I wonder, with the right choice of mapping/filters, how much variation you can admit on the input side while still ending up with something that sounds basicall

I wonder, with the right choice of mapping/filters, how much variation you can admit on the input side while still ending up with something that sounds basically okay at the end. Context: I've been thinking about this recently whenever something to do with machine-written music comes up. My best attempt to describe what music is to me is something like "the patterned construction and resolution of rhythmic and tonal dissonance" — i.e. start neutral, create something that feels slightly uncomfortable, but hint through the pattern how it might be resolved to make you comfortable again. Kinda like storytelling, come to think of it. So what if instead of a series of Sierpinski triangles, you used hailstone sequences, or anything else that contains interesting complexity and eventually returns to the origin. Could you mix and match different filters ("reject anything that doesn't fit this equality") and maps ("natural numbers to pentatonic scale", "natural numbers to commonly used chord progression"), and then apply them recursively? Could you end up with something that people would interpret as "real music", or would it always still sound like obviously computer generated "not-quite-music"?