Creating an "overfitted" index on how well swe bench submitters do on problems OpenAI rated as: > 3: It is almost impossible to understand what you are being asked to do without further information.