fix: make wait verdicts evidence-based

[thymikee]

Summary

• Classify wait failures from readable-capture evidence instead of poll index.
• Add structured wait reason/evidence fields; make capture stalls retriable and ordinary absence explicit.
• Gate current-surface decoration on structured reasons and document the taxonomy in CLI and website help.

Tier 1 only; native scroll-to-find remains unchanged.

Fixes #1515

Validation

• pnpm check:affected --run (358 files, 3,390 tests passed)
• pnpm build
• Live source CLI on iPhone 17 / Agent Device Tester verified wait_deadline_exceeded with readableCaptures: 1 and wait_target_absent with readableCaptures: 5; all sessions closed.

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.95 MB	1.94 MB	-12.1 kB
JS gzip	624.7 kB	620.7 kB	-4.1 kB
npm tarball	744.6 kB	741.1 kB	-3.5 kB
npm unpacked	2.61 MB	2.60 MB	-12.1 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.7 ms	26.9 ms	-0.8 ms
CLI --help	67.0 ms	63.9 ms	-3.1 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/sdk-selectors.js	+42.0 kB	+13.4 kB
dist/src/cli.js	-5.3 kB	-1.7 kB
dist/src/selector-vocabulary.js	-5.4 kB	-1.5 kB
dist/src/cli-help.js	-438 B	-106 B
dist/src/runtime.js	-43 B	+80 B

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://callstack.github.io/agent-device/pr-preview/pr-1570/
Built to branch gh-pages at 2026-08-03 15:13 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[thymikee]

Reviewed exact head e6149c5d9018108639efcf74e49210f8ace9d345 — no actionable findings. The wait taxonomy is now evidence-based end to end: target absence requires a completed readable observation, zero-evidence expiry remains retriable capture stall, and a deadline-consuming capture after readable evidence is classified separately. The production selector-wait route, current-surface decoration, landmark mismatch behavior, and typed retriable propagation are covered by regression tests; the claimed live iPhone evidence exercises both absence and deadline outcomes. CI is green and GitHub reports the PR mergeable. Ready for human review.

Residual risk: no separately authorized cross-vendor review was run.