Yeah, we could easily prove with a video that L-cancelling is in, by finding 2 cases of the same air attack with 2 different recovery times. However, there's no way to 100% prove that L-cancelling isn't in the game other than having someone try it first-hand and report back (and even then, you'd have to trust it was a reliable source and that they are skilled enough that they would be 100% sure and try it with different timings, in case that changed, and so on).
However, we can speculate its removal based on A) the other changes made that seem to simplify things, B) the benefit to WiFi prediction algorithms its removal would have, and most importantly, C) the fact that in the videos, the lag time has been reduced to L-cancel speeds even though we know these guys aren't skilled enough to pull one off.
So, if its in, it will be much easier to prove it exists than to prove it doesn't. Hence why I'm not 100% sure like I am with Wavedashing, where there is video evidence to show it clearly can't be done any more.
I'll admit Kix's theory is interesting though - maybe it only "auto L-cancels" in cases where the attack collision frame hasn't yet started, but after it has its full recovery time? That could explain some of what we are seeing. I'll have to go back and study the frame timings closer, and try to notice if the recovery lag is different when it is before or after the hit frame.
Or...I could be lazy and just wait until someone from E for All reports back and get a definitive answer. I'm sure its one of the first things they'll test.