FlyBrain-HalfLife

Off the scale, because no connectome data is used. The README badges MaleCNS v1.0 and FlyWire and reports a '12,260 neuron / 428,400 synapse MaleCNS v1.0 subgraph', but the graph the pipeline actually runs is generated in code: main.py calls ConnectomeDataLoader.build_canonical_flywire_connectome(), which - with no data file present - wires every tract with the closed-form index formula (arange(src)*3 + shift) % dst_len and invents the synapse counts, while the parallel core/connectome.py builds its edges with rng.integers at fixed connection probabilities and places somata at random gaussian coordinates. No connectome file ships with the repo (no .feather, .csv, .npz, no cache), and dataset/flywire_downloader.py is never imported by any pipeline code, so nothing ever fetches the real tables either. The LIF engine and the 3-factor KC->MBON STDP rule in dopamine.py are genuinely implemented and the STDP does write weights - but into that synthetic matrix, so the write says nothing about fly synapses. main.py also imports env.arena, which does not exist in the repo, so the documented entry point cannot run as shipped. To rate this, a reader would need a code path that loads real MaleCNS or FlyWire edge tables into the matrix the LIF engine steps.