166,700 real neurons.

The MaleCNS connectome — published in 2025 by Janelia, the MRC Laboratory of Molecular Biology, Cambridge, and Google — is a complete, real wiring diagram of a male fruit fly's brain and nerve cord. Roughly 166,700 neurons. 25.6 million synapses.

The haze around you is a schematic stand-in for that scale — not individually mapped positions. What's about to light up in the center is real.

We only need 2,476 of them.

Smell, taste, courtship, wing control, digestion — none of it has anything to do with dodging a log. We traced outward from the fly's real obstacle-detecting neurons to the real motor neurons they connect to, two hops deep, and stopped there.

Every position you're seeing is the neuron's own real reconstructed location in the connectome — not a layout algorithm.

435 neurons that watch for danger.

Highlighted here: real LC4, LC6 and LPLC2 neurons — looming-detector cells wired to fire when something in the visual field is getting bigger, fast. In a real fly, that's the circuit that notices an incoming swatter.

1,821 neurons of fixed, biological wiring.

This middle layer is mostly left alone. It's the fly's own real circuitry — not something we designed. We only ever adjust the connections feeding the final motor decision, the same way this circuit's actual plasticity is thought to work.

220 neurons that shout "turn."

Descending neurons — real cell types like DNp01, DNp02, DNp11, DNg40 — carry the final decision out of the brain. These are the fly's actual, empirically strongest escape-command neurons. We map their output straight onto Wiggly Gator's steering.

Dale's principle, enforced.

Every real synapse here is signed from actual predicted neurotransmitter data — acetylcholine and glutamate mostly excitatory, GABA inhibitory. A connection never flips sign during training. That constraint is real fly biology, not a modeling convenience we get to ignore.

The fly's own bilateral symmetry.

Every neuron here is tagged with which real hemisphere it lives in. Pick a side.

Eyes → wiring → escape.

Same real neurons, same real synapses — just spread apart along the direction signal actually travels, so you can see the shape of the circuit instead of a tangle.

Watch a decision happen.

These pulses are riding along real synaptic connections, sensory neurons toward motor neurons — a stylized visualization of activity flowing through the actual graph, not a literal spike-by-spike readout.

Only the last mile learns.

The eyes and the wiring in between stay fixed — real, unedited biology. Training only ever adjusts the synapses feeding the final motor decision, using a reward signal (did that turn help or hurt?) applied through real reinforcement learning. Nothing about the outcome is scripted.

Highlighted in gold: the actual synapses that were allowed to change.