flyverse

The whole MaleCNS v1.0 graph (166,700 neurons, 24,559,135 signed edges, loaded whole with no pruning) runs as leaky integrate-and-fire in Rust with every constant documented at the top of the engine: 0.1 ms timestep, rest and reset -52 mV, threshold -45 mV, membrane tau 20 ms, synaptic tau 5 ms, refractory 2.2 ms, delay 1.8 ms, 0.275 mV per signed contact. Membrane voltages change under a documented neuron model, so L1. Not L2: there is no plasticity of any kind in the source - a grep across the Rust for plasticity, STDP, dopamine, eligibility, learning rate and any weight increment returns nothing, and the synaptic weights are read from the packed connectome and never written. This is one of the more honest repos in the batch: the README opens with a per-layer accounting of what is measured versus what is an engineered surrogate (the ventral-cord sensory drive is canned replay, ground locomotion is a speed cap, the room is invented), and it reports its own negative result - the haltere perturbation probe bounds any steering response at about plus or minus 0.02 and the authors say outright that the haltere-on/off difference 'is not evidence of anything' because the system is chaotic.