Focus traversal answers what TalkBack requests when it reaches a control. An interaction scenario also records what happens after the user activates it. Use it to investigate a radio choice whose amount or checked state is missing from an announcement, or whose state is wrong after moving focus away and back.

This is an explicit diagnostic command. Ordinary aloud android audits never activate controls. Use a disposable emulator and synthetic data: an activate step invokes the real focused control and can change application state.

Prepare the pinned TalkBack companion and navigate the test app to the intended window. The command uses an already enabled companion; it does not install packages, change accessibility settings, launch or navigate the app, or dismiss blocking dialogs. Its handshake verifies the target window.

aloud android-interaction \
  --package org.irs_public.aloud.fixture \
  --scenario examples/android-radio-interaction.json \
  --out aloud-report/radio-interaction

The example expects the native acceptance fixture's radio-en screen, with three radio choices followed by a disabled choice. On Android 14, its native radio state description is “selected” or “not selected”, while the checked value is a separate boolean. Adapt the fixed sequence and exact speech assertions to the app, screen-reader configuration and language under test. A different focus order should fail the assertions, not silently choose another control.

Steps use TalkBack's own first, next, previous and activate gesture actions. Each command executes once. Activation is never retried and is bound to the previously observed focused node. An app process, service session, window or unexpected focus change stops the sequence.

expect can require checkable, checked, stateDescription and case-insensitive literal speechIncludes / speechExcludes strings. Pair a positive checked-state phrase with exclusion of its opposites (the example excludes “not selected” and “unselected”); “selected” alone also occurs inside both opposite phrases. Speech checks use the utterances for that action, not a concatenated whole-screen transcript. Use independent assertions for the initial amount, activation announcement and refocus announcement; speech from an earlier action cannot satisfy a later check.

Each command retains the focused node before and after the action, including its native checkable/checked/selected values and state description. Bounded native click and content-change events preserve their source identity and timing. Raw broadcasts, command records and diagnostics remain available when a command or assertion fails.

Observation completion and assertion success are separate. A settled, silent activation can be a complete observation with a failed speech assertion. A click or unrelated content-change event alone does not prove the selected state or the expected announcement. Missing or inconsistent evidence never becomes a passing assertion.

The default source is TalkBack's speech-request listener. --tts logging additionally requires an already configured logging TTS engine and validates its independent receipts and queue accounting. Record the tested APK hash/version alongside the capture; repository provenance alone does not identify an installed APK. The engine emits synthetic silence. Neither mode records the sound a person hears. The result is diagnostic evidence, not a WCAG or Section 508 conformance verdict, and it is not fed into the ordinary audit ratchet or OpenACR automatically.

The companion is built from pinned Google OSS TalkBack source. It does not instrument the TalkBack version installed on a Samsung phone. Rooted emulator setup and the silent logging engine also differ from a physical device's configuration. Do not replace a tester's everyday screen reader to claim a vendor reproduction.

For a physical-device retest, retain app version/build, Android/One UI version, TalkBack version, language, TTS engine/voice and verbosity settings. Record the exact utterance on first focus, after each selection, and after focus leaves and returns. Keep English and Spanish outcomes separate. An emulator capture can narrow the defect to exposed state, native events or TalkBack speech requests; that evidence does not substitute for the affected device result.

The synthetic native fixture and test/android-interaction-smoke.mjs exercise activation, repeated selection and refocus with English and Spanish amount labels. Unit tests exercise malformed or stale evidence and failed assertions. Native execution is a separate CI check; unit fixtures alone are not a native speech pass. The acceptance workflow retains the raw evidence from each run.