The change adds one edition candidate, its paired draft assessment, a spend record, a run record, a research-status change, and two harvested prior arbiter verdicts; AGENTS.md is untouched and the tooling found no new citation identifiers under content/, so there is no fabrication surface I can test. Provenance is stamped throughout (runId, producedBy, model, date, promptVersion, ledgerHash) and the assessment header states plainly 'Append-only; NOT human reviewed', with role: draft and humanReviewed: false — no AI record is dressed as human and no draft as ratified (§3.15, §4). The edition's stated purpose is to tighten discipline, not loosen it: evidence ids formerly spent on neighbouring claims (E032 on C002/C005, E024 on C005, E021 on C020) are now cited beside the claims they attach to, with cross-claim inference disclosed as inference (§3.2, §3.6, §3.8); every anchor-only featured claim (C039, C073, C077, C143, C174) is named as anchor-only and graded no higher than unresolved; and records resting on the unread SRC-JHONG-2026A (E044, E045, E050–E052) are explicitly treated as carrying no weight, consistent with the provisional/unread-source rule in §4. Competing accounts are presented symmetrically with their own decisive tests (§3.3), negative results (E024 null resistivity, E026 absence of pain-free controls, E016) are carried as spine (§3.11), credibility is separated from diagnosticity throughout (§3.4), and independence caveats (same NIH lineage, proponent authorship, second-hand review) are stated (§3.10). The research-status change to VASO-R007 is mirrored identically in research.yaml, history.yaml (append-only, actor and aiAssisted recorded), and the run's status file, so nothing is misattributed (§3.14).

The assessment and edition are explicitly stamped with model, date, prompt version, producing run, role:draft, and humanReviewed:false, so neither is presented as a human or already-ratified judgment. Claims based on SRC-JHONG-2026A are repeatedly identified as anchor-only, from an unread and unverified source, assigned no evidential weight, and graded unresolved; even the page anchors are accompanied by that verification caveat rather than represented as checked. The synthesis distinguishes credibility from diagnosticity, names contrary and null findings, discloses shared research lineages and essay-derived dependence, and presents three serious competing mechanisms with calibrated uncertainty. The research-status transition and history entry are append-only and identify the responsible model run, while AGENTS.md and the constitutional checks are untouched; mechanical verification also found no newly added citation identifiers requiring resolution.

The change adds a new edition candidate and accompanying draft assessment for the vasocomputation case under protocol edition-v10, along with run, spend, and harvested arbiter records. The assessment is properly stamped as an unratified AI draft (`humanReviewed: false`, `role: draft`) and adheres strictly to the rule that evidence records are cited only for claims they are attached to or as explicit disclosed inferences. Anchor-only claims (C039, C073, C077, C143, C174) are transparently labeled as held on anchors without admitted evidence, records from unread source SRC-JHONG-2026A are treated as absent, and VASO-R007 is legitimately recorded as answered following the execution of study VASO-S001. No files are omitted, AGENTS.md is untouched, and all epistemic provenance requirements are satisfied.

The diff adds a stamped draft assessment (role: draft, humanReviewed: false, runId/model/date/promptVersion/ledgerHash) and matching edition candidate under edition-v10, plus an append-only history line and research.yaml status flip for VASO-R007. The rationale and synthesis explicitly correct prior cross-claim spending of evidence IDs, mark every anchor-only featured claim (C039/C073/C077/C143/C174) as held solely on its unread anchor with no admitted record and grade none above unresolved, treat SRC-JHONG-2026A-derived records as absent, and keep the case verdict unresolved with competing accounts and calibrated language. No new citation identifiers appear (tooling empty), AGENTS.md is untouched, checks are tightened rather than loosened, and provenance/history are honest and append-only.

The diff adds only edition/assessment candidates, run and spend records, an append-only history entry, a research-lifecycle status change, and harvested past arbiter verdicts; AGENTS.md is untouched and the mechanical verification found no newly added citation identifiers, so no new fabrication surface is presented. Provenance is stamped throughout (runId, model, date, promptVersion, humanReviewed: false, 'NOT human reviewed'), anchor-only claims (C039, C073, C073/C077, C143, C174) are consistently graded unresolved and said to be held on their anchors with no admitted evidence, and the unread-source records (SRC-JHONG-2026A: E044, E045, E050–E052) are explicitly treated as absent under the provisional rule rather than being spent. The edition's whole purpose is to tighten compliance with edition-v10 — citing each evidence record only beside the claim it is attached to, with cross-claim uses stated as disclosed inferences — which strengthens, not weakens, the §3.15 checks; the VASO-R007 answered status is a reasoned lifecycle update tied to already-admitted study S001 and evidence E019–E021, not a new record. Disagreement, negative evidence, and strongest objections are displayed rather than hidden, and the harvested pr-357/pr-359 files are verbatim governance records, not constitutional edits.