What the wiring diagrams say about fly navigation

Four findings that survived an exhaustive pass over every fan-shaped-body population and a literature sweep for prior claims (2026-09-12), from the MaleCNS (2026) and hemibrain (2020) connectomes. Each page is self-contained and states what was already in print. A fifth page collects results that turned out to be published.

Against the models

1. The compass has a built-in brake that no published model includestwo brainssimulated

The shifter neurons get three times more compass input from the position they write to than from the position they read from, on both sides, so they anchor the bump. In simulation rotation collapses. Hulse 2021 saw the dense EB interconnection; the ratio, placement and consequence are new.

The missing middle

2. The home vector is stored in synapses, not activity, at hΔB's synapses onto a family of hΔ typestwo brainssimulatedhypothesis

No population has recurrence for an activity integrator. hΔH, hΔA, hΔI, hΔG carry column-matched hΔB input, walking-driven dopamine (FB4M, FB1H) and the reward octopamine neuron OA-VPM3. Only a vector integrator reproduces the behaviour.

3. Every store delivers the displacement sign-preserved; hΔM is the only inverter, so it is the return pathphases measuredsimulated

Anchored to Hulse 2021 and Lyu 2022 conventions, hΔB lands on the axonal arbor of each store, so FC2 receives "where I have gone". FC2 → hΔM → PFL3 at 178° turns it into "go back". One memory: dispersal read directly, return read through hΔM.

Unnamed inputs

4. The velocity signals come from cells nobody has namedtwo brains

PS196_b feeds the rotational (GLNO) and translational (FB3A) velocity inputs plus ExR2/ExR4; FB3A carries an ascending signal into PFNd and supplies 12 % of its input.

Already in print

Confirmations and minor additionsaudit

The LAL steering stage (Rayshubskiy 2025, Liao 2025), PFR (Lyu 2022), mushroom-body bridges (Hulse 2021, Li 2020), the goal-to-oviIN link (Weber-Langstaff 2025) and ring-class-specific EL feedback (Plitt 2025), each with what this analysis adds.

Method and audit: exhaustive-search.md. Repository docs: findings.md, uncharacterised-neurons.md, behavioral-task-map.md. Data: MaleCNS v1.0 flat connectome; hemibrain v1.2 traced adjacencies.