[chubes4]

Summary

Shell extraction refused to share a header across routes when the only difference was the capture's ordinal dialog panel handle (data-blocks-engine-capture-dialog-N in the panel id, the panel attribute and the trigger's aria-controls). Shell identity now ignores that ordinal in the same normalization that #2960 added for the projector's hashed dialog id, so such headers become one shared header template part. Before: the 3-page fixture's diagnostic listed only [{"area":"footer","slug":"footer"}]. After: a header part (3/3) and a footer part, with page and front-page including both.

What changed

• normalizeNestedChromeMarkup() in ShellExtraction.php maps data-blocks-engine-capture-dialog-<N> to the bare handle for identity only, so the id and every reference to it normalize the same way. It uses CaptureEvidence::DIALOG_CLASS rather than spelling out the name.
• Emitted markup is unchanged: the shared header part keeps the first route's handle on both the panel id and the panel attribute.
• New regression test tests/unit/captured-dialog-ordinal-shared-header.php (registered in composer.json) uses the issue's 3-page fixture. It checks that the header and footer parts are shared and that page/front-page include both, plus a negative control: headers whose dialogs differ in content stay unshared.
• Additive: only shell identity comparison changes. Headers that differ in anything other than the ordinal dialog handle are still not shared.

Verification

• passed on b7585ca. Base trunk@4f62f31 unchanged.

set -e
cd php-transformer
(test -d vendor || composer install --no-progress --no-interaction)
( while sleep 60; do echo "gate heartbeat: composer test still running"; done ) &
hb=$!
trap 'kill $hb 2>/dev/null || true' EXIT
composer test

• CI: pending

Cook run

• Gate 1: tree 0dfaf5720227da8e0551e2b6801a333df690df9c, fingerprint dd6dfeca89aecb9eef915019bb3abf5b615a71f72597b42fabbc82ded983d239

AI assistance

Tool: Homeboy (pi) · Model: claude-opus-5-5 (tier roadie/cook-heavy) · Used for: The AI read the existing #2960 normalization, wrote the reduced-fixture regression test with a negative control, applied a one-line normalization fix, and ran the php-transformer suite.

Closes #3549