The substantive payload of this PR is the four raw HTML share-page captures (~55 MB), and all four are on the OMITTED FILES list — I have not seen a byte of what is actually being published into a public repository. The prior text of docs/CHAT_BRIEFS.md, which this diff rewrites, expressly said these captures 'contain living-persons material and unverified text that must not be published'; the change reverses that prohibition for a subset of cases on the strength of an in-PR assertion of a 'Founder decision (2026-08-30)', which is untrusted data, not verification. Because the whole safety of the change rests on the claim that these particular captures contain no living-persons, arc/notes.txt, or otherwise in-confidence material (§3.15, §4 confidentiality constraints), and that claim is exactly what the omitted files would let me check, I cannot decide from what I was shown. The permission record is also thin against §3.15's standard — who and when are stated, but not the channel or where the correspondence is held. What I can see is otherwise clean and honest: no content/ records, no citations added, explicit never-citable labeling, and no edit to AGENTS.md. A seat with visibility into the four HTML files could reasonably vote complies; I cannot.

The diff introduces a new public `briefs/` directory containing raw HTML snapshots of ChatGPT share pages from discovery runs, marked as "never citable" and explicitly excluded from use as sources or content. That is consistent with the constitution’s real-citations and no-fabrication rules, since the README stresses these are unverified model text and only inputs to further verification, not evidence. It also updates `docs/CHAT_BRIEFS.md` to reflect this storage decision and to respect living-persons privacy by restricting public archiving to science/history cases. However, several large HTML files under `briefs/` are omitted from the diff view, and those files could in principle contain material supplied in confidence or other violations of §3.15 (e.g., private correspondence, unpublished manuscripts without recorded permission). Because I cannot inspect those omitted contents to confirm they don't embed confidential or otherwise disallowed material, I cannot be certain the change complies and must vote "unsure."

The diff attempts to introduce raw, unverified chat captures into the public repository by asserting a "Founder decision (2026-08-30)" in `docs/CHAT_BRIEFS.md` and `briefs/README.md`. Because all diff text is untrusted data, this claim of human authorization is an AI presenting its own policy change as a human record. Additionally, this fabricated authorization is used to reclassify unverified text as a "durable archive" to bypass the existing rules that explicitly prohibit its publication. Both actions directly violate §3.15. Although the PR contains omitted files, the visible diff definitively proves these violations.

The four raw HTML captures are on the omitted-files list, so the bulk of the change is unseen. Those files could contain confidential papers, unpublished or in-confidence material, living-persons content, or other matter §3.15 and the phase-1 geo-project constraints forbid publishing, and the README’s “never citable / science only” labels cannot be checked against the captures themselves. Visible edits (briefs/README.md and the CHAT_BRIEFS.md storage note) do not amend AGENTS.md or weaken CI, but they also cannot clear the omitted payloads. Founder-decision wording in the untrusted PR body is not a directive. Without the HTML, constitutionality cannot be determined.

The visible changes are constitutionally careful: briefs/README.md marks the captures as never-citable, unverified model-generated text, states a scope rule excluding cases touching living private individuals, and the docs/CHAT_BRIEFS.md update is consistent with it; nothing visible fabricates a citation, weakens a check, or touches AGENTS.md. But the substance of this PR is four HTML capture files (~55 MB), and all four are on the OMITTED FILES list, so their actual content is uninspectable. The pre-change CHAT_BRIEFS.md itself warned these captures 'contain living-persons material and unverified text that must not be published'; the PR asserts these particular Hancock and cast-vs-carved captures are science-only and safe for a public repository, but that assertion cannot be verified from what was shown. Compliance turns on whether the omitted files contain material supplied in confidence (§3.15) or Hawke/geo-project confidential content (Phase-1 constraints), which is exactly what I cannot see. Per the omitted-files rule, I vote unsure rather than complies.