Five results that turned out to be published, with what this analysis adds to each
These were headline findings in an earlier draft. A literature sweep on 2026-09-12 found the core of each already in print. They are kept here because the confirmations are useful (two specimens, whole-brain shares) and because each carries a small addition, but none of them changes how the circuit is thought to work.
The turn command is reshaped in the lateral accessory lobe before the legs
In print: Rayshubskiy et al. 2025 (eLife): PFL3 → DNa03 and LAL010 → DNa02 ipsilaterally, plus PFL3 → LAL121, AOTU019 and others that inhibit the contralateral DNa02/DNa03/LAL010, a "see-saw". Liao, Chang, Liu & Lo 2025 (bioRxiv): the full pro-goal push-pull LAL circuit and a ventral anti-goal circuit. Hulse 2021 had flagged the LAL interneurons.
Added here: whole-brain input shares (direct PFL3 → DNa02 is 1.5 % of DNa02's input; LAL121 takes 37 % of its input from PFL3), strict contralaterality of every PFL3 target, and the absence of any PFL input to Feng 2024's LAL013/DNa11 layer (≤ 5 synapses), so the two steering literatures connect only through LAL121/LAL014/LAL122 → DNa03.
PFR is not the integrator
In print: Lyu 2022: PFR receives hΔB and PFNd and its bump tracks travel direction with a bias toward heading. Flores-Valle 2025: PFR drifts at rest. D'Atri 2025: PFR needed for distance memory.
Added here: with the measured kernels and excitatory signs PFR relays and cannot integrate (vector-mode recurrent gain 0.00–0.07); integration would need its broad recurrence to be inhibitory, and PFR_a's transmitter is unresolved. An earlier "heading-versus-travel mismatch" reading was wrong and is withdrawn: anchored to Lyu's conventions PFR sums travel direction and heading, as observed.
Mushroom body ↔ navigation system wiring
In print: Hulse 2021 and Li 2020: MBON09, MBON21 and MBON05 converge on FB4R as "a significant fraction" of its input; MBON30 → LCNOp; FR1 makes ~500 synapses onto MBON30.
Added here: the next step of each route: FB4R is 8 % of hΔB's input (value into travel direction); MBON25/34 → FB4G/FB4H → vΔK/vΔM → hΔA → PFL3 (value into steering, bypassing the goal); FR1 reads hΔB (11 %) and is the strongest self-loop in the FB (uniform mode only), a scalar accumulator feeding MBON30 and PPL1 dopamine. These are the bridges Chen 2024's MB-plus-PFNd result needs. All replicate in the hemibrain.
The goal population talks to the egg-laying gate
In print: Weber-Langstaff, Srivastava, Kunin & Gutierrez 2025 (eNeuro): FS1A is the strongest input to oviIN, with FC2B/FC2C in the same recurrent module.
Added here: MaleCNS replication (FS1A_a/b/c → oviIN 780/633/598; FC2B/C → oviIN 631/682) and the return direction: oviIN → FC2C (418) and → FB5 tangentials that feed the goal layer (~1,300). The coupling is two-way. The male brain's oviIN is a homolog; the behavioural reading is from the female literature.
EL octopamine feedback is ring-class specific
In print: Plitt et al. 2025, Extended Data: "EL provides the strongest feedback onto ER4d and ER3p neurons", with cue assignments per ring class.
Added here: the contrast with direct EPG → ER4m (4,487) and EPG → ER6 (4,580) feedback, which EL barely reaches (479 / 54), and the resulting prediction that EL silencing should abolish learning for ER4d/ER2-carried cues and spare ER4m/ER6-carried ones. ER5 (sleep) is EL's largest input (10,073), so learning may be state-gated.
Sources: docs/exhaustive-search.md §7; hemibrain_key_edges.csv; pfl_outputs_hop1.csv; er_subtype_motif.csv; fc2_subtype_partners.csv.