Five questions we keep separate
Model validity: does the model measure what it claims? Interpretation method: are the conclusions drawn appropriately? Content accuracy: is the explanation faithful to the evidence? Workflow safety: are safeguards working? Outcomes: what changed for people, under which conditions? Evidence for one question is not proof of another.