Estimating and Incentivizing Imperfect-Knowledge Agents with Hidden Rewards