The change is the second answer step in a chain of panel objections: VASO-C001 was found compound and refused, split into atomic VASO-C190 (occurrence) and VASO-C191 (commonness), with the refusal reason, splitting run, model, promptVersion and date stamped in claims.yaml, dispositions.yaml and history.yaml — §3.2 and §3.14 are served, not dodged. Records that cited the compound were relinked and then re-read against the parts: E028 was refused as non-atomic and reissued as E108/E109 with reader-act trails showing the dropped C190 link and the supports→context downgrade (§3.5, §3.9); E083 was refused because its quote did not state the proposition its title assigned it; E012/E015/E018 are marked provisional with 'source text not retrievable' reasons and the head assessment explicitly treats them as absent and carrying no weight (§4 provisional rule). The head edition adopts the objecting seat's point in full, moving C191 from weakly_supported to unresolved, deleting the unsupported 'the literature uniformly treats the phenomenon as common' justification, and rewriting the three article sentences so the prose says only what the admitted records hold (§3.13, §3.6). Assessments are new append-only files stamped runId/model/date/promptVersion with humanReviewed: false and role: draft, and C190's provisional grade discloses that it leans on sibling records (E087/E090/E081) rather than its own context records. No citation identifiers were added; AGENTS.md is untouched; no check in §3.15 is loosened. My one review note, not a violation: the mechanical re-parenting of child claims (e.g. C059, C084-C086, which are specifically about community-sample prevalence) to BOTH C190 and C191 is a blanket relink rather than a reading of which parent each bears on, and should be tidied in a later pass.

The visible run account shows the substantive correction behaving constitutionally: C001 is split into atomic C190/C191, contextual records remain context, provisional records carry no weight, and C191 is now unresolved rather than supported. However, the added packet.json and two reply.json files are omitted from both the diff and run account, so their published contents cannot be checked for confidential supplied text, fabricated provenance, or inconsistent model/run stamps. This matters because working packets can contain source text and prompts even when the resulting edition is sound. I would need those three omitted files, or a mechanical account establishing that they contain no confidence material and that their provenance matches the recorded runs, before finding compliance.

This change properly resolves the review notes regarding compound claim VASO-C001 by cleanly splitting it into atomic claims VASO-C190 (occurrence) and VASO-C191 (general-population prevalence) under §3.2. Sibling and dependent evidence records were re-examined and relinked without unearned inheritance of evidentiary bearing (§3.5), including splitting compound evidence VASO-E028 into atomic context records VASO-E108 and VASO-E109. Unretrieved source records remain strictly provisional and unweighted in assessment calculations in full accordance with §4. In response to panel feedback, VASO-C191 was calibrated to 'unresolved' at low confidence because only context-level records bear on it, satisfying §3.13. All modifications maintain append-only histories, dispositions, and complete run provenance (§3.8, §3.14).

The change answers the prior §3.2/§3.13 and §3.5/§3.6/§4 objections: VASO-C001 is refused as non-atomic and split into C190 (occurrence) and C191 (commonness); linked evidence is re-read, E083/E028 refused or re-split into atomic context-only E108/E109, and provisional unreadable sources are labeled and excluded from weight. The head assessment grades C190 provisionally_supported on disclosed sibling imaging and C191 unresolved/low, stating every admitted record is context and no population count exists, matching the article and dropping the prior weakly_supported/essay-weight and 'uniformly common' phrasing. Assessments are append-only drafts with run stamps; no AGENTS.md edit, no fabricated citations (none added), and no confidence material is published.

This change is the constitution's machinery working as designed: a panel seat's §3.2 objection led to VASO-C001 being refused as compound and split into atomic C190/C191, with records relinked and re-read against each part rather than inheriting the old link (§3.2, §3.5). The second readers refused records that failed checks — including their own splitting artifacts, e.g. the E028 part whose locator pointed at the Keywords list when the quote sits in the Introduction, which was caught and refused rather than published, and the mislabeled E083 part — exactly the honest-refusal behavior §3.8 and §4 demand; no invented citation, quote, or locator is visible anywhere in the additions, and the records the head assessment leans on (Croft 1994/PMC2540791, Ballyns 2011/PMC3493620) were retrieved and read during the runs. The flawed intermediate grade on C191 (weakly supported while its own reasoning admitted no admitted record and a provisional record was spent on it) is corrected in the head edition, which grades it unresolved at low confidence, strips the 'uniformly' claim, and states in the article that the essay's report carries no weight; the superseded assessment is kept append-only and labeled draft/humanReviewed:false, so no unratified or superseded judgment is presented as standing (§3.13, §3.14). Provisional records (E012, E015, E018) are stamped, displayed, and explicitly given no weight, satisfying the founder amendment on provisional records. One item I cannot verify from the packet and would want the operator to answer on the record: the PR body asserts the cap-blocked edition ran under a dated founder exemption (#411), which the shown run records do not document — a budget-cap exemption is not one of the §3.15 epistemic checks and the spend ledgers are otherwise transparent, so this does not rise to violation, but the exemption's record should be pointed to in docs/DECISIONS.md or the change notes.