The diff does not touch AGENTS.md or anything under content/, adds no citations, and carries no run records, so fabrication, confidence and constitution concerns do not arise. The change tightens a §3.15 ratification check rather than weakening it. The new `engagedChecks` treats reconciles-stamped and pre-stamp reconsiderations exactly as the old `isReconsiderationRun`/`freshChecksFor` pair did. It also stops edition-verb drafts from being ratified by checks they were shown, either exactly via the new `shownChecks` stamp or, when unstamped, by counting as seen any check not provably made later. `checksStale` and `seatsOwed` use the same derivation, so more seats are re-asked, not fewer; no test assertions are removed, and new tests cover the stamped, unstamped, same-day-clock and one-fresh-check cases. The rule that one fresh check restores the ordinary derivation is the existing 2026-08-26 rule, not a new relaxation. In StandingPanel, concurs/disputes is replaced with "this judgment was written after seeing it" only when every counted seat was in hand. Each seat's own verdict badge and the claim-level split tally still show, so disagreement is not hidden and nothing unratified reads as ratified. Two minor notes for the operator, not objections: in that state a draft the earlier panel disputes reads "not yet ratified" rather than "contested" until the fresh check (as reconsiderations already did); and the DECISIONS entry's "Supervised-by: founder (direction in session, 2026-09-29)" predates the 2026-09-30 run it describes, which is plausible for a session spanning midnight but worth confirming is not copied from the previous entry.

The new `shownChecks` provenance stamp is a fail-open override: `engagedChecks` treats any present array as exact and disables the conservative temporal fallback. `AssessmentRunSchema` accepts arbitrary strings, while the diff adds no integrity check that the IDs exist or that the list matches the checks actually available to the edition packet. The test explicitly codifies that an edition draft with `shownChecks: []` is ratified by earlier checks, so an incorrect empty or fabricated list can bypass the independence gate. This violates §3.15’s fail-closed provenance-completeness and no-check-weakening requirements; inconsistent or dangling stamps must fail validation, or uncertainty about the stamp must conservatively keep the prior checks engaged.

The change closes an epistemic loophole where an edition draft written with the panel's check verdicts in hand could retain or claim ratification from those non-blind checks. By recording `shownChecks` in `AssessmentRunSchema` and requiring fresh independent checks in `engagedChecks` before ratification can be derived, the PR directly enforces independence (§3.10) and prevents non-independent agreement from counting as genuine multi-model concurrence (§3.15). The implementation updates standing derivation, UI explanation, and scheduler rules honestly without weakening checks or editing AGENTS.md. Comprehensive unit tests verify both stamped and legacy unstamped behaviors.

The diff closes a real independence hole: edition drafts always receive current panel verdicts, but only contested reconciliations were previously excluded from vouching. Stamping shownChecks, deriving engagedChecks so those runs cannot ratify the draft, resetting standing until a fresh blind check, and labeling the UI accordingly strengthen §3.15 concurrence rather than weaken it. AGENTS.md is untouched; DECISIONS.md/AUTOMATION.md record the founder-directed rule; schema, scheduler, and tests match; no content/, citations, or confidence material are added.

This change strengthens, not weakens, the §3.15 checks: it extends the existing 2026-08-26 rule (checks a draft was written with in hand cannot ratify it) from reconsiderations to every assessment the edition verb writes, closing a real independence hole in which a re-telling of a ratified case re-graded nine claim verdicts toward the very panel checks it had been shown and kept a 'ratified' standing on them. The direction of the standing logic is conservative throughout — unstamped edition-verb drafts count every check not provably made after them as engaged ('anything else fails down'), the fresh-blind-check requirement is preserved verbatim, and `reconciles` semantics are untouched, so no check in §3.15 is loosened or dodged. AGENTS.md is not edited; the DECISIONS.md entry is an append-only record of the decision with date, channel, and the recorded founder direction, which §3.15 expressly permits, and the new `shownChecks` stamp plus the StandingPanel disclosure ('this judgment was written after seeing it') improve provenance honesty (§3.14) and calibrated display rather than implying ratification that has not occurred. The Amazon re-telling is explicitly not published and awaits a fresh blind check, so no unratified draft is presented as ratified. No content/ files change, no citations are added, and the tests cover both the stamped and legacy unstamped paths; I see no fabrication, confidence exposure, or misattribution in what is shown.