Fast rolling aggregates are swell. I meet a lot of people who are building trading systems and want this sort of thing, but it usually isn't a great choice because the perfectly rectangular kernel probably isn't the best possible version of the feature and because arbitrary kernels can be well approximated using a state of constant size rather than a large buffer storing a sliding window.

Are you aware of any efforts to apply DBSP's theory to a general programming language/environment? From my perspective, DDlog was the most inspiring project in the field of incremental computation, but it seems like all of these projects just lead to implementations of streaming databases or other similar commercial products that fit into Data™ pipelines (no offense). Incremental computation pops up everywhere, from databases to business logic to UI rendering and video game graphics, and I have this hunch that if the problem could be solved at a fundamental level and in an accessible way, we could have revolutionary gains for programmers and programs.