usually what i did is to take an abstract spec, derive thick layers / modules to decompose the problem, and then look at the deadline to see what MVP i can draw in that space.

whenever that mvp is not what was expected, if i'm lucky enough, the decomposition allows for easy adjustements to match the need