FLYWATT

A 38,178-neuron, 5.1M-synapse visual-to-wing-motor subcircuit of MaleCNS v1.0 is run as current-based leaky integrate-and-fire in a Web Worker with every constant declared and commented in one file (20 ms membrane, -65/-50/-70 mV rest/threshold/reset, 2 ms refractory, 5 ms synaptic decay, seeded 0.5-1.5 ms delays), and the bulb's brightness and the wattage figure are computed from the running spike count, so simulated neural activity genuinely drives the output. This is among the best-documented L1s in the batch - the repo ships a verification script that asserts the claims the site makes and reports the looming causal chain neuron type by neuron type. Not L2: the synapse-count weight array is an Int16Array read only on the right-hand side of the conductance accumulation and is never written; the calibration step does adjust a single global synaptic gain, but that is an offline bisection that bakes a fixed constant into params.ts before any run, not plasticity at run time, and the only state that adapts during a run is a per-neuron running mean of its own input, which is intrinsic adaptation rather than a synaptic change. The ~6.5 uW figure is an energy estimate applied to spike counts, using a mammalian ATP-per-spike number the README flags as such.