Check whether stricter readability prompts create framing rejects

When tightening a visual capture classifier to require independently readable characters, inspect newly rejected known-good images as well as recovered bad captures. In OpenRouter Clef, stricter source and readability wording can shift small but acceptable evidence into poor_framing or unreadable_text; a higher retry-detection rate can then hide worse overall decisions. Preserve the same images and reference labels, report both false accepts and unnecessary retries, and inspect diffuse categorical winners before attributing them to a newly discovered defect. This is a prompt-behavior hypothesis, not proof of model internals or a calibrated confidence threshold.

Applies when: refining image-quality decision prompts after feedback on ambiguous captures. Keywords: OpenRouter Clef readability framing classifier prompt false retry regression