Starting with a disclaimer, I don’t know, what follows is speculation only.

I suspect that the only requirement is that the slope of the chirp at encoding and decoding is the same.

Ultimately, it’s used to remove noise in relation to a reference level.

I haven’t had an opportunity to do the math to confirm this and I’ve bookmarked a few LoRa videos to see if they get into the mechanics, but I’m not holding my breath.