docs(lab): HEADARC3 — logged-children ordering law (multi-day) + regression lock

[mikegreiling]

HEADARC3 — the logged-children ordering law, multi-day discrimination

HEADARC2's captures were all same-day, so its "most-recently-completed first" could not distinguish stopDate DESC from date DESC then index ASC from index ASC. This probe builds a multi-day fixture with a same-day tiebreak pair that separates all three.

Fixture (lab/scripts/research-headarc3.sh, golden things-lab-golden-v1 · Things 3.22.11 · guest clock advanced 2026-07-05 → 2026-07-09): project HA3-Foo-P, heading Foo, children A,B,C,D,E created in index order, completed B (day1), C (day2), A (day3), then D then E same day (day4). So index order A<B<C<D<E ≠ stopDate order B<C<A<D<E, and the same-day D/E pair (completed in index order) renders E,D under a stopDate-DESC law but D,E under an index law.

Verdict — all three contexts render E, D, A, C, B:

context	shape	sort
flat project logged toggle (heading OPEN)	continuous list, per-row date + muted heading sublabel, no day sections	stopDate DESC, within-day by completion instant
archived-heading group (Foo archived)	Foo grouped section header + nested children, no per-child sublabel	stopDate DESC, within-day by completion instant
global Logbook	day-grouped (Yesterday / month headers), per-row date + project sublabel	most-recent day first, within-day stopDate DESC

The same-day pair renders E before D (E's stopDate is 5.8 s later) → the within-day tiebreak is the completion instant, NOT child index. The maintainer's literal B,C,A-over-three-days answer is A, C, B. HEADARC2-B's in-project-toggle-labels-HEADING vs Logbook-labels-PROJECT asymmetry re-confirmed.

Shipped rendering CONFIRMED — no divergence. src/read/project-view.ts already sorts both the flat logged region and the archived-heading group children by stopDate DESC (full-timestamp, so same-day ties resolve by instant, matching the GUI), and the flat continuous shape matches our flat logged[] / loggedHeadings[].items — no shape change. The contract's stopDate DESC "Read views" claim is confirmed (accrues in the assumption register per the version-stamping doctrine, not a contract edit).

Changes

• docs/lab/headarc3-logged-ordering.md — campaign doc (byte evidence, HEADARC3-1/2/3 verdicts).
• lab/scripts/research-headarc3.sh — build driver (multi-day fixture + byte snapshots; GUI phase via guest screencapture + one vncdotool toggle click).
• test/cli/render.test.ts — two HEADARC3 regressions locking the law (index≠stopDate fixture asserts E,D,A,C,B for both the flat region and the archived-heading group).
• docs/reference/assumption-register.md — RD-7 (logged-region ordering = stopDate DESC), Confirmed under 3.22.11.
• docs/reference/README.md — probe-id index row.

No behavior/contract change; no CHANGELOG (not user-visible). VM headarc3-lab torn down (only the golden remains).