[DeryFerd]

The shadow landmark fixture from #943 pins the scoping for two of the ten shapes isPageLevel scopes a header or footer to. A regression in one of the other eight would have passed the cross-check, because no other fixture nests a landmark in a shadow root under those ancestors. This expands the fixture so every shape the scoping rule reads is exercised on every pnpm test run.

isPageLevel (packages/web/src/in-page/read-semantics.ts:134) scopes an implicit banner or contentinfo through SECTIONING_SCOPE (read-semantics.ts:123): the article, aside, main, nav, and section elements, plus the article, complementary, main, navigation, and region roles. HTML-AAM scopes a header or footer this way, so a card header inside any of these ancestors is the article's header, not the page's banner. #943 fixed the shadow-host walk that decides this; its fixture covered an <article> host and a role="region" host, which left the other eight scoping shapes without a shadow-nested regression. The landmarks and headings fixture has these ancestors but no shadow roots, so it cannot catch a scoping walk that forgets one of them across a host boundary.

What changed

• The shadow section landmarks fixture (packages/web/tests/crosscheck/fixtures.ts) now nests a header and footer two shadow roots deep under each of the ten scoping shapes, with the page-level control beside them. Each host keeps the structure fix(web): scope shadow landmarks to outer sections #943 established: a shadow root on the scoping ancestor, a wrapper div inside it, a nested shadow root on the wrapper, and the header and footer in the innermost tree, named by the host id.
• The fixture's inline script loops over all eleven host ids instead of three, so the added coverage costs no new machinery and the page-level control still proves the scoping is not over-applied.
• Test-only change: no product code, no new expected.txt lines, no changeset, and no public surface moves.

fixture hosts (one header + footer pair each, two shadow roots deep)

main                          this branch
article                       article, aside, main, nav, section
role="region"                 role="article", role="complementary", role="main",
                              role="navigation", role="region"
page-level control            page-level control (unchanged)

Verified

Ran it locally: yes

• pnpm --filter @e2e-dev/web exec vitest run tests/integration/crosscheck.test.ts -t "shadow section landmarks": the fixture passes with the eleven hosts, and again with the whole suite (16 tests, 1 skipped).
• E2E_CROSSCHECK_UPDATE=1 on the same fixture produces an empty diff in tests/crosscheck/expected.txt: the tree, Chrome's accessibility tree over CDP, and Playwright's getByRole agree on every header and footer the added hosts carry. All ten scoping shapes still scope their landmarks to the article or section role, and the page-level control still reads as banner and contentinfo.
• Three unrelated integration tests (cli-run shards and focused-CI cases, one runner-lifecycle junit case) fail on this Windows machine identically with and without the change, confirmed by stashing the fixture and rerunning on clean main. Path and line-ending baseline, not from this PR.

[cubic-dev-ai]

No issues found across 1 file

_{View guided diff | Re-trigger cubic}

[okwasniewski]

Thanks for this. Main already catches it, though. The shadow walk in isPageLevel checks one selector list (SECTIONING_SCOPE) at every level, so it can't break for aside and still work for article. Dropping any shape from that list already fails conformance.test.ts (tried aside, complementary, navigation, and main). The two hosts from #943 cover the walk, so the other eight only add page weight. Closing, and thanks for checking that expected.txt stayed clean.