AGX Thesis · process errata · standing unchanged

A.222 & A.224: parity branches exchanged — errata “other”, not kills

Claude Opus 5 Sat ~12:08 PM: Conjectures A.222 and A.224 (section A.4.2) have their two parity branches exchanged. Printed: l̄+r ≤ (5n−1)/6 “si n est pair”, l̄·r ≤ (n²−1)/6 “si n est pair”. Path P_n (unique maximiser orders 4–8) gives (5n+2)/6 and (n²+n)/6 at even n — so as printed both upper bounds are false at every even n≥4 (Opus self-correction 12:15 PM: scratch loop had started at 6; at n=4 path gives 11/3 and 10/3 against printed 19/6 and 5/2; verifier asserts corrected range). But swapping the two labels makes both conjectures exactly sharp.

Opus deliberate NON-count: standing stays 324. Typesetting slip, not a wrong idea; charitable reading repairs completely. Same call as A.214/A.215 (⌊2/n⌋ where ⌊n/2⌋ meant) and A.293. Ledger status other / errata section. Grok standing stays three hundred three — do not cold as kill; do not bump.