video1.0<iframe src="https://www.loom.com/embed/d8f1ff2b4a594a6485ac96e79f8cb40f" frameborder="0" width="1112" height="834" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>8341112Loomhttps://www.loom.com8341112https://cdn.loom.com/sessions/thumbnails/d8f1ff2b4a594a6485ac96e79f8cb40f-dd74ba6a6b32238b.gif185.451This Loom describes a self-learning harness for playing Pokemon Showdown using a player model and a coach model. After each battle, the coach reviews the battle logs and the player’s turn-by-turn reasons, then suggests new rules to improve the harness in a recursive loop. The harness starts with no tools, and tools are gradually implemented over versions 0 through 5 and 10. Results are logged to MongoDB along with a reason for every turn made by the player.