FlyBrain

The plasticity here is real and it is in the fly circuit. `Plasticidad.tick` multiplicatively depresses KC-to-MBON synapses in the network's own sparse weight matrix, in place, gated by the coincidence of a roughly 1-second Kenyon-cell eligibility trace with dopaminergic firing above each DAN's own running baseline; which MBON a given DAN modulates is read from the connectome's actual DAN-to-MBON synapses rather than hand-assigned, PPL1 carries punishment and PAM reward, and a floor keeps weights at 10 percent of their measured value. It runs every tick of an actual fight (`aprender=True` by default), on a private copy of the connectome so one fly's learning does not leak into others, and the repo ships a paired no-plasticity control. No validation mark: there is no published comparison against real fly measurements, and the project's own verdict is honest that it has not got there - the script prints that accuracy does not improve more with plasticity than without, and the source states punishment cannot make the fly cautious because the approach MBONs barely depend on Kenyon-cell input (12.8 to 12.2 spikes even when that input is depressed).