If the named model were assigned this stage, which of its supported effort levels is the lowest likely to complete it correctly with no rework, about 9 times in 10? Judge the work remaining given the issue, not the size of its text. Judge this model independently of which model class another question selects. More effort cannot compensate for every capability gap: choose insufficient if no supported level meets the bar. Choose not_needed if no work remains for the stage. These are routing recommendations, not execution settings.