02:01:59 -0700; rewritten before any compile because backticks were shell-expanded in the first write)
Base: k01 (8064 [sha], frame 0x126c, 74 hunks).
Change (LAB, HELD class: indeterminate value copied on the reply path): remove the function-body
  "short play_type = _unit_play_speech_none;" initialiser and declare "short play_type;" in the speech block after
  animation_weight (/Od: word [ebp-0x119c] declared after speech_rating 0x1190 / animation rating 0x1198; its only
  store is the consider_speech result at 0x490172).
Evidence: J: January's else-body prologue has 6 zero stores (protagonist_actor, look_type, recipient_look_priority,
  protagonist_groups, is_reply, candidate_delay) and NO play_type store; its reply path 0x17a9 "mov edi,[ebp-0xf4]"
  reloads the previous iteration's play_type for the possibility store (0x18a9 mov [..-0x560],di).
  D: /Od stores play_type only from the call.
Prediction (replication of structural v10u on an older base): the 0xcd9 extra zero store and most of the
  0x1680-0x1830 consider_speech/possibility hunks disappear (hunks 74 -> ~40), but frame grows to 0x1274 (+8) and
  padded size 8096. Other rows unchanged.
Decisive test: frame line, hunk count, gate size.
