mirror of
https://github.com/Cesarus85/ringfall-mjolnir-vr.git
synced 2026-10-29 04:01:10 +01:00
Jonas VolmanandClaude Opus 5
068038491f
My render_actor re-baseline note said the function's branch-width family is a "shortening fixpoint" in which three seed branches one byte over the line block five others. Wave H's research refutes that on every point. The "span 128" figures were near-form distances; the short displacements are 122-123, so all 27 near-but-fits branches fit on their own. No final-bytes model reproduces our object, while a relaxation reproduces all 526 of January's widths. Width here is non-local and chaotic: one deleted call flipped 29 branches, 18 with byte-identical spans, and 330 single-statement deletions moved the count from 0 to 32. Padding the frame to January's size left the count unchanged, so the claim that the frame "pays for the branch tax" is unsupported too. The consequence worth keeping: on this function padded size mostly measures this noise, so candidates should be judged by width-normalised code length. The same probe refutes the BASIS of an older closed-form impossibility argument for _actor_look_idle_find_prop in R21 - that the span alone decides the width. The site remains unreachable, but because the state is non-local, not because of that argument. Both records now carry the correction; the colour-argument fix stands, having been verified from January's bytes independently. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>