Phase 3 · the wiring behind the number

Why the flight-steering code fails here: HS reaches DNg02 only through relays that cancel, and reaches DNa02 directly

Synapse counts from the MaleCNS graph, left-hemisphere HS cells shown. Edge width ∝ synapses. There are 0 direct HS → DNg02 synapses in the whole graph.
visual input (optic lobe) relays (posterior slope · gnathal) descending neurons 17,000+ other synapses AN07B004 + · IB008 − · PS117 − · DNp54 − … HSS (left) horizontal-system LPTC · 1,173 output synapses HSN (left) 614 output synapses PS080 (left) GABAergic · 4,495 input synapses, 259 of them (6%) from HS OA-VUMa4 octopaminergic, unpaired · sign +1 under the Shiu / Xenova convention DNg02 (right, 14 cells) Namiki 2022 flight-steering population 17,896 input synapses · these relays are < 1% DNa02 (left) walking-turn DN · Xenova's turn channel lateralises 26 / 6 Hz from HS alone 112 + 126 synapses 125 synapses 172 synapses, inhibits 28 synapses, excites 47 synapses, direct What the numbers say A relay that inhibits (PS080) and a relay that excites (OA-VUMa4) meet on the same DNg02 cells; in the un-refit LIF they cancel to within 1 Hz. Inhibition can only lateralise a DNg02 that is already firing. Here that drive is one hand-set constant; in the fly it arrives from the body (AN07B004). DNa02 has the direct line, so it steers the loop. It is a walking-turn neuron: a labelled deviation from the plan, not flight biology.
bench/paths.mjs on the MaleCNS graph (Xenova/fruit-fly-simulation data) · bench/hs-inject.mjs · bench/out/dng02-sweep.md
github.com/ZeroXClem/closed-loop-fly