> It's far too fast to be re-encoding the whole video and is too precise to be cutting at the nearest keyframe.

Correct! For example, if you trim the end of a video in the middle of a GOP, it includes the entire GOP as is, but only plays up to the point where you trimmed.

Which is a very bad solution! Imagine cutting a video to remove some very privacy sensitive material, and some of those frames still ending up in the file, just not viewed because in theory the video file politely asks the player to skip until a later timestamp.

I'd for sure rather have tools that actually do what they claim to do. Using container metadata tricks feels to me as lazy and misleading to the user.