On a _very_ high-level it is like a spreadsheet model of computation: build the dependencies between calculations, propagate value changes to the dependent calc

On a _very_ high-level it is like a spreadsheet model of computation: build the dependencies between calculations, propagate value changes to the dependent calculations, but with two important improvements: per-node memoization of the previous calculations and ignoring updates that will not lead to the change of something visible to the user in the UI.