The diff touches only docs, an experiment README and script, and one UI label. No content/ record, citation or AGENTS.md line changes, and the citation tool found nothing to check. The count.mts change is a comment and an output string only; `higher`/`lower` were already computed over `far`, so the corrected wording ('every seat more than one step from the judgment') now matches what the script counted, and the README's output matches the new template exactly (20/14/0/6). The DECISIONS.md fix is made in place but discloses the first wording, cites review note #454 and gives the AMZ-C023 counterexample. That is the §3.15 review-note process working as intended and keeps the revision visible under §3.14. The stamp-rule edit narrows a doc sentence to match unchanged code (`engagedChecks`, the 2026-08-26 decision); documenting an existing exception loosens no check and improves inspectability. The LatestStrip relabel ('records last changed <date> · the change history') fixes a label that was false on three cases and better serves §7's 'last content-update date, linking to the case changelog'. Minor note, not a violation: the Before Sputnik line was strengthened from 'grade it higher' to 'all five seats are more than one step above' without mention in the PR body. It is consistent with the split rule and the 14-of-20 tally, but the operator should confirm it against the README's per-seat table.

The change corrects the description of split direction to match the shown computation, which filters seats to those more than one step from the judgment before counting higher and lower verdicts. The decisions entry preserves the provenance of the former wording and identifies the concrete AMZ-C023 counterexample and review note, while git retains the prior revision. The UI label is corrected so it no longer claims that the displayed non-housekeeping date is the history’s literal last entry. No citations, content records, ratification records, constitutional text, or checks are added, removed, or weakened.

This pull request directly addresses review note #454 and panel observations from PR #453. It accurately clarifies the definition used for the direction of a split across documentation and analysis scripts, corrects an overstatement regarding seat directions on split claims, clarifies the rule on stamps for older reconsiderations, and corrects the UI text in `LatestStrip.tsx` to honestly reflect that the displayed date is the last substantive record change rather than strictly the latest history entry. No constitutional rules are violated, AGENTS.md is untouched, and the changes enhance precision and transparency.

The diff only corrects prose and a UI label that #453 got wrong, answering review note #454 and two smaller seat notes. The split-direction wording now matches what count.mts already computed (seats more than one step from the judgment); DECISIONS and AUTOMATION give the honest counts and the AMZ-C023 counterexample; the stamp sentence now states both the edition-verb rule and the pre-existing reconciles path without changing code or loosening §3.15; LatestStrip relabels lastContentUpdate so it is true when the newest history entry is housekeeping. No AGENTS.md edit, no citations, no case ledger, no confidence material, no fabricated provenance.

This PR makes three documentation and labeling corrections in response to a review note, touching only docs, a counting script's output wording, and one UI label — no case content, no assessments, no citations, and no edit to AGENTS.md. The corrections move in the constitution's direction: the split-direction statistic is restated to match what the code actually counts (seats more than one step from the judgment), with the earlier miswording openly recorded in DECISIONS.md rather than silently erased, which honors §3.14's requirement that changes record what changed and why. The LatestStrip label fix replaces a demonstrably false label ('the change history, last entry') with an accurate one ('records last changed <date> · the change history') plus a code comment explaining the discrepancy, serving §3.9's honest distinction between what a record shows and what a reader might infer. The AUTOMATION.md sentence about stamps is narrowed to match the 2026-08-26 decision as implemented in engagedChecks — a clarification, not a weakening of any §3.15 check; the code is unchanged. No fabrication, confidence exposure, constitution edit, or check-dodging is visible anywhere in the diff.