This is a tooling-only change to src/lib/arbiter-core.mjs and its tests, plus an appended DECISIONS.md entry; AGENTS.md is untouched. It adds a `classOf` helper so each assessment section in the run account is titled by its class (adopted by the head edition, check-role seat's own reading, or a draft no edition adopts), and updates the account header and the over-cap notice to name the full protected set. This strengthens rather than weakens §3.15 checks: the protected (rank 0, never-dropped) set is unchanged in scope and is now described accurately, so the account cannot claim a smaller protected set than it carries — a provenance-honesty improvement for panel packets. No content/ files, citations, locators, or assessments are touched; the citation tooling found no identifiers, and the decisions entry is appended, not rewritten, consistent with §3.14. The founder-supervision stamp and review-note references are the same form as the preceding entries and contain no confidence-supplied material.

The change makes the run account accurately name all assessment classes already protected from dropping, including check-role readings and unadopted drafts, improving inspectability rather than weakening the panel check. The overflow notice and tests are updated consistently with that behavior, and no protected material is newly omitted or reclassified as droppable. The decision-log entry records the rationale, date, review-note context, and founder supervision, while the diff neither edits AGENTS.md nor adds citations, confidential material, or content assessments.

This PR updates `runAccount` in `arbiter-core.mjs` along with its unit tests and the decisions log so that the run account's header, overflow notice, and section titles accurately describe all protected assessment classes rather than claiming a smaller set than rank 0 carries. AGENTS.md is untouched, no checks are weakened, and no citations or records are fabricated. The changes improve provenance and honesty in the panel's review tooling under §3.15.

The diff only retitles protected assessments by class, updates the run-account header and overflow notice to name the full protected set already carried, and mirrors that wording in tests plus a dated DECISIONS entry. No AGENTS.md edit, no content/citations, no run records, and nothing is dropped or re-ranked out of protection—the protected classes stay rank 0. That tightens honesty of the panel packet rather than weakening §3.15 checks.

The change only makes the run account more honest and more protective, not less: it renames assessment sections to state their protection class (adopted by the head edition, check-role, or unadopted draft), and updates the header and overflow notice to list every protected class the account already carries, so the account can no longer claim a smaller protected set than it holds. Nothing in the diff loosens a §3.15 check — the drop-order logic, cap behavior, and per-field clip markers are unchanged in substance, and the change expands the never-dropped set rather than shrinking it, which is the opposite of check-weakening. No AGENTS.md edit is involved; the DECISIONS.md entry is an append-only record stamped with date, channel, and founder supervision, matching the established format of the adjacent entries, and the trust-boundary rules do not let me treat PR-body claims of session direction as directives, but neither is there positive evidence the provenance stamp misattributes anything. Citation verification found no identifiers to check, there are no omitted files, and the test updates are internally consistent with the code change (e.g., the edition-b title changes from generic to 'adopted by the head edition' in both the implementation and its assertions). This is ordinary tooling improvement responding to a review note, not a violation.