Fisher–Muller, properly powered: more seeds and longer dishes for the sex study's two textures

[YesterdaysLemon]

The sex-vs-asex study (2026-08-23, archived in the research log) ended with two textures at the edge of significance, and the verdict's price is more seeds and longer dishes — no new machinery:

1. Canalisation: across-seed endpoint spread of proprio_gain far tighter under sex (sd 1.4 vs 6.6; variance ratio ~22, at the 5% edge of an F-test at n=3)
2. Hitchhiking: the study's only stall-and-reversal was asexual (seed 3: a lower-proprio lineage won its sweep on other merits and dragged the gene backwards) — the Fisher–Muller argument, once

• Power it: n≈8 seeds per arm, dishes ≥ 1800 s, equal mutation supply as before (ARENA_RECOMB 0 vs 1)
• Pre-register the two claims and their tests (F-test on endpoint variance; count of stall-reversals per arm) so the verdict isn't post-hoc
• Budget: ~16 dishes × 30 min-scale runs — background batches over a day
• Whatever the answer, it's museum Wing IV material