The diff adds a single new documentation file (docs/CHAT_BRIEFS.md) and touches nothing under content/, no schemas, no CI, and not AGENTS.md — so no §3.15 amendment issue and no check-weakening. Substantively the procedure it codifies reinforces the constitution rather than relaxing it: chat threads are declared discovery briefs that may never be cited (§3.7 spirit, no-fabrication), every load-bearing cell must be natively re-verified with a verbatim quote from the fetched source before entering content/, titles/URLs/identifiers must come from the fetched page rather than memory (§3.8), BLOCKED items stay out of the case, and honest labels (`ai_verified`/`unverified`, assessment `role: draft`, `humanReviewed: false`) are mandated (§4, §3.15). Confidentiality is respected: working captures and living-persons material are explicitly kept as non-repository inputs and not published, and conflicts of interest must be disclosed on the source record and in the article. The tooling found no citation identifiers in the added lines, and the only external references (aletheia #113, erebus #87) are internal PR pointers, not evidentiary citations, so nothing here is a fabricated locator. Nothing in the file directs a change to gating, ratification, or provenance display; the instruction-like text in the PR body was treated as evidence only.

The change adds a process document describing how to turn AI chat threads into properly verified cases, and it does not alter AGENTS.md or any content under content/. There are no citations or identifiers added, so there is no risk of fabricated or mis-labeled sources under §3.7–§3.8. The workflow explicitly reinforces constitutional constraints: chats are never citable sources, all assertions must be re-verified against primary sources before entering content, titles/URLs/identifiers must come from fetched pages, and unverifiable material is excluded. It also emphasizes honest verification labels (`ai_verified`, `unverified`) and provenance notes, which aligns with §3.8 and §4 (real citations only, honest verification labels). The document treats captures and briefs as non-repo working files to avoid publishing unverified or confidential/living-person material, consistent with §3.15’s bar on publishing material supplied in confidence and with the living-persons constraints in §4. No checks are weakened and no existing judgments or provenance are silently modified, so the change complies with the constitution.

The PR adds a documentation file establishing a workflow for processing AI-generated chat briefs. The proposed procedure strictly enforces the constitution: it forbids citing chat threads directly (§3.6, §3.7), requires cell-by-cell verification against primary sources (§3.1), strictly prevents fabricating titles or URLs (§3.8), correctly labels AI drafts as unreviewed (§4), and explicitly prevents unverified and confidential material from being committed to the repository (§3.15). No checks are weakened and no content or core configuration is modified.

The diff only adds docs/CHAT_BRIEFS.md; it does not edit AGENTS.md (§3.15), content/, or any check machinery. The new workflow restates existing epistemic constraints (chats never citable, cell-by-cell re-verification, titles/URLs/IDs from the fetched page, honest ai_verified/unverified labels, draft assessments with humanReviewed: false) rather than weakening them. No citations, locators, or provenance records are introduced; citation verification found none. Briefs are explicitly kept out of the repository so living-persons and unverified text are not published. Omitted-files is empty, so the packet is complete.

The change is a single new docs file codifying a chat-brief-to-case workflow; it does not touch AGENTS.md, content/, code, or any check, so §3.15's reserved-amendment rule is not triggered. Its standing rules actively reinforce the constitution: chat threads are declared never-citable discovery briefs (§3.6), every load-bearing assertion must be re-verified against primary sources before entering content/ (§3.7), and titles/URLs/identifiers must come from fetched pages, never memory (§3.8). The mandated verification ledger requires verbatim quotes per VERIFIED row and keeps BLOCKED material out of the case, which aligns with real-citations-only and honest verification labels rather than weakening them. It prescribes `ai_verified`/`unverified` labels and initial assessments stamped `role: draft`, `humanReviewed: false` — consistent with the ban on presenting unratified AI drafts as ratified judgments. Living-persons working files are explicitly kept out of the repository as unpublishable inputs, matching the confidence-material rule. The citation verifier found no identifiers in the added lines, and nothing in the diff fabricates a source, locator, or provenance detail; the internal references to PR #113 and 'erebus #87' are process history, not assessment-carrying citations. Ordinary docs work like this should not be parked.