The time of the event doesn’t necessarily coincide with any of the times that you’re checking. That’s the whole point of looking for visual cues. Again, if the event happens 34 minutes into the video, and it leaves AI detectable visual cues for 10 minutes, the AI will never find it using binary search. It will skip to 30 minutes, see nothing, skip to 45 minutes, see nothing, skip to 52:30, see nothing, skip to 56:15, see nothing, and fail at some point when it can’t divide the video further. Binary search would fail in this scenario. It’s not just useless, it’s an abject failure, and the AI was a waste of processing power when you could have scrubbed forward five minutes at a time instead. That would have found the visual cue, but would not be a binary search.