[mziolkowskii]

Summary

The first four nightly E2E runs from master (2026-09-19 → 09-22) each had one or two red legs, for four unrelated reasons, none of them in the library. This PR removes three of those failure modes at the layer they live in, without retrying anything the flows assert, and documents the fourth.

Night	Red leg	Cause	Here
09-19	android 0.82 / API 35	android-settle.sh: uiautomator dump returned 0 without writing the file; the adb shell cat under set -e killed the script before Maestro started	fixed
09-20	android 0.82 / API 35	Gradle: 403 from Maven Central for gson-2.9.1.jar while resolving the build classpath	fixed
09-21, 09-22	ios 0.87 / iPhone 17 (iOS 26)	smoke-mute-volume: the btn-mute tap never reached JS, so evt-muted never appeared	transport retry, root cause open (#5130)
09-22	android 0.87 / API 36	DeviceServerDiedException ... UNAVAILABLE on the first command of the first flow, 3 s after the driver came up; the other nine flows passed	documented only

Motivation

Nightly is meant to become a required check after 10–15 consecutive clean runs. Each of these failures would reset that streak while telling us nothing about the library.

For the iOS one, the simulator device.log from both failing nights shows UIKit delivering touch-down and touch-up to the app's window, followed by no onPress. In both failures the touch-down was followed by ~120 ms without touch updates, then a UIKit system-gesture state change, then the touch-up; in every passing run of that flow the state change came after the touch-up. The tap went 2.4 s after onEnded, so the settle added in wait-for-end.yaml after the 09-15 incident is not a complete answer. Evidence and open questions are in #5130.

The Android driver death gets no mitigation here. A warm-up maestro test before the suite (tried in an earlier revision of this PR) does not help: every maestro test process installs, connects and uninstalls its own driver, so the suite's first connection is still a first connection. A transport-level retry of flows that die in the driver is a possible follow-up.

Changes

• Test app: eventLog.press() counts presses per control and EventLogPanel renders one pressed-<testID>-<n> marker per press in the fixed marker row, above the scrolling log, so a flow can wait for a specific tap on an element that is never off-screen. Unit tests added.
• e2e/shared/press.yaml (new): taps a control and waits for that marker, re-sending the tap (max 2) only if it never appeared. The marker is set synchronously in onPress before the player is touched, so the retry covers tap transport only; every player marker is still asserted exactly once, and the controls set absolute values, so a second delivery is harmless. Used by the five flows that press on an idle player (smoke-mute-volume, smoke-seek, smoke-hls-seek, smoke-loop, smoke-replay-after-end); taps issued during playback stay bare tapOn, since iOS delivers those only once the player is idle.
• scripts/e2e/android-settle.sh: one guest shell per attempt; the dump file, not the exit status, decides whether an attempt produced a dump; only a dialog still on screen in the last dump fails the leg. New android-settle.test.mjs (fake adb/sleep on PATH), 6 cases including "dialog dismissed, then no dumps".
• .github/workflows/_e2e-android.yml: assembleDebug gets one more attempt after 30 s only when its log shows a 403/429/5xx from a repository; a compile error or a 404 still fails once.
• Docs: e2e/CONTEXT.md (lost-tap gotcha linking E2E: iOS 26 tap on an idle player reaches the app window but never fires onPress #5130, the second transport-only retry, why a driver warm-up does not work), e2e/README.md.

Platforms affected

• Android
• iOS
• visionOS
• tvOS / Android TV
• Windows
• Web

Type of change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that changes existing behavior)
• Documentation only

Test plan

• Automated: test-app/src/e2e/eventLog.test.ts (press counting, snapshot identity, reset), scripts/e2e/android-settle.test.mjs (6 cases). bun run test 197 pass, bun lint and bun typecheck clean.
• Flows: no assertion changed. Each flow asserts the same player markers once; only the tap transport is wrapped, and only for taps on an idle player.
• CI: the PR gate runs the Android legs with the new settle and build steps and the iOS leg with the press subflow on five flows. The nightly matrix is the real test for the iOS change (the failure showed 2 nights out of 4 on one leg); a re-sent tap shows up as a retry attempt in the flow's commands.json in the Maestro debug artifact, so the rate of lost taps stays observable.
• Not reproduced locally: the iOS tap loss needs a hosted iOS 26 simulator under load; evidence is from the nightly artifacts (runs 35553773836 and 35678849766).

Checklist

• I read the contributing guidelines.
• I added or updated tests that cover this change, or explained under "Test plan" why none apply.
• For a bug fix the E2E suite can exercise, this PR includes a Maestro flow that fails without the fix, or "Test plan" explains why none does.
• bun run test, bun lint and bun typecheck pass locally.
• I updated the documentation / README where relevant.
• If this changes how the library is used (API, props, events, behavior), I updated the AI agent skill (skills/react-native-video/) to match. (Not applicable: no library change.)
• This PR is focused on a single concern.
• I tested my changes on at least one platform, or described under "Test plan" what CI (which runs both) should show.