Test simpler classifier outputs instead of assuming they help

A simpler output contract is not itself evidence of better visual classification. When replacing a multi-reason capture classifier with a binary action, benchmark source recognition, false accepts and unnecessary retries on the same fixed images and labels. In OpenRouter Clef, separating diagnostics or shortening the question can leave source/completeness errors intact even when the output is valid and confident. If wording and question count change together, report the joint candidate rather than attributing the result solely to binary versus multiclass choices. Preserve the stronger baseline when a cleaner-looking design regresses.

Applies when: redesigning image capture decision questions after reviewing classifier failures. Keywords: OpenRouter Clef binary classifier prompt visual source completeness regression