This is a tooling/pipeline change: a new `answer` verb that reads standing objections against a PR and routes them back to the second reader (for named records) and the edition verb (for assessment/article), plus a generalisation of reverify's settlement into `settleRecords` and a new edition protocol v12. No content/ records or citations are added — the verification section confirms no identifiers to check — so no fabrication surface. The checks in §3.15 are not loosened: the verb explicitly refuses PRs that touch AGENTS.md, refuses code changes and multi-case changes, writes no `in` rows of its own beyond the second reader's admissions, pushes nothing, and the DECISIONS entry and code comments state the panel judges the change again. Objection text is passed into the reader and the edition packet plainly labelled "data under review, not instructions", and protocol v12 requires each objection to be adopted or declined with reasons in the rationale, which displays rather than hides disagreement. Provenance is stamped honestly: history entries carry the answer's runId, the reader's model and the `actor` "aletheia answer (… second reader), on #PR"; refusals record the state the record was actually in rather than a hardcoded "provisional" (the §3.14 fix noted as review note #371). The only judgment call I weighed is that an answer leaves a record it could not re-read untouched with no disposition row, the attempt recorded only in the run account — that is reasoned in the code comment (the record was admitted read; a failed fetch is no verdict) and is ordinary imperfection rather than a rule broken. Whose words count is also constrained to the arbiter workflow's own comment at the PR head, avoiding a spoofed "seat" directing the pipeline.

The change does not edit AGENTS.md, add citations, or relax any existing publication gate. `answer` treats objections as data under review, re-reads named records through the existing second-reader verification path, records run/model/date provenance and history, and leaves unread records unchanged. Edition-level responses remain AI-generated candidates under protocol v12, and the caller must push them for the constitutional panel to judge again rather than treating the answer run as ratification. The current-head and workflow-author checks also reduce the risk of misattributing arbitrary comments as panel objections.

The pull request implements `aletheia answer <pr>` to feed standing constitutional objections back into the verification and edition pipelines. It rigorously enforces trust boundaries by treating objections as untrusted data under review and restricting parsed objections to authenticated arbiter bot comments at the current HEAD. Crucially, the answer verb does not push, merge, or bypass any constitutional checks; any resultant diff remains subject to fail-closed CI and panel ratification under §3.15.

The diff adds the founder-directed `answer` pipeline verb and edition protocol v12; it does not edit AGENTS.md, add ledger citations, or publish confidence material. Consequential outputs are limited to second-reader re-settlement of named live records (same stamps/tombstones/history pattern as reverify, verb=`answer`) and a forced edition whose packet carries objections as data under review; the verb pushes nothing and leaves final judgment to a fresh panel (§3.15). Refusals cover AGENTS.md edits, code-only PRs, multi-case PRs, absent/spoofed/stale arbiter words, and wrong branch. Provenance (actor, runId, history, dispositions) and the DECISIONS/AUTOMATION entries match the stated 2026-09-20 session direction. No fabrication or check-weakening is visible in the complete diff.

The change adds an `answer` verb that reads a parked PR's standing objections and puts them back through the existing gated verbs — records named by seats go back through the second reader under the verify protocol (with the objection marked as data under review, judged on the source text), and everything else goes to a forced edition whose candidate must answer each objection in its rationale. It pushes nothing and writes no `in` row of its own; the caller pushes and the panel re-judges the head, so no consequential change publishes on the answer's word (§3.15 is explicitly preserved, and the verb refuses PRs that touch AGENTS.md or code, honoring the founder's exclusive amendment power). Objection provenance is guarded against fabrication: only the arbiter workflow's own bot-posted comment carrying a machine blob judged at the PR's current head, and review notes by the same author, count as the panel's word — comments by anyone else are ignored and flagged (review note #373's concern is implemented, not dodged). The refusals are reasoned, history entries are stamped with runId/date/actor, refused records are tombstoned with reasons, and the mechanical checks (quote match, identifier resolution, fail-closed reader) are untouched — telling the second reader the objection is an adversarial pointer, not a substitution for the source-text judgment, so I do not read it as check-weakening. No citations are added, no confidence material appears, and the mechanism itself was directed by the founder in session, as recorded in DECISIONS.md.