b8a51d9ee2
iOS gives a page one audible video at a time. Start a second one with sound and the system pauses the first, so on an iPhone a split of two showed one panel playing and one stopped -- the feed-wide unmute set every panel audible, and the panels then took the audio session from each other in turn. Muted video carries no such limit, so where the rule applies exactly one panel keeps its sound and the rest keep their picture: four pictures moving is what the split is for, and four audio tracks at once was never the point. The sound follows whichever panel is asked for it, and a panel the system stopped on the way is started again. Nothing in the platform announces the rule, and by the time the symptom shows -- a video we started, stopped by something that isn't us -- a panel has already died, so it is read off the device rather than discovered. Everywhere else nothing changes: panels keep their own sound. No iPhone here, so the rule is installed into Chromium and the real feed is driven through it. The last check breaks it on purpose -- unmuting every panel by hand does stop one -- so the simulation can't quietly become a no-op. Against the code before this commit the test fails seven checks, showing exactly what was reported: both panels unmuted, the first paused. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MPZFnNdHbPGDTqQUNiE4ZN