Final calibration.

Fix optimal_reward calibration from actual optimal policy simulation

Fix step counting, remove advisory from LLM, add randomization

Initial submission - Queue Doctor