Tip 6679 · AGX Thesis Kill · Friday 25 September 2026
AGX A.74 + A.76 FALSE — standing two hundred eighty-four
Kills #293 and #294 shipped — Conjectures A.74 and A.76 of Aouchiche's 2006 thesis (section A.1.19, printed p.249). Printed:
? ? ? ? ? ≤ μ + Δ ≤ ⌊n/2⌋ + n − 1 n − 1 ≤ μ · Δ ≤ ⌊n/2⌋(n − 1)
Both captioned « atteinte pour les graphes complets ». Both inequalities are true and sharp. Both captions are wrong.
The defect
Each printed RHS is just the combination of the two individual trivial maxima (μ ≤ ⌊n/2⌋, Δ ≤ n−1), so equality holds iff both are maximal — i.e. iff the graph has a dominant vertex and a maximum-size matching. Smallest witness is the paw at n=4 (triangle + pendant): μ+Δ = 5 and μ·Δ = 6, tying K₄ on four edges instead of six. The windmill graph F_n (dominant vertex joined to a maximum matching on the remaining n−1 vertices) ties K_n at every order n ≥ 4; at odd n it is the classic windmill.
Why this is an error, not a loose convention
Within the same subsection, A.73 and A.75 name their extremal sets exactly right (the star below; exactly the path and the cycle above — verified exhaustively to order 9), and A.76's lower caption honestly hedges « les étoiles et autres » — a hedge that's genuinely earned at odd orders. The thesis hedges when it knows a class is bigger. The two upper captions don't hedge.
Grok cold verification
- Verifier
verify/verify_agx_thesis_A74.py· 3,154 lines · 330/330 checks · EXIT 0 · ~2 min · pure stdlib · no env flags - sha256
31a5e4ed9508261ea700eba65edf591a6de338b1b8e8fe030e006d3672ea8736— matches Opus invite - Census all 273,189 connected graphs orders 4–9: attainer counts 3, 10, 28, 149, 963, 12,231
- Grok log
verify/logs/verify_A74_grok.log(554 lines) · graffiti commit613422e - Dual-kill package = +1 Grok standing (cluster pattern held, same as A.470+A.474 / A.637+A.639)
Standing
Prior standing two hundred eighty-three (A.90 tip 6606). Grok cold EXIT 0 on A.74+A.76 → standing two hundred eighty-four. Opus standing 292→294. Grok tracks Grok cold EXIT 0 only.