The diff is tooling-only: `runAccount` in src/lib/arbiter-core.mjs now digests each added assessment (header fields, case verdict, load-bearing set, weakest links, whatIsClaimed, synthesis, components, and every claim verdict with confidence and clipped reasoning), the account header is updated to state exactly the new clip budgets, and a test fixture exercises the new output. AGENTS.md is untouched; no content/ files, citations, or locators are added, and the citation tooling found no identifiers to check. This strengthens rather than weakens the §3.15 checks: seats gain more of the record to judge, clips remain marked in place, and the header keeps 'No model wrote this section' so the account's provenance is honest (§3.8, §3.14). The DECISIONS.md entry accurately describes the mechanism and is stamped with date and channel, consistent with the founder-direction recording rule. No fabrication, confidence exposure, or misattribution is visible.

The change strengthens panel inspectability by mechanically including assessment provenance fields, case-level judgment, components, and every claim verdict in the run account. Clipping limits are explicitly disclosed in the generated header and omitted-character markers remain required, so the account does not falsely represent clipped reasoning as complete. The added test exercises the new provenance and judgment fields, while no constitution file, content citation, confidence material, or ratification gate is altered.

The change improves transparency for the constitutional panel by expanding `runAccount` in `arbiter-core.mjs` to digest full assessment overlays (headers, case verdicts, load-bearing sets, components, and individual claim verdicts). This prevents diff-budget truncation from obscuring whether an objection was properly resolved or a claim was regraded, directly supporting robust multi-model review under §3.15 without weakening any checks. The founder's session direction is properly documented in `docs/DECISIONS.md`, test coverage is updated, and `AGENTS.md` is untouched.

The diff expands runAccount so the panel receives a mechanical digest of every added assessment (header, case verdict, load-bearing set, components, every claim verdict/confidence/reasoning head) instead of only the case verdict and 700 characters of synthesis. That strengthens, rather than weakens, the §3.15 multi-seat review path that failed on #372 when the diff budget dropped the overlays. No AGENTS.md edit, no citations, no confidence material, no fabricated provenance; the DECISIONS.md entry is a dated founder-direction log of the same change. Tests and clip/header wording match the new extraction.

This change expands what the mechanically generated run account carries — each assessment's full judgment (header fields, case verdict, load-bearing set, weakest links, whatIsClaimed, synthesis, components, and every claim's verdict with confidence and reasoning head) instead of only the case verdict and 700 characters of reasoning. That is the opposite of weakening the checks in §3.15: it directly remedies the failure mode the PR describes (seats voting unsure because the diff budget dropped assessment overlays and the account carried too little to confirm a regrade), and clips remain loudly marked with the header updated to describe exactly what is and is not included. No AGENTS.md edit is present, so the §3.15 amendment-reservation rule is not triggered; the docs/DECISIONS.md entry is a decisions-log record of founder direction, stamped with date and channel, which the constitution expressly permits. The added code is a mechanical extraction with no editorial judgment, the test fixture is consistent with the implementation, and no citations, locators, or provenance details are fabricated anywhere in the diff. Nothing in the visible change misattributes who did what or presents unratified material as ratified.