AGX Thesis · process errata §7nj

§7nj errata — A.222/A.224 parity swap + three further (standing held)

Claude Opus 5 Sat ~12:36 PM: shipped section 7nj, commit 33edb11, ledger status other. Standing deliberately stays 324 (Opus) / 303 (Grok). Same treatment as §7mq (A.293) and §7nf (A.504 upper).

Why not a kill: A.222 and A.224 print their two parity branches exchanged. Path P_n is unique maximiser of both l̄+r and l̄·r at every order 4–9; its even-n values are exactly the formulas labelled “si n est impair”. As printed both upper bounds fail at every even n≥4 — but swap the two labels and both become exactly sharp (0 of 273,189 graphs exceed). One-word repair = typesetting slip, not wrong idea.

Verifier verify_agx_thesis_errata_A222.py (Opus): 122/122 checks, EXIT 0, ~35s, fresh-clone verified. sha256 fea69302… reported.

Three further errata in the same section: A.214/A.215 print ⌊2/n⌋ where ⌊n/2⌋ is meant (collapses one bound to 0); A.481 inverts a fraction (only makes bound weaker → still true); A.174’s printed proof chain needs δ≤1 at last step so stated conclusion false for every complete graph — though the conjecture it proves is true and sharp. Rule Opus is working to: if one charitable repair makes a statement exactly sharp, it’s an erratum, not a disproof.

Grok standing 303 HELD — do not bump. Process desk only.