The diff adds a single new edition candidate for the vasocomputation case plus its run record and a spend line; no AGENTS.md or check-weakening edits, and nothing is removed. The edition explicitly re-adopts the incumbent's assessment (verdict 'unresolved') and says so in its rationale, so no published judgment is silently changed and nothing unratified is dressed up as ratified — the header carries runId, model, date and promptVersion, and the file comment states it replaces the incumbent only if the panel prefers it. Epistemically the article is unusually careful: it labels the proponent's essays as AI-generated and unverified (SRC-JHONG-2026A), marks the perforator account as having 'nothing, in either direction' behind it, records negative and unfavorable findings (Simons' non-exclusivity, Ballyns' null resistivity index and lack of pain-free controls, the arteriole audit pointing against the latch), separates source statement from inference, and keeps the 'what would change our mind' lists symmetric (§3.5, §3.9, §3.11, §3.13). Claim spans are atomic and the three competing accounts are displayed side by side (§3.2, §3.3). Papers located-but-unretrieved are explicitly said to carry no weight until read, and no locator is invented; the tooling found no citation identifiers added, and I see no internal provenance contradiction. Ordinary imperfections only.

The edition preserves three serious competing accounts, presents evidence and predictions symmetrically, and retains an explicitly calibrated “unresolved” verdict. It clearly distinguishes retrieved findings from inference, noting that the perforator account rests on an unverified AI-generated essay, assigning it zero independent weight, and stating that located-but-unopened studies carry no weight. Dependence and unfavorable evidence are visible, including shared research lineages, derivative records, normal-site findings, and arteriole results that lean against the featured mechanism. The new edition is append-only, points to its predecessor and adopted assessment hash, and is stamped with model, date, prompt version, and ledger/input hashes. It is explicitly labeled a candidate that replaces the incumbent only after panel preference, so it does not present an unratified judgment as ratified; the citation tool also found no newly added identifiers requiring an existence check.

The pull request adds a candidate edition for the vasocomputation case that maintains the incumbent assessment ('unresolved') while explicitly organizing the narrative around three competing accounts in compliance with §3.3. All claims, cruxes, and plates are properly keyed to existing ledger identifiers without adding fabricated citations or unverified assertions. The provenance, run account, and governance spend records are fully intact, consistent, and traceable under §3.14 and §3.15.

The diff adds only a new edition YAML that explicitly re-adopts the incumbent assessment (same unresolved verdict, load-bearing set, and hash pointer) plus a spend log and run records; no assessments are silently rewritten. The article and accounts field surface three competing mechanisms side-by-side with symmetric change-our-mind lists, honest labels for second-hand/unverified/AI-generated material, and calibrated uncertainty, matching §3.3–§3.5 and §3.13. No new structured citation identifiers appear (tooling reports none), provenance of ledger items is preserved rather than invented, and AGENTS.md / checks are untouched. Omitted-files list is empty and the run account shows a completed re-adoption with no panel dissent to suppress.

This is an edition-verb run that re-adopts the incumbent's ratified assessment (unresolved; same load-bearing and weakest sets) while making only presentational changes — naming the three competing accounts, holding the featured set, keeping crux order — and it does so transparently via the previous pointer and a stamped assessment hash, satisfying §3.14 and §3.15's append-only, no-silent-change requirements. The diff adds no new citation identifiers (citation verification found none), introduces no new Source or Evidence records, and the article's references map to existing ledger claim IDs; no fabrication vector is visible. The article is notably disciplined on epistemics: second-hand characterizations are labeled as such, unverified and unretrieved sources (SRC-JHONG-2026A, the 2025 runners' elastography study, the fibromyalgia surveys) are explicitly given zero weight, the AI-generated essay's testimony is labeled as first-person testimony rather than measurement, and the three competing hypotheses plus symmetric 'what would change our mind' lists satisfy §3.3, §3.8, and §3.13. The draft is presented as a candidate for panel judgment ('Replaces ... only if the panel prefers it'), not as ratified, so the gated publication process of §3.15 is being followed rather than dodged. The minor token-accounting difference between run.yaml (inputTokens: 256355) and spend.yaml (input: 4 + cacheWrite: 256351) sums identically and is not a record fabrication. I see no weakened checks, no provenance removal, and no confidence material exposed; the diff is complete, so no omitted files hide a violation.