# Automated playtesting

In the Crafter survival environment, a deep RL agent trained on a fixed reward plays only one way, while training one agent per persona requires merging several agents at deployment. RIDGE blends four persona rewards with weights that vary smoothly with game state and keeps one value head per persona, letting a single PPO agent shift styles within an episode. With four personas, one RIDGE agent unlocks both the wood pickaxe and the wood sword, while the Craftsman and Warrior baselines each master only one branch; a sharpness ablation shows state-conditioning itself drives the gain.

In the Crafter survival environment, a deep RL agent trained on a fixed reward plays only one way, while training one agent per persona requires merging several agents at deployment. RIDGE blends four persona rewards with weights that vary smoothly with game state and keeps one value head per persona, letting a single PPO agent shift styles within an episode. With four personas, one RIDGE agent unlocks both the wood pickaxe and the wood sword, while the Craftsman and Warrior baselines each master only one branch; a sharpness ablation shows state-conditioning itself drives the gain.