Is anyone here familiar enough with Conal Elliott's version of FRP to speak to whether there are limitations on the sorts of equations you can use to create animation with it? For instance, is it possible to satisfy the continuous time semantics of FRP.. while using an animation function that relies on a rough integration method (eg: Euler, Taylor, some ODE integrator, RK4, yadda yadda)? My hunch is "no", since you're introducing error, which might compound if you use the result of one integrated animation function in another integrated animation function. That error will produce non-determinism and violate continuity and introduce path dependence and 🤮 If that's the case, then does that mean you can only use FRP with exact / closed-form equations? BUT... Conal's examples use functions like sine, which (AFAIK) don't have an exact closed-form implementation and are approximate. (EDIT: Turns out trig functions are closed-form, though most implementations I've found are approximate, seemingly due to π being irrational) Am I just conflating the denotational semantics of FRP with the operational semantics of whatever functions are being composed? Does FRP even care about determinism? Context: I'm writing something that you could think of as a game engine / authoring tool, and I'm interested in using FRP for its continuous time semantics (basically, everything forming a DAG of continuous functions with uniform compositional properties, with events that can reconfigure the topology of the graph). My goal is to allow extreme uses of slow motion and rewinding in the authoring tool, with perceptually perfect determinism. But I can't square how this would work with, say, mesh-to-mesh 3d collision detection, or fluid dynamics. These algorithms always seem to use approximate integration, and they tolerate temporal/spatial error by saying "the user will never see it". That assumption doesn't hold in my tool.