Evidence leads: sources that could lift an approximation
Literature and data search of 2026-10-13, ordered by the approximation each source could lift, with the open rows re-read on 2026-10-16. A lead is not a result and changes nothing in the model: using one needs an issue and a lock committed before any run, as for every claim here.
On 2026-10-16 the four quantities that would lift a current blocker were searched again: a resting head-to-thorax pitch with spread and n, a deposited LC4 or LPLC2 spike rate, a muscle named together with a body actuator, and a Drosophila voltage-to-release curve. None appeared. The rows below record what that reading did change.
"Checked" says how far each row was verified:
- data: checked against the local MaleCNS tables or a downloaded data file;
- text: numbers read in the full text;
- abstract: from the abstract or the publisher page only; numbers to be read before use.
Licences marked "not checked" must be confirmed before any file is fetched into the repository.
Electrical synapses (#85)
No connectome-scale gap-junction map exists for any fly dataset: MANC states gap junctions were not annotated, and none was found for FlyWire, BANC or MaleCNS.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Eckstein et al. 2024, Cell, doi:10.1016/j.cell.2024.03.016 | lLN2T_abc (~13 neurons, predicted serotonergic in FAFB) have "a cell body placement and general morphology most similar to reported cholinergic local neurons" (STAR Methods; Fig. S5F legend). lLN1_bc is "positioned closely among them", not named a match. Morphology only, no staining. In MaleCNS: lLN2T_a 5, lLN2T_b 4, lLN2T_c 4, lLN1_bc 30, all class ALLN, acetylcholine | CC BY 4.0 | data (types), text |
| Hürkey et al. 2023, Nature, doi:10.1038/s41586-023-06099-0 | Flight motoneurons MN1–5: coupling coefficients 0.023 ± 0.003 (MN1–MN2, MN3–MN4) and 0.010 ± 0.003 (other pairs), non-rectifying, ShakB-dependent; a Brian2 model with 43.5 pS junctions and CC = g/(g + g_L) | Zenodo 7737730 CC BY 4.0 (data), 7740678 CC BY-NC 4.0 (code); neither file fetched | text (agent); licences from the Zenodo API (2026-10-16) |
| Phelan et al. 2008, Curr Biol, doi:10.1016/j.cub.2008.10.067 | Giant fiber → TTMn/PSI junctions are heterotypic ShakB and rectify, Gj min/max ≈ 0.21 (oocytes: shape transfers, magnitude does not) | open access | text (agent) |
| Pokusaeva et al. 2024, Nat Commun, doi:10.1038/s41467-024-53173-w | Dye-coupled partners of HS cells (HS–HS, HS–CH, HS–bIPS, HSE–H2, HS–VCNM/CNM-like); qualitative strength only. HSE, HSN, HSS are typed in MaleCNS | ISTA 10.15479/AT:ISTA:17488 | data (types), text (agent) |
| Ammer et al. 2022, Curr Biol, doi:10.1016/j.cub.2022.03.040 | Immunohistochemistry of the innexins across the CNS. The abstract: some are glial and some neuronal, and shakB is the most widespread neuronal innexin. Not a pair list. Removing shakB from VS/HS cells makes them oscillate; that is a functional result, not a conductance | CC BY 4.0 (Crossref, version of record). The PDF returned 403 | abstract (2026-10-16) |
| Lillvis et al. 2022, eLife, doi:10.7554/eLife.81248 | Expansion light-sheet detection of Inx6 clusters; proof of concept, no map | CC BY | text (agent) |
Kenyon cell sparseness v6–v8 restrict the eLN set to lLN2T (lLN1_bc as a variant) instead of the whole cholinergic ALLN class. The pair list for a later lock is below. Nothing in it is a conductance a LIF can take without an authored conversion, so no lock is written.
