Underneath behavioral-mechanics-hub.md, sitting at equal status with every properly-cited page in the hub, are 17 pages that don't cite anything. No line numbers. No <sup>n</sup> footnotes. Just a line at the bottom of each one reading "Provenance: Synthesized from..." and a status value (vivid, vibe: high-density-tactical) that doesn't exist anywhere in the vault's own frontmatter standard. They use terms like "Authority Recognition Mechanism (ARM)" and "the Flower and Root" — vivid, plausible-sounding, and, after ten independent full readers went through all 26,217 lines of the actual book, absent from the source entirely.1
[RESONANCE] This wasn't a hostile actor slipping something in. Something inside the vault's own prior process generated a stylized, uncited rewrite of the book and it got filed as if it were the book — and it sat there long enough, formatted confidently enough, that the next ingest almost trusted it by default. The read-until-saturation pass only caught it by accident, because the 10 section readers were working from the actual source text and simply never encountered those terms — if they'd started from the existing vault pages instead of the primary source, Corpus B would have looked exactly as credible as everything else.
The vault has a name for citation discipline — inline <sup>n</sup> footnotes, full-path backlinks to source stubs, no claim without a traceable line. Corpus B violates every part of that spec at once (no footnotes, bare <span class="void-link-broken" title="Not in vault: wikilinks">wikilinks</span>, invalid status values) and still got folded into a hub as equal-status material. The formatting rules aren't decorative. They're the thing that would have caught this months ago, if anyone had lint-checked the hub against them.
Promotion criterion: pair this with a genuine LINT/audit session across other large hubs — is Corpus B a one-time incident from a single lost-and-rebuilt-skills-folder period (per CLAUDE.md's own May 2026 rebuild note), or is there a second hub somewhere carrying a similarly uncited, differently-formatted layer that's never been checked against its own citation-rule? If a second instance turns up, this stops being an anecdote about one book and becomes a real vault-hygiene finding worth its own audit protocol.
Flag for the Corpus B remediation session (rebuild-to-spec or demote out of ARCHIVES, per PRD) — and when that session runs, check whether the same lost-skills-folder period that produced Corpus B produced anything else format-noncompliant elsewhere in the vault.