One spike.
Then takeoff.

What is real here,
and what is assumed?

Real: leaky integrate-and-fire dynamics in 16 visual pooling units and one DNp01-style giant fiber, driven by physical retinal expansion from an approaching disc. The escape follows the first spike after an 8 ms motor delay.

Assumed: convergence counts, synaptic weights, and the motor mapping are project choices inspired by the published LC4/LPLC2 to DNp01 pathway. They are not measured MaleCNS synapse counts and the fly mesh is an artistic adaptation. Full detail in docs/methods.md in the repository.

Controls: the lesion switch clamps the giant fiber while upstream keeps firing. The rewired switch keeps every in-degree but destroys retinotopy, and the escape fails. That failure is the point.

Food mode runs a second circuit on the same neurons-as-LIF core: odor sensors feed four walk-OFF inhibitory units that cancel a steady descending walk command, so the fly brakes at the sugar. Lesion walk-OFF and it strides past dinner. Rewiring still stops, because contact drive is global rather than topographic. A useful disagreement between two controls.

Anatomy from flybody, developed by Google DeepMind and HHMI Janelia, under Apache 2.0. Pathway biology from von Reyn et al. 2014 and Ache et al. 2019.

Made by Royal Pinto at Facetworks.