Couldn’t you implement the same logic with a static reader effect? That is, instead of implementing a new dynamic effect SomeAction, why not define someAction = ask >>= SomeActionWithCtxt, then interpret the static effect SomeActionWithCtxt r without running ask? So I wouldn’t say that dynamic effects are absolutely mission-critical for your use case.
Not sure if that’s helpful or not, but the “substitution metaphor” sounds a bit like the difference between effectful by-name evaluation (dynamic) vs. effectful by-value evaluation (static). In terms of this thread, static effects seem to be “squeezed dry” at the interpretation site before substitution, whereas the interpretations of dynamic effects are substituted wholesale into use sites.