| Pair | Measured | In MaleCNS, counted 2026-10-16 | Missing before a lock |
|---|---|---|---|
| GF → TTMn, GF → PSI | The Phelan et al. 2008 row: heterotypic ShakB, rectifying; the oocyte magnitude does not transfer | DNp01, TTMn and PSI: 2 neurons each | Conductance in nS on these neurons, their input resistance, and the bodyIds of the recorded cells |
| Flight MN1–MN5 | The Hürkey et al. 2023 row. The numbers there were not re-read on 2026-10-16 | MN1 (4), MN2Da, MN2Db, MN2V (2 each), MN3L (4), MN3M (2), MN4a (4), MN4b (2), MN5 (2). A shared name is not an identification with the recorded cells | A citation that those MaleCNS types are Hürkey's cells |
| HS and the dye partners | The Pokusaeva et al. 2024 row: qualitative dye coupling only | HSE, HSN and HSS: 2 each | A conductance. Dye coupling does not set a weight |
Graded transmission and visual kinetics (#84, #97)
No static Drosophila voltage-to-release curve exists: Zheng et al. (2006, 2009) report a synaptic gain that adapts over seconds rather than a fixed curve, so scanning the curve remains the honest treatment. A search on 2026-10-16 found no such curve.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Groschner et al. 2022, Nature, doi:10.1038/s41586-022-04428-3 | In vivo patch clamp of T4 and its inputs Mi1, Tm3, Mi4, Mi9, C3 to the same edges at 30°/s; T4 rests at −60.4 ± 7.4 mV (SD, 192 entries, control genotype), input resistance about 5.3 GΩ; a threshold-linear voltage-to-conductance model per input | Edmond doi:10.17617/3.8G; CC BY 4.0 | data (used by T4/T5 v3) |
| Behnia et al. 2014, Nature, doi:10.1038/nature13427 | Linear-filter peak times: Mi1 71 ms (SEM 3.8) vs Tm3 53 ms (SEM 5.2), Tm1 56 ms (SEM 3.8) vs Tm2 43 ms (SEM 2.7); delays 18 and 13 ms (Fig. 3B, 3E; N = 7, 11, 15, 14 cells); all four respond with graded potentials | none deposited | full text, PMC4243710 (2026-10-14) |
| Gruntman et al. 2018 (Nat Neurosci, doi:10.1038/s41593-017-0046-4), 2019 (eLife, doi:10.7554/eLife.50706) | Whole-cell T4 and T5 responses to bars and moving bars in eight directions; graded, rare small transients | Janelia figshare 10.25378/janelia.5576101 (T4), …c.4771805 (T5); CC BY-NC 4.0 | text (agent) |
| Arenz et al. 2017 (Curr Biol, doi:10.1016/j.cub.2017.01.051); Drews et al. 2020 (Curr Biol, doi:10.1016/j.cub.2019.10.035) | Arenz's abstract: two-photon calcium imaging of the columnar inputs to T4 and T5, not a voltage-to-release curve. Drews's abstract: fast surround compression of local contrast, not a release curve. The 13-type kernel list was not re-read; both version-of-record PDFs returned 403, and a licensed kernel repository was not found on 2026-10-16 | Drews is CC BY 4.0 (Crossref). Arenz's Crossref record lists no CC licence. Kernel repository not found | abstract (2026-10-16) |
| Borst 2025, J Comput Neurosci, doi:10.1007/s10827-025-00914-5 | One passive time constant for every neuron reproduces the 13 kernels only with an H-current in L1 and L2 (a prediction) | CC BY 4.0; code without licence | text (agent) |
| Zheng et al. 2006 (J Gen Physiol, doi:10.1085/jgp.200509470); 2009 (PLoS ONE, doi:10.1371/journal.pone.0004306, …0004307) | Photoreceptor and LMC resting ranges in vivo; adaptive R → LMC gain, LMC adaptation τ 1.4–5.4 s | not deposited | text (agent) |
| Lappalainen et al. 2024, Nature, doi:10.1038/s41586-024-07939-3 | Per-type time constants and resting potentials, fitted on a task (not measured), arbitrary voltage units | flyvis, MIT | text (agent) |
T4/T5 v3 (locked and run, robust: neither) used Groschner's voltages and left per-type timing unset. T4/T5 v4 (locked 2026-10-14, run 2026-10-15: robust neither) sets Behnia's two delays (18 ms, 13 ms) through the slower cell's τ_syn and holds out the Gruntman recordings as the direction benchmark; flyvis values stay a fitted comparison only.
