EidolonZero builds a job-shop schedule one decision at a time. Before each decision it searches: it imagines how the schedule would unfold with the world model it learned, and scores those futures. This page draws that search as a tree, from the real position on the left to imagined states on the right.
The line on the left is the game so far: the instance's start and each move actually played (▸). The last circle on it is the real position now.
Branches from that position are the legal moves. Each one's imagined future gets its own colour: everything past the first move is the model's imagination. Before a search they are sized by the network's instinct (prior π₀); after it, by how much the search thought about them.
Size is simulations through a step; colour is the value picked under Color (by default how good the search thinks it is). Levels replays how the search grew.
Click a node to see its numbers and its stat card.
Simulations sets how hard it thinks (the time per move is measured on your machine); Show sets how much of the tree is drawn.
Keys: S search · P play its choice · A autoplay · space skip the growth · ? this card · arrows walk the tree · double-click folds a branch · wheel zooms, drag pans.