[andreivms]

Summary

Check the live isDefault flag before expensive focusability geometry when finding a default element. Unmarked candidates cannot be the answer, so they do not need those reads.

This is a live predicate reorder, not a cache or traversal rewrite. Visibility/accessibility checks still run for actual default candidates, and live property changes remain observable.

Evidence chain

Public Mover/findDefault search context
-> source performs geometry before a cheap live default-flag rejection
-> generic real table/default-query reproductions expose unnecessary reads
-> reject non-default candidates before geometry without caching or changing traversal

Implicated stacks during unresponsive renderer

Public Mover, findDefault and context/virtual-parent traversal establish the search context. The dominant context/traversal work is not reduced by this change; no entire-workload or renderer-unresponsive causal claim is made.

Baseline result

• A 15-cell table default query with no default candidate: 15 style plus 15 rectangle reads -> zero; its null result is unchanged.
• Trusted keyboard entry into the table: 61 style plus 61 rectangle reads -> one plus one.
• Walkers, filters and virtual-parent calls remain 5 / 94 / 662.
• Focus results and the exercised keyboard sequences are unchanged.

Geometry API reads can be cached browser work. They are not forced-layout counts or measured production CPU savings.

Scope and limits

• Regression coverage includes live defaults, hidden/disabled/fieldset/programmatic/shadow/modal/root modes and real keyboard/DOM behavior.
• No cross-query cache, dependency or public-type change is introduced.
• This independent change does not contain the backward-enumeration repair.