Closed-Loop Fly

The full 166,700-unit MaleCNS graph runs as a leaky integrate-and-fire network with documented tauM/tauS/threshold/refractory constants re-exported unchanged from the upstream reference, plus an optional activity-dependent threshold shift - that shift is an intrinsic (spike-frequency adaptation) property, not a synapse. Not L2: the only line in the repo that writes a synaptic weight is the octopamine flight-state gain, and it multiplies every synapse onto an LPTC once at network construction from a fixed query parameter. Nothing changes a weight while the loop is running, and no optimizer touches the graph either. Notably honest: docs/followups.md retracts the project's own octopamine hypothesis and reports that a readout artefact inflated the published haltere and vision results.