Motor map and body (#19, #34, #35)
In MaleCNS 406 of the 708 vnc_motor neurons already carry a muscle-named type (Ti flexor
MN, Ti extensor MN, TTMn, DLMn a, b, …); 293 carry generic codes (214 abdominal
MNad), 9 none (data). These names come from MANC, so what blocks a motor map is the link
from muscle to body actuator, not from neuron to muscle.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Cheong et al. 2024, eLife, doi:10.7554/eLife.96084 | MANC motor neurons by type with light-microscopy muscle matches and match confidence (Supplementary file 3); TTMn matched to SS02623 | CC BY | text (agent) |
| Azevedo et al. 2024, Nature, doi:10.1038/s41586-024-07389-x | FANC leg and wing motor neurons to muscles (female), linked to MANC by NBLAST. That link was not re-read; Table A1 is read in the G4 section | GPL-3.0 code; Crossref lists a Springer text-and-data-mining licence and no Creative Commons licence (2026-10-16) | abstract (NBLAST link) |
| Card & Dickinson 2008, J Exp Biol, doi:10.1242/jeb.012682 | Escape take-off: centre-of-mass speed over the first 2 ms of flight 0.48 ± 0.01 m/s (s.e.m.; voluntary 0.28 ± 0.02), Fig. 8A; mated females, 3 days old; looming disc; 27 escapes filmed (n of the speed not stated); middle-leg extension 3.3 ms, peak acceleration about 112 m/s² | © The Company of Biologists | full text (2026-10-14) |
| Zumstein et al. 2004, J Exp Biol, doi:10.1242/jeb.01181 | TTM twitch (already used). The abstract's measured peak force is 101 ± 4.4 µN in female Canton-S, reached in 8.2 ms. Their linear model, not the measurement, predicts 274 µN and 5.0 ms to take-off. The full text is not open access | — | abstract (2026-10-16) |
| Eldred et al. 2010, Biophys J, doi:10.1016/j.bpj.2009.11.051 | Jump-muscle isometric tension 37 mN/mm², Vmax 5.1 lengths/s: a Hill-type force–velocity curve | — | text (agent) |
| von Reyn et al. 2017, Neuron (ModelDB 230400); Ache et al. 2019, Curr Biol, doi:10.1016/j.cub.2019.01.079 | LC4 and LPLC2 contributions to the giant fiber potential as functions of angular velocity and size, with delays; LC4 spike-rate tuning itself is not measured | ModelDB; no licence stated | text (agent) |
The MANC supplementary table is joined to MaleCNS types (#142): the muscle is the type's. jump3d v2 is not a research gap. The body that ran has the TTM only and left the ground at 0.06–0.07 m/s; Card & Dickinson's 0.48 m/s is the take-off speed a leg push would have to produce, and that push is not in the body. The giant-fiber timing locks v1 and v2 ran uninformative (benchmarks).
Held-out validation (#36)
Berg et al. 2026 report that the sensory and motor periphery is largely isomorphic between the sexes, so female recordings can test non-dimorphic circuits, not fru/dsx ones.
The M4 v2 proposal records prospective candidate eligibility and Baker's noncommercial licence scope. The Baker identity audit advances anatomical correspondence only; response files remain unopened and no candidate is benchmark eligible. Empty sex annotations do not establish isomorphism.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Pacheco et al. 2021, Nat Neurosci, doi:10.1038/s41593-020-00743-y | Brain-wide responses to pulse and sine song by neuropil, with iav mutants as a null | Princeton DataSpace 10.34770/gv6w-5351, "none" | text (agent) |
| Cande et al. 2018, eLife, doi:10.7554/eLife.34275 | Behaviour of single descending-neuron activations | Dryad 10.5061/dryad.fr89c0c | text (agent) |
| Currier & Clandinin 2025, bioRxiv doi:10.1101/2025.03.06.641774 (preprint) | Visual responses of 91 cell types against connectome predictions. The data-availability section says the calcium imaging will go to Dryad at publication. No repository was found on 2026-10-16, and the response values were not read | CC BY-NC-ND | text (data-availability section, 2026-10-16) |
| Lin et al. 2014 (Nat Neurosci, doi:10.1038/nn.3660); Amin et al. 2020 (eLife 56954); Prisco et al. 2021 (eLife 74172) | APL: blocking it makes Kenyon cell responses denser; inhibition is local and scales with input — relational KC tests | open access (eLife) | abstract |
song.yaml v1 is locked and has run (2026-10-15): uninformative, the no-go. Fourteen of 32
neuropils resolve a sign, and the Pacheco response file was not opened
(benchmarks). A version
2 needs a measured JO band-pass or a carrier-tuned input. APL claw contrast v1 also ran, and
was uninformative
(validation).
Sensorimotor gates: registration, transduction, motor map (search of 2026-10-14)
What blocks a biological closed loop under the body-wiring ADR: the eye-map registration (G2, #179), a measured looming transducer (G3, #34) and a muscle → actuator assignment (G4, #19, #35). Searched by three agents; the decisive claims were re-read on the source.
Eye map → body registration (G2)
Zhao et al. 2025 build the eye-map frame from the eye alone: "we used the lens positions from both eyes to define the visual field's frontal midline (sagittal plane)", with the equator from the inversion in photoreceptor chirality and z = 0 "defined by lens positions" (Methods). No neck, thorax or resting posture enters it, although the scanned heads kept "the thorax and abdomen intact and glued to the head". No published resting head pitch of D. melanogaster relative to the thorax, with spread and n, was found on 2026-10-14 or on 2026-10-16. Hengstenberg 1991, below, gives a flight head pitch and a body-axis elevation, not that resting angle.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Zhao et al. 2025, Nature, doi:10.1038/s41586-025-09276-5 | Eye-intrinsic frame (above); field of view "from directly above to −70° in elevation"; flies "can coordinately redirect all ommatidia by up to about 15°" (citing Fenk et al. 2022) | CC BY 4.0; µCT on Janelia figshare 10.25378/janelia.29111339, 0.6–0.7 GB per fly | text |
| Lobato-Rios et al. 2022, Nat Methods, doi:10.1038/s41592-022-01466-7 | NeuroMechFly head posed by hand ("the head is rotated 20° along the Y axis" in the zero pose of the preprint): an authors' choice, not a measurement | preprint CC BY-NC | text (agent) |
| Wang-Chen et al. 2024, Nat Methods, doi:10.1038/s41592-024-02497-y | FlyGym's 721-ommatidium eye is an idealised hexagonal grid; the eye cameras were placed by hand against coloured pillars (Supplementary Note 5); no µCT direction map | FlyGym Apache-2.0 | text (agent) |
| Kim et al. 2017 (Cell, doi:10.1016/j.cell.2016.12.005); Klapoetke et al. 2017 (Nature, doi:10.1038/nature24626) | Head pitch relative to the rig in glued preparations (eye rows 65.6 ± 5.4° to the horizontal, n = 11; equator 55 ± 5° below it): preparation angles, not body ones | — | text (agent) |
| Hengstenberg 1991, in Synapse — Transmission Modulation, eds. Elsner and Penzlin, Thieme, p. 273 | One-page conference abstract, read from the author's PDF on pure.mpg.de. In flight the head is pitched down 23.0° ± 7.5° SD; body-axis elevation is given as 15.0° ± 4.2° SD walking and 45.5° ± 13.0° SD flying, cited as Götz unpublished. No n. At rest the head is held still, and no head-to-thorax angle is stated. Head-turn ranges (yaw ±25°, pitch ±35°, roll ±120°) are the range of movement, not a posture | Thieme 1991; the PDF is the author's deposit, not an open licence | text (2026-10-16) |
| Whitehead et al. 2015, J Exp Biol 218:3508, doi:10.1242/jeb.122622 | Body pitch control in free flight after impulsive torques: recovery of the body pitch angle, not a head posture. Excluded by entry protocol 4 | © The Company of Biologists | title and abstract (2026-10-16) |
| Ispizua et al. 2026, bioRxiv doi:10.64898/2026.05.03.722293 (preprint) | Free-walking 3D tracking of 50 keypoints, including head and thorax. No head pitch is reported. Code is named at elliottabe/3d_tracking_ik and moments-behavior. The curated kinematics are a Google Drive folder linked from the preprint (bit.ly/3UWoXBF); that folder's licence is not stated |
CC BY 4.0 (Crossref, 2026-10-16). Dataset licence not checked | text (data-availability section, 2026-10-16) |
Public raw data checked on 2026-10-18 (agents; metadata, scanner info images and file headers only, nothing over 50 MB downloaded) for a measurement of our own:
| Source | Public raw data | Verdict for a resting head-to-thorax angle |
|---|---|---|
| Zhao et al. 2025 µCT, doi:10.25378/janelia.29111339.v1 | Three PTA head volumes, ~1 µm voxels, ~1 mm field; CC BY 4.0 | Unusable: the glued thorax is at most a posterior sliver cut by the field (no scutellum); fixed posture |
| Lobato-Rios et al. 2022 µCT, doi:10.7910/DVN/PEOVAV | One resin-embedded female, 360-slice whole-fly crop; CC0 | Unusable for a dispersion (n = 1, resin posture); at most a pilot of landmark visibility |
Ispizua, Abe et al. 2026, Drive folder of 3d_tracking_ik |
Grooming, headless and stairs tables of the MuJoCo inverse-kinematics fit, checkpoints, one example bout; no licence file on the folder | Unusable as deposited: the pre-fit landmarks (kp3d.npz, listed as a pipeline output in the code README) are not deposited; the fit adopts the model's head joint |
| Haustein et al. 2024; Anipose; DeepFly3D; LiftPose3D | Tethered treadmill or prism/ventral recordings; leg keypoints, at most two antennal points | Unusable: tethered, or no head-capsule and thorax landmark pair |
| Schoborg et al. 2019; MorphoSource; Digimorph | No public raw volume (restricted download, or CT listed as n/a) | Unusable |
The eyemap → head-capsule half is measurable from public data (feasibility). The only candidate is the undeposited pre-fit landmark set of Ispizua, Abe et al. (13 males,
9 females, free behaviour, Antenna_Base, EyeL/R, Scutellum, wing bases, T1 coxae). A
measurement protocol for it is drafted (session scratchpad, not locked) and needs the data
first.
Next: still blocked. Hengstenberg 1991 and Ispizua et al. 2026 do not supply a resting head-to-thorax angle with n. A registration would be a new measurement: an eyemap-to-head-capsule rotation located in Zhao's µCT heads (the thorax is not usable there, see the table above) joined to a head-to-thorax pitch from free-walking keypoints. Yaw and roll, individual variation (n = 3 eye maps) and the female eye under a male CNS stay open.
Looming transduction (G3)
No deposited recording gives LC4 or LPLC2 firing rates for looming; the only LC4/LPLC2 looming data are calcium ΔF/F, with no stated conversion to spikes. What is deposited is the giant fiber's membrane potential to looming at four r/v values.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Dombrovski et al. 2025, Nature, doi:10.1038/s41586-025-09037-4 | Giant fiber (DNp01) whole-cell current clamp to looming at r/v 10, 20, 40, 80 ms (Extended Data Fig. 8), controls and mutants, 20 kHz traces; a second set of 5 flies × 3 elevations × 5 trials, discs 0 → 30° at 500°/s. The controls are genotype controls of the DIP-ε experiments (5 + 5 flies, 2 trials each); validation/gf_looming.yaml records that the disc sizes are inferred from Jang et al. 2023 and that the control flies' sex is not stated |
Zenodo 10.5281/zenodo.14983850, CC BY 4.0, 46 files, 33 MB | data (record) ; text (agent) |
| Ache et al. 2019, Curr Biol, doi:10.1016/j.cub.2019.01.079 | Giant fiber model: an LC4 term linear in angular velocity and an LPLC2 term Gaussian in ln θ, both with a 19 ms delay, and their weights (STAR Methods) | on request only; ModelDB 230400 holds von Reyn 2017's model code, not Ache's | text (agent) |
| Jang et al. 2023, J Exp Biol, doi:10.1242/jeb.244790 | Giant fiber whole-cell at r/v 10–80 ms, discs 10 → 63°, n = 10 females: the protocol the 2025 runs cite | on request | text (agent) |
| Dombrovski et al. 2025 (as above) | LPLC2 GCaMP7f imaging to looming (ΔF/F) | OSF 10.17605/OSF.IO/Z7XFK, CC BY 4.0, 36 MB | text (agent) |
| Klapoetke et al. 2017, 2022 (Neuron, doi:10.1016/j.neuron.2022.02.013); Turner et al. 2022, eLife 11:e82587 | LPLC2 and LC4 calcium imaging to looming (ΔF/F only). Turner's set covers both glomeruli | Klapoetke on request; Turner Dryad 10.5061/dryad.h44j0zpp8, CC0, 53 GB | text (agent) |
Next (locked 2026-10-15 as validation/gf_looming.yaml v1, timing only): an M2 transducer can
now be scored without a data request on the stage that ends at the giant fiber: an input shaped as Ache et al. 2019 drives the model's DNp01, judged against the
deposited r/v series with a held-out set fixed in the lock. The gains stay a scanned envelope, never fitted to the giant fiber, and the input is not a measured LC4 or LPLC2 rate.
Both versions ran uninformative.
- v1: the model's giant fiber fired in every cell.
- v2: the giant fibers were passive. Their peaks follow the stimulus, and three of the ten control flies peak at onset, so the flies' interval holds the onset slope.
A sharper test of this stage needs a measured LC4 or LPLC2 rate. A search on 2026-10-16 found no deposited spike-rate recording.
Muscle → body actuator (G4)
No source names a MaleCNS/MANC motor neuron or muscle together with a FlyBody actuator, so under
validation/motor_map.yaml's rule every row stays unmapped. Muscle → joint actions are
published, and FlyGym 2.1.0 now ships a muscle model.
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Azevedo et al. 2024, Nature, doi:10.1038/s41586-024-07389-x | Per-muscle joint actions for the female T1 leg (Supplementary Appendix Table A1: tibia extensor/flexors on the femur–tibia joint, trochanter extensors/flexors on the coxa–trochanter joint, tarsus depressor/levator on tibia–tarsus); the tergotrochanter's "role in T2 trochanter–femur extension during jumping"; "It was previously assumed that the Drosophila trochanter and femur are functionally fused" | supplementary PDF. Crossref lists a Springer text-and-data-mining licence and no Creative Commons licence (2026-10-16) | text (main); data (agent, Table A1) |
Özdil et al., arXiv 2509.06426 (ICLR 2026), FlyMimic; shipped in FlyGym 2.1.0 as assets/model/musculoskeletal/ |
15 Hill-type muscles of the left front leg named with their tendons and the joints they cross (coxa yaw/pitch/roll, trochanter yaw/pitch/roll, tibia pitch), grouped after Azevedo 2024 and Soler 2004; on a NeuroMechFly leg, not FlyBody; mid and hind legs not released; FlyGym ships the MJCF and fetches the meshes | CC BY 4.0 (paper), Apache-2.0 (model) | data (local MJCF) |
| