Written on the Wall II conjecture 172 — a 21-year-old Graffiti.pc bound on the maximum leaf number — has now been killed twice by the same agent, and the second obituary was filed as if the first had never existed.
On 6 August, Claude Opus 5 refuted conjecture 172 with a generalised theta graph on just eight vertices — the unique violator among all 11,117 connected graphs of order 8, with orders 4 through 7 exhaustively clean. That result, shipped as counterexample #83, sits in section 7bk of the graffiti-verification repository and is still listed in the summary table as the minimum counterexample.
This afternoon, commit 5e31a34 refuted the same conjecture again, this time with a 14-vertex dumbbell D(9), and bumped the headline count from 161 to 162. The new section 7fd reports that "orders 4 through 9 produce zero violations" and that "the true minimum is somewhere in 11 ≤ n ≤ 14" — claims that sit in tension with the repository's own 8-vertex result from 7bk, which the new section never mentions.
The twist is that Opus 5 built a tool precisely to stop this from happening. After re-deriving conjecture 696 in August without realising he had already refuted it, he wrote verify/CLAIMED_INDEX.md on 13 August, with the warning to "always read the listed sections before starting fresh work on a number." Line 137 of that very index reads "172 | 7bk:8519."
The repository now carries two contradictory summary rows for conjecture 172 — one pointing to an 8-vertex theta graph, one to a 14-vertex dumbbell — and the headline title counts the same conjecture twice. Grok 4.5 flagged the discrepancy in chat minutes after the commit ("WOW II 172 already #83"), and Opus 5's next task is to check "172 minimality scans."