This is infrastructure only: no content/ files, no citations (verification found none), and AGENTS.md is untouched. The change extracts the existing weekly-window arithmetic (epoch, canon paths, declared-supervised set, first-parent git log, lane split) into scripts/gate-window.mjs and imports it from both scripts/arbiter.mjs and the new schedule, so the two cannot count differently — the cap itself is unchanged (CONTENT_MERGES_PER_WEEK still bounds `room`, and `room <= 0` exits), so no §3.15 check is loosened. The scheduled job only re-runs the PR's own arbiter workflow run against today's window; the panel judgment is re-executed rather than reused or upgraded by hand, and parks for seat objections are untouched (parkedOnBudget matches only verdict=park with a spent-budget reason, and the tests assert violation-parks and passes are excluded). The kill switch in governance/operation.yaml is honored before anything is re-run. The two newly-carried declared-supervised hashes are moved verbatim from arbiter.mjs, not newly asserted, and each is stamped with PR, date and channel; the DECISIONS.md entry is likewise stamped with date and the founder's in-session direction, which §3.15 expressly permits recording. Remaining risk is only that a re-run could merge a change whose budget park was the sole brake, but the budget is a self-imposed rate limit that this code still enforces, so I see no rule broken.

The change does not edit AGENTS.md, publish content, add citations, or alter any assessment record. It preserves the existing budget gate by moving its epoch, canon paths, supervised declarations, and first-parent counting into a shared helper used by both the arbiter and scheduler. The scheduler only reruns the full arbiter workflow for PRs whose arbiter data says they were parked on the content-merge budget, and it exits unless the recorded operational state is `live`, so it does not bypass panel judgment or the kill switch. The founder-direction entry in DECISIONS.md records a date, session channel, and quotation in the form expressly permitted by §3.15, while the tests cover parsing and oldest-first slot selection.

This pull request automates the re-triggering of the constitutional arbiter workflow for pull requests that were parked solely due to reaching the weekly autonomous content-merge budget once window capacity opens up. It extracts the existing gate window logic from `scripts/arbiter.mjs` into `scripts/gate-window.mjs` so both the gate and the recheck schedule evaluate the budget identically, without loosening any verification or rate-limiting constraints. The recheck script explicitly respects the kill switch in `governance/operation.yaml` and re-runs the existing arbiter workflow rather than bypassing it. No constitutional rules in `AGENTS.md` are edited or weakened.

The diff is pure gate/ops plumbing: it extracts the existing weekly content-merge window arithmetic into scripts/gate-window.mjs (epoch, CANON paths, SUPERVISED_DECLARED, first-parent log, lane split) shared by arbiter and the new schedule, and adds a kill-switch-honoring cron that re-runs the last arbiter workflow only for open PRs whose sticky comment is a budget-only park, oldest first, one per free slot. SUPERVISED_DECLARED and the arbiter’s rate-limit behaviour are moved, not loosened; AGENTS.md is untouched; no content/, citations, confidence material, or run overlays are added. DECISIONS.md records the founder direction that authorized the fix. Tests cover the pure parkedOnBudget/pickForRejudge helpers. No §3 fabrication, provenance mis-stamp, confidence, or check-weakening issue is visible in the complete diff.

This is an infrastructure change, not content: it adds a scheduled re-judge of PRs parked solely on the weekly content-merge budget, re-running the same arbiter run on the same commit once the window has room. It does not weaken any §3.15 check — the budget itself still holds (re-judging happens only when `room > 0`, and the total re-judged PRs cannot exceed the free slots), the panel verdict and checks must still pass, the kill switch in governance/operation.yaml is honored in the new script, and the arbiter's own gate logic is unchanged (the window arithmetic was moved verbatim into `gate-window.mjs` and shared by both callers precisely so the two cannot count differently, which is check-strengthening, not weakening). No AGENTS.md edit, no citations added, no content changes, no fabricated provenance; the DECISIONS.md entry accurately describes the change and carries a recorded founder direction. The declared-supervised list is relocated verbatim with no entries added or removed. The `parkedOnBudget` matcher is narrow (park verdict plus a spent-budget reason string) and its edge cases are unit-tested, so it will not silently re-judge PRs parked for substantive panel objections. I see no rule broken; at most one could ask for a merge-time budget re-check against races between the 3-hour schedule and concurrent merges, but that is an operational hardening note, not a constitutional violation.