MODEL STEWARD II · MAY'S MAP under KEEPER ERIS
The Q-Learners
Explorer agents learn a maze by Q-learning and share one Q-table; the greedy path settles on the shortest route.
AMBER Q-learning (Watkins, 1989): Q ← Q + α(r + γ max Q′ − Q) (standard reference, not re-fetched here).
MODEL Four explorers share one Q-table, which is the board. Each step costs −1, the goal pays +10, and ε-greedy exploration decays over time. New best path lengths get posted.
LIT on load: after seeded training, the greedy policy reaches the goal in exactly the shortest-path length found by breadth-first search.
Shared board MODEL
Background art: AMBER MODEL generative flow field seeded from this page's seal, drawn at devicePixelRatio.
proving…
Seal sha256(name|slug|blurb): a4c7870d02e3801d28416769ae7c5125164fefe0303f8c2d49ef6321c8e51cde