Fly Chess Lab
FlyWire chess experiment with a Rust/WASM LIF simulation, spike traces, and disconnection controls. Uses an engineered, currently untrained move readout.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.Documented LIF constants (timestep, rest, threshold, delay, refractory) at the top of the simulation core.
https://github.com/tolatolatop/fly-chess/blob/0855b5fd90872e8f3c9669eab9e325292399cae6/native/src/lib.rs#L5-L9The membrane integration and spike/reset loop; the connectome weight array appears only on the right-hand side of a read, never as an assignment target.
https://github.com/tolatolatop/fly-chess/blob/0855b5fd90872e8f3c9669eab9e325292399cae6/native/src/lib.rs#L77-L95The move score is an untrained fixed +/-1 projection whose weights come from hashing the UCI move string; there is no optimiser or training step.
https://github.com/tolatolatop/fly-chess/blob/0855b5fd90872e8f3c9669eab9e325292399cae6/src/brain.mjs#L28-L41
A full FlyWire v783 graph (138,639 neurons, 15.1M edges, no thresholding or subsetting) run as a documented Shiu-style LIF in Rust/WASM: rest and reset -52 mV, threshold -45 mV, 20 ms membrane and 5 ms synaptic time constants, 2.2 ms refractory, 1.8 ms delay, 0.275 mV per synapse, validated against an independent Brian2 run. That earns L1. Not L2: the synaptic weight vector is built once from the connectome file and is only ever read during simulation, never written. The move readout is a fixed +/-1 projection derived deterministically by hashing the move string, so nothing is trained anywhere in the project. The README says so itself - each move resets from rest, no cross-move memory, no synaptic plasticity - and reports that the resulting play is weak.
Record
Nearby
All projects →MaleCNS simulation connected to ViZDoom through modeled visual inputs and a fixed button readout. Includes plasticity experiments and negative validation results; learned survival has not been demonstrated.
Research framework comparing a MaleCNS-constrained sparse recurrent controller with rewired graphs, conventional neural networks, and other controls on simple VizDoom tasks.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.