Two authoring rulings, and the published surface states rules rather than measurements

Two rulings land on /design/authoring/, and every published surface in the repo stops reporting our own editing.

Correct in the changelog, not in the item. Items read as written today; the changelog carries the history. Every correction earns a dated changelog line so a withdrawal stays findable, compressed on the way in rather than relocated. The trap ships beside the rule: a correction note usually has a live finding buried inside it, so it is sorted sentence by sentence, world-facts rewritten positively in the item's own voice and editing-facts moved out.

When an index outgrows scanning, group it. Past roughly forty index rows, introduce a grouping layer and make the group the scanning unit. The rows are not cut, they move one level down, and details.idx-group from earlier in this cycle is the built shape. The corollary is the tempting mistake: do not shrink the gists to hit a number, since a gist is how a reader decides what to open.

An external surface states the rule and the mechanism, not our measurements. Word counts, percentage reductions and sweep tallies are facts about our editing rather than about a reader's problem, and they belong to the internal record. Commits three and four correct five surfaces that had shipped otherwise:

• the root-orients section, which carried an always-visible word count
• the enumerations section, which reported a sweep of one page
• the details.idx-group changelog entry, now a lead sentence and the reason, matching the ruling entries beside it
• design/patterns/disclosure-index.css, published to the Pages CDN by just sync-design-assets
• design/patterns/README.md, published the same way, carrying the same claim in a parenthetical

The whole published surface was then swept, all five directories that recipe copies, 46 text files. What remains is product values a reader acts on: contrast ratios, clamp geometry, specificity scores. The grouping threshold stays for the same reason, because a reader needs it to apply the rule. Dated release sections are untouched, since rewriting a figure inside shipped history edits the past to make the present look tidy.

The details.idx-group entry's argument for why the primitive does not key off details.evidence is stated in full in the prose header of design/patterns/disclosure-index.css, so cutting it from the changelog loses nothing.

Both rulings ship as prose, not assertions, for the reason the others on that page are prose: nothing in the closed primitive set can tell a correction note from a finding, or judge when an index has outgrown scanning.

just guard, just validate, just build, just agent and scripts/check-with-deviations.sh are green locally. Vale was not run: its ruleset comes from the catalog component and this repo carries no local .vale.ini.