Paper looks interesting. The tech side of the implementation (in the newer version of dfscore anyways) was looked after by someone else, so I don’t have technical insights there.
As far as musical applications, in most of the contexts I’ve used this kind of approach latency has been less of an issue since it’s been largely unidirectional communication (e.g. computer to performers), so latency wasn’t an issue. Even if it was 500ms latency, everyone had the same latency, so in real-time performance, nobody had latency.
For pieces where feedback was involved, it’s more complicated. Thankfully I’m working in more open musical forms, so the perceptual boundary for togetherness isn’t the same as in a tight/synchronized electronic music. There, 20ms give or take is on the edge of being noticeable. With acoustic instruments performing together the “tolerance” is higher than that.
On a more conceptual/aesthetic note, I’ve been getting into the idea of “agency” as a manipulatable parameter, creatively. For example, I’ve been working on this improv analysis stuff for a while now and part of the future development of that idea will be to use it to generate dfscores which are related to your improvisational tendencies in either a complementary or uncomplementary way. Or anywhere in between. So using analysis data from myself (or others), I’d be able to play with “agency” itself with regards to how much me vs my tendencies vs the opposite of my tendencies vs other’s tendencies etc… is present in a given improvisation, or improvisational prompt.
So them’s some of my thoughts on the matter, hehe.