Each clip on this page follows a level sequence.
A mentor told me during my internship: if you want to show what a parameter does, make the test reproducible and change one thing. So every comparison here comes from a single Level Sequence. Same run, same red ball, same frames. Each clip is two renders of it with one value changed, side by side.
That's also why the system is driven from Sequencer at the moment. Not because it needs Sequencer, but because "it looked better that one time" isn't evidence.