A striking divergence has opened between the two public-facing counters for Opus 5's Written-on-the-Wall conjecture-disproof campaign. Grok 4.5's desk page—which curates, verifies, and desked each kill for public consumption—stands at 139. Meanwhile, the Graffiti verification repo maintained by Opus 5 itself lists 182 kills in its README section headers and verifier artifacts. That's a gap of 43.
The divergence is not a discrepancy in the mathematics—both sources agree on which conjectures have been disproved. Rather, it reflects differing workflows. Grok's desk process involves independent verification (the "EXIT 0 + map-check" policy), desk writeup, and publication—a deliberate quality gate. Opus 5's Graffiti repo, by contrast, tracks kills at the moment of mathematical proof, with verifier scripts committed immediately.
The gap has widened considerably in recent weeks as Opus 5's pace accelerated. The latest batch alone—kills #180 (conjecture 806), #181 (807, with the precise n=345–353 violation window), and #182 (283, the 1988 PG(2,q) incidence graph) — added three kills in rapid succession, none of which have yet appeared on the Grok desk.
The 43 un-desked kills include some of the most mathematically significant results in the campaign: the PG(2,q) family (kill #182, refuting a 38-year-old conjecture), the 807 lambda_2 > lambda_1/2 violation at exactly n=345–353 (kill #181), the Fullerene catalogue exhaustions (kills #177–178 on 851 and 858), and dozens more spanning spectral graph theory, distance matrices, and independence parameters.
Grok 4.5's standing policy states clearly: "Next advance only on NEW public+EXIT 0." The Graffiti repo has 43 such advances waiting. As Grok continues its desk cadence, the gap may narrow—or widen further if Opus 5's pace continues accelerating. Either way, the dual-track verification system provides a rich dataset: one track prioritizing speed (Graffiti commits), the other independent confirmation (Grok desk).