The workflow half of the change is unobjectionable: moving the trigger from per-push to an hourly cron, setting `cancel-in-progress: false`, and keeping `workflow_dispatch` does not touch the ratification gates, the fail-closed CI, or the contested-parking rule — batching only shifts when the same gated panel runs. The DECISIONS.md half is the problem. The header line `docs/DECISIONS.md | 2 +-` shows one line removed and one added: the diff deletes the existing entry `**2026-08-26 — Villarroel permission recorded; the history rewrite drops from urgent to optional.**` and re-emits its entire body glued onto the tail of the new batching entry, under the unrelated heading `**2026-08-26 — The content-response ripple batches hourly (founder direction).**`. The file's own first line states it is append-only, and the PR body describes this as an addition ("one file + decision log"), so the destruction of a prior entry's heading is undisclosed. The record that got mangled is precisely a §3.15 permission record (who granted, when, by what channel, where held); burying it inside a CI-scheduling entry degrades the discoverability and integrity of that provenance and silently rewrites a published decision rather than appending a new one (§3.14, §3.15). Secondary, and not by itself decisive: the new supersede loop closes *any* open `respond/*` PR, including one that may be sitting as needs-approval or parked-as-contested — the rationale (re-derivation from current main) is plausible but the step should exclude contested/parked PRs to stay clearly inside §3.15's "disagreement does not publish silently". Fix is small: restore the Villarroel entry as its own bullet and append the batching decision separately.

The change does not touch AGENTS.md and involves only a workflow file and docs/DECISIONS.md, so most core epistemic constraints are unaffected. However, the final bullet in docs/DECISIONS.md is malformed: it appends the entire Villarroel-permission/history-rewrite paragraph directly after the new decision about batching the content-response workflow, inside the same list item. That text is not introduced as a separate decision and appears to have been accidentally concatenated, which changes the meaning and clarity of the prior, already-ratified decision record. Given §3.15’s emphasis on transparent, inspectable records and not weakening or confusing the checks, silently garbling a prior decision entry into a new one compromises the integrity of the decision log. Because this log is the authoritative record for operational rules under which AI must act, corrupting its structure is a material violation. No citation or provenance fabrication appears, but the mis-edit of DECISIONS.md breaks the requirement that checks and their governing records not be weakened or obscured.

The PR alters the `content-response.yml` workflow to trigger on an hourly schedule rather than per-push, and adds a step to close older unmerged `respond/*` PRs as superseded. This does not weaken any editorial or validity checks (§3.15), as the newly generated PRs still re-derive all judgments from `main` and undergo the standard validation and gating before merging. No content or citations are modified, so epistemic rules are untouched. The update to `DECISIONS.md` clumsily concatenates the new entry into the preceding entry, overwriting the old entry's title, but the historical text is preserved; this is an ordinary formatting imperfection, not a constitutional violation.

The diff only retimes the existing content-response workflow (push/path filter → hourly cron plus workflow_dispatch), sets cancel-in-progress to false, and closes stale open respond/* PRs after a new batch PR is opened; it does not edit AGENTS.md, CI gates, ratification, verification labels, or overlay append-only rules. The same scripts still detect staleness mechanically, still open a labeled PR through the existing risk/auto-merge path, and still terminate the loop on overlay-only merges, so §3.15 checks are not weakened or reclassified. No citations, locators, assessments, or confidential material are added. The DECISIONS.md edit records the scheduling change on an existing 2026-08-26 bullet rather than a fresh append, which is sloppy log hygiene, but the prior Villarroel text remains and git history still versions the file—ordinary documentation imperfection, not a silent rewrite of a published epistemic judgment.

The change reschedules the content-response workflow from per-push to hourly cron, disables cancel-in-progress, and adds a step closing superseded respond/* PRs with a public comment. None of this touches the §3.15 gates: the produced PRs still flow through classify-pr-risk labeling, the same scripts, and the same required checks before merge (`gh pr merge --auto` still waits on checks), so no consequential judgment publishes on one model's unilateral say-so and no check is weakened or reclassified. Closing an unmerged automated draft PR does not alter published assessments (§3.14) — overlays on main are untouched, and the closure leaves a public comment and branch record rather than silent removal; the superseded content is re-derived from current main if still warranted. The PR body's claim of 'founder direction' is untrusted, but the change needs no special authorization: only AGENTS.md amendment is founder-reserved, and AGENTS.md is not touched. The DECISIONS.md edit is ugly — the new hourly-batching bullet absorbs the previous Villarroel-permission bullet verbatim into one merged entry rather than cleanly appending — but the §3.15 permission record's content is preserved in full, so the append-only log loses no provenance. No citations are added, and the verification packet confirms there is nothing to resolve. Ordinary sloppiness in the log formatting is not a constitutional violation.