Card L-S (STRIP TEST of candidate L-1; three compiles) - written 2026-09-26 17:12:29 -0700 (clock-read) BEFORE compiling
Candidate L-1 = HEAD + A + B + C:
 A (RF-E RR-1) argb argument = modulate_pixel32_by_real_alpha(0xFFFFFFFF, alpha)   [/Od 0x66a020 +0x31d]
 B (structural R1 / RF-E RR-2) clip handled through the pointer (clipped = *clip; clip = &clipped; clip->... +=)
                                                                                   [/Od +0x220..+0x2a4, RTC "local_clip"]
 C (RF-L L-1) helper widget_instance_get_cumulative_alpha_modifier in the /Od 0x668190 shape (parameter reused,
   while loop, no `parent` local)                                                  [/Od 0x668190]
Each change has its own first-party attestation; the strip test shows what each buys by removing it.
 S-1 = HEAD + C       prediction: the loop fold flips to January's [ecx+eax] (VN ordinal of `definition + 0x4c`
                      depends only on the prologue), but the frame keeps HEAD's alpha/clip homes -> residual,
                      differing from January only in the RF-E frame instructions (A/B's territory). 70%.
 S-2 = HEAD + A + C   prediction: residual; clip moves to [ebp+8] but the clip phi stays a register (RF-E RR-1
                      finding) -> a few frame instructions differ, +0x78 matches January. 70%.
 S-3 = HEAD + B + C   prediction: residual; alpha takes [ebp+8] (RF-E: R1 alone) -> frame differences, +0x78
                      matches January. 70%.
(RR2 = HEAD + A + B, i.e. C stripped, is RF-E's measured one-instruction residual: C buys exactly +0x78.)

OUTCOME (appended 2026-09-26 17:13:22 -0700): ALL THREE PREDICTIONS HELD.
 S-1 (HEAD + C): gate 97/5/0; keyed diff vs build/base 1 changed (render_recursive, residual), 0 added/removed.
      +0x78 = `mov dx,[ecx+eax]` (January); the remaining hunks are HEAD's frame differences (alpha/clip homes).
 S-2 (HEAD + A + C): 97/5/0; 1 changed (residual); +0x78 January; frame hunks remain (clip phi in a register).
 S-3 (HEAD + B + C): 97/5/0; 1 changed (residual); +0x78 January; alpha in [ebp+8] instead of clip (4 hunks).
 RR2 (HEAD + A + B, C stripped; RF-E): +0x78 `[eax+ecx]` only.   L-1 (A + B + C): EXACT.
 => C alone buys exactly +0x78 (in every combination); A and B together buy the frame (RF-E). No change is
    decorative; each has its own /Od attestation. In every strip variant only render_recursive changes.
 Alignment hunk counts incl. the 4 literal-spelling hunks every EXACT fn shows: RR2 5, S-1 28, S-2 28, S-3 8, L-1 4.
