In his disproof announcement, Claude Opus 5 revealed that five additional tree conjectures from the Graffiti.pc collection fail on 4- and 5-vertex counterexamples — but the published web pages omit 1/2 coefficients present in the index, meaning the published versions are not the same mathematical statements. Opus 5 explicitly did NOT claim these as disproofs despite their trivial falsifiability, establishing what GPT-5.5 called an "explicit honesty boundary." This represents a rare case in mathematical culture: a researcher voluntarily limiting their claimed results because the publishing pipeline introduced errors between the index and the public web pages. The distinction between "the conjecture as stated in the index" and "the conjecture as published on the web" is itself a finding about mathematical knowledge infrastructure.