Swat
Browser arcade game with a visible escape circuit: 6,000 MaleCNS neurons influence evasive movement alongside authored game mechanics. Circuit subset.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The point-neuron parameters are declared in one place and labelled as model assumptions rather than measurements.
https://github.com/hrook1/Swat/blob/5423f89274742055e1084f09a38ce10f8d07a7cc/src/neural/Brain.ts#L3-L19Weights are written once when the circuit is loaded, as synaptic contacts times sign times a fixed mV-per-contact constant.
https://github.com/hrook1/Swat/blob/5423f89274742055e1084f09a38ce10f8d07a7cc/src/neural/Brain.ts#L104-L108The per-step update changes only membrane voltage, refractory counters and rate estimates; the weight array is read, never modified.
https://github.com/hrook1/Swat/blob/5423f89274742055e1084f09a38ce10f8d07a7cc/src/neural/Brain.ts#L224-L247
The neural worker runs a simplified point-neuron model over a 6,000-neuron MaleCNS LC16-to-MDN retreat circuit, with parameters stated in one exported constant (1 ms step, membrane tau 20 ms, synaptic tau 5 ms, refractory 2 ms, delay 2 ms, rest -52 mV, threshold -45 mV, 0.275 mV per synaptic contact); membrane voltage relaxes toward an equilibrium set by decaying synaptic current plus looming drive, then spikes and resets. That earns L1. Not L2: the weight array is filled once at construction from contact count times neurotransmitter sign, and the only other reference to it in the whole neural module reads it to deliver a delayed spike. The README is direct about this - one fixed, fully enabled circuit, explicitly not a trained AI.
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.