The diff does not edit AGENTS.md, config or any check, adds no citation, identifier or locator (verification found none), and touches no claim, evidence or source record. It adds an edition, append-only draft and check assessments (stamped runId/model/date/promptVersion, humanReviewed: false), two research-status updates, a stamped history line, spend ledgers and run records. The article moves toward ledger fidelity: it withdraws the incumbent's unsupported sentence that a radical-pair model reproduces the xenon potencies, tells ORCH-C046 and C057 as reports, and marks the corpus survey and xenon re-analysis as the site's own work. The one verdict change (ORCH-C046 to unresolved) is reasoned openly, with no attached record and the prior grade disclosed. All four seats disagree on C046 and C012, and that disagreement is displayed through the installed seat readings, not hidden. No file asserts 'ratified'; standing is left to derive from the checks, and the reported split list (C012, C046) is consistent with a within-one-step majority on the load-bearing C010, C030, C002 and C013. The Anthropic seat's max_tokens failure is disclosed in run.yaml and the spend ledger (128,000 output tokens, $2.96). R008 and R010 are marked answered with statusBy, statusDate and a history entry anchored to S001/S002 and E035–E040, and the residual work is pointed at R014/R011. The omitted drafter replies are fully accounted for by the run account. Minor notes, not breaches: the Grok seat's `sensitivity` block was dropped at install (the raw reply survives under proposals/, but the verb should carry or note such fields); the draft's C012 'mixed' reasoning leans on diagnosticity rather than the claim's truth condition; and the PR body's '5 exact' does not match the seven unanimous-exact claims.

The edition and adopted assessment are explicitly stamped as AI-generated drafts with runId, model, date, prompt version, and humanReviewed:false, while the four independent check assessments preserve their own verdicts and the failed Anthropic call is recorded rather than concealed. Three of the four returned seats are within one step of the draft’s unresolved verdict, and the account reports no split on any draft load-bearing claim; the contradicted seat and the two non-load-bearing claim splits remain visible. The article carefully separates measurements from interpretation, identifies same-program or partial independence, presents contrary animal results and underground nulls, and uses calibrated labels rather than a truth score. It also withdraws a proposition carried only by refused records and frames ORCH-C046 and ORCH-C057 as reports or assertions rather than verified findings. The two research-status changes are accompanied by status provenance and an append-only history entry naming the responsible run and rationale. The only omitted files are drafter replies that the RUN ACCOUNT compared against the installed edition and assessment, so they do not leave a material unseen gap.

The pull request cleanly updates the Orch OR edition to bring the article within length guidelines while preserving the unresolved case verdict and faithfully representing the underlying evidence ledger. The multi-model check panel operated under check-v2 protocol and achieved concurrence (three of four installed seats within one step of the draft's unresolved verdict: Gemini weakly_supported, GLM unresolved, Grok unresolved, and GPT-5.6 contradicted), with Anthropic's empty max-token response transparently logged in spend and run manifests. Research agenda updates (ORCH-R008 and ORCH-R010 marked answered) are properly justified by existing studies in the ledger and recorded in history.yaml. No constitutional rules were modified or weakened, and provenance records across all spend and run files are complete and honest.

The change is an append-only edition rewrite plus a blind check panel on the existing ledger: no claim, evidence, or source records are added or altered, and citation verification found no new identifiers. The draft assessment stays labelled role: draft / humanReviewed: false, stamps runId/model/date/promptVersion, and the article now says only what the ledger holds (withdrawing the refused radical-pair potency sentence; reporting C046/C057 as reports). ORCH-C046 moves to unresolved because no admitted evidence is attached to that claim while the same bound remains weighed on C030 — honest provenance, not fabrication. Research status flips for R008/R010 are stamped to the edition run with notes tying them to S001/S002, and history is append-only. Four of five check seats installed (Anthropic empty at max_tokens, recorded); three of four are within one step of unresolved, disagreement is displayed rather than hidden, and AGENTS.md is untouched.

No AGENTS.md edit, no new citations (verification found none added), and no claim/evidence/source record is touched, so the append-only and real-citation rules are not implicated. The ratification path is honest and follows the concurrence mechanism: four independent vendors' seats answered blind (the Anthropic seat's failure is recorded in the run notes and spend ledger with the exact token ceiling, billed, and the seat is owed a re-ask rather than silently dropped), three of four sit within one step of the draft's unresolved verdict, and the two split claims (ORCH-C012, ORCH-C046) are not in the draft's load-bearing set, so the standing 'ratified by three of four' matches §3.15's scale, with every seat's word displayed. Provenance is properly stamped throughout (runId, model, promptVersion, date, humanReviewed: false, role: check/draft), the withdrawn radical-pair sentence was dropped because its records were refused at intake — the article now says only what the ledger holds — and the ORCH-C046 move to unresolved is explicitly labeled as following the no-attached-evidence rule rather than doubt about the paper, with the bound still weighed at ORCH-C030. The one blemish — the xai seat's raw reply carrying a sensitivity field the installed assessment omits — is itself printed by the repository's own comparison tooling and the raw file is committed, so nothing is hidden or misattributed; the two research-status changes are stamped with statusBy/statusDate and a history entry. I see no fabricated citation, no confidence exposure, no weakened check, and no misattributed provenance.