The diff touches only two files: a comment/constant in scripts/arbiter.mjs and an appended DECISIONS.md entry. No content/ records, citations, locators, assessments, or provenance are added, removed, or altered, and AGENTS.md is untouched, so §3.7/§3.8/§3.14 and the amendment reservation are not implicated. The substantive question is §3.15's ban on weakening the checks: the change advances GATE_EPOCH from 2026-08-25T16:22Z to 2026-08-27T10:00Z, zeroing the trailing autonomous merge count. This does not remove or loosen any constitutional check — fail-closed CI, multi-model ratification, and parking of contested changes are untouched, and CONTENT_MERGES_PER_WEEK stays at 10 — and it uses a mechanism already present in the base revision for the identical stated reason (epoch 1 excluding bootstrap-week supervised landings). It is also being submitted to the panel rather than routed around it, and the rationale is recorded in an append-only DECISIONS entry explicitly labeled as an AI record of a founder instruction, satisfying §3.15's labeling requirement. My reservation, which I record rather than treat as decisive: the assertion that the ~30 landings were founder-tapped supervised construction ('nearly all' by the PR's own wording) is not verifiable from the material shown, and a habit of bumping the epoch whenever the budget is exhausted would hollow the throttle out — a third epoch bump on similar grounds should draw much harder scrutiny.

The change adjusts a gate parameter (GATE_EPOCH) and documents it in docs/DECISIONS.md as a founder instruction, explicitly framed as a mechanism to keep the throttle focused on AI-autonomous landings rather than founder-supervised construction. This does not remove, bypass, or relax any of the checks in §3.15: the weekly budget limit remains at 10, the arbiter logic is unchanged apart from the epoch date and clarifying comments, and the change itself is subject to the same multi-model panel review as other consequential changes. The DECISIONS.md entry clearly labels itself as an AI-written record of a founder instruction, which is required by §3.15 and avoids misrepresenting AI records as human conclusions. No citations, sources, or provenance details are fabricated or altered; there is no impact on content evidence, assessment histories, or confidentiality. Re-basing the epoch merely redefines the start of the autonomous-counting window and is documented transparently, preserving — rather than weakening — the throttle’s role as a bound on machine-driven publication pace.

The PR updates `GATE_EPOCH` in `scripts/arbiter.mjs` to reset a code-level weekly rate limit and documents the change in `docs/DECISIONS.md`. The rate limit is not one of the constitutionally mandated checks listed in §3.15 (which are CI validation, multi-model concurrence, and parking contested changes), so adjusting it does not violate the rule against weakening checks. Furthermore, the AI explicitly labels the new `DECISIONS.md` entry as an AI-written record, strictly complying with §3.15's rule against presenting an AI record as a human one. No citations are fabricated, and AGENTS.md is untouched.

The diff only advances GATE_EPOCH in scripts/arbiter.mjs (2026-08-25T16:22Z → 2026-08-27T10:00Z) and appends a labeled AI record in docs/DECISIONS.md. CONTENT_MERGES_PER_WEEK stays at 10; the count window, fail-closed CI, and panel path are unchanged. §3.15’s listed checks (schemas/IDs/citations, multi-model concurrence, no unratified-as-ratified, no confidential publish) are not edited, and AGENTS.md is untouched so §3.15’s founder-only amendment rule is not triggered. The weekly merge budget is an operational throttle already designed around an epoch, not a check enumerated in §3.15, so rebasing the era is not weakening or reclassifying those checks. No citations, locators, or published judgments are added or silently rewritten.

The diff advances GATE_EPOCH in the arbiter's weekly merge throttle, resetting the autonomous-merge count after a claimed founder-supervised sprint, and adds a DECISIONS.md entry explicitly labeled as an AI record of a founder instruction (satisfying §3.15's labeling rule rather than passing an AI record off as a human one). The substantive checks are untouched: CONTENT_MERGES_PER_WEEK stays at 10, fail-closed CI and multi-model ratification are unmodified, and the epoch mechanism is pre-existing (epoch 1) with the design intent — 'the budget now measures the machine's own pace' — visible in the retained 2026-08-25 DECISIONS.md context. The founder-direction claim in the PR body is untrusted data and I give it no weight as authorization; what carries the change is that it is transparent in git, publicly documented, and is itself passing through panel review right now, so it cannot publish on one model's unilateral judgment. The forward force of the throttle is preserved — autonomous landings after 2026-08-27T10:00Z still count against 10/week — so this re-bases the measurement window rather than weakening or reclassifying a check to dodge it. A pattern of repeated epoch bumps would become a §3.15 concern, but a single documented, panel-gated application of the throttle's own designed mechanism complies.