Mouse Maze uses actual data from the MICrONS cubic-millimeter mouse cortex scan. Google Research contributed connectomics tools, static meshes and hosting; the dataset is credited to the MICrONS Consortium.
Embedded in this file
The neuron view retains the source skeleton coordinates and branch topology, with each cell centered and uniformly scaled. Those v661 morphology examples are separate from the v117 synapse sample; no identity match between versions is assumed.
How the rat decides
Real directed synapse counts become normalized connection weights. Artificial inputs represent walls, unexplored corridors and the cheese direction. Three recurrent updates produce directional preferences. These scores guide a depth-first exploration with backtracking. At 0.5 fees, shortest-path search takes over; at 0.8, selected walls become passable.
The firing activity, sensory inputs, motor assignments, and movement rules are invented for this game. The scan is not a trained navigation policy and does not recreate this animal's mind.
Sampling limits
The sample is incomplete and biased, not a representative circuit. Segmentation roots may be fragments or contain reconstruction errors. The full scan is not embedded.
Source files
The MICrONS Consortium (2025), Functional connectomics spanning multiple areas of mouse visual cortex. Source URLs and SHA-256 hashes are stored in the embedded data. See the citation policy.
Coin fees come only from the server's configured contract event. A community reward milestone is not an issued token or an on-chain claim; that contract integration will be added with the coin.