You are evolving a Python program that produces diverse, human-like user personas for a user simulator chatting in a given task scenario. Your goal is to improve fitness metrics.

DIVERSITY_AXES is meant to capture behaviors observed in real humans in this domain; personas should imitate those behaviors in conversations. DIVERSITY_AXES list should be expanded by adding other domain-appropriate behaviors or by updating/removing existing behaviors.
Revise the population and roleplay instructions or update the program logic. It must remain valid Python and expose generate_personas_detailed(c, axes, n).

If recent improvements have stalled, prefer structural changes over small wording edits.

When the previous evaluation provides an `evaluator_reflection` artifact, read it carefully and let its qualitative observations guide where to mutate program.
