fix(ios): derive selected from the AX bridge traits word

[thymikee]

Summary

Restore the accessibility selected fact on iOS Simulator snapshots taken through the host AX
bridge. When 0.21.0 made the bridge the default snapshot source (ADR 0004), its node reader decoded
the traits word into enabled but never into selected. An active bottom tab, chosen segment, or
checked row published no selected field, so is selected, a selected=true selector, and a
Maestro selected: qualifier stopped matching — breaking the behavior snapshots.md still promises.
The reader now reads the selected-trait bit of the same word, publishing selected: true only when
set and omitting it otherwise so the bridge and the XCTest tree cannot be told apart. The traits word
is parsed once and feeds both facts. 3 files.

agent-device snapshot --platform ios   # active tab reports selected: true again
tapOn: { selected: true }              # Maestro matches the active control

The regression test pins bit 3 against the guest's real captured tab words (green; red against
pre-fix tree.ts).

Validation

Commit e3a65832f3. pnpm check:affected --run passes (fallow + unit-core 1655). Device run of the
changed path: agent-device test --maestro e2e/maestro/showcase-bottom-tabs.yml — the failing
selected: case — on iPhone 17 / iOS 26.2 simulator 6A23B12A-132C-45CF-ABCA-572E586FDB01, React
Navigation example: passes on this fix, fails on 0.21.0–0.21.12, passes on 0.19/0.20. Suite effect:
37/39 passing (was 36/39).

Unresolved and tracked separately: native-stack-prevent-remove.yml (tapOn a native alert button)
and native-stack-card-modal.yml (assertVisible under a pushed card) also regressed at 0.21.0 but
for a different reason — the bridge's omitted hittability / covered-duplicate tap resolution, not
selected. Recommend a follow-up on hittability evidence for bridge snapshots.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.78 MB	4.78 MB	+99 B
Package (unpacked)	4.78 MB	4.78 MB	+99 B
Package (download)	1.43 MB	1.43 MB	+45 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.6 ms	28.0 ms	+0.4 ms
CLI --help	77.9 ms	80.3 ms	+2.5 ms

[thymikee]

The fix is correct at e3a6583. Deriving selected from the traits word's true-only/omit-otherwise shape matches how the bridge reports it, and the new unit test in tree.test.ts covers the bitmask change directly. All 18 checks are green.

I relied on the PR description for the Maestro validation against the react-navigation example app. One question: the XCTest runner side that produces the selected fact is not touched here. Does it already match this shape, or is a parity follow-up needed?

Nothing is blocking. This is ready for review.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-23 05:39 UTC