# Reward Program Repair

Repair the current `reward(state, action, next_state)` implementation using the provided mismatch examples.
Do not rely on any benchmark or task name; use only the provided interface, schema, current program, and mismatches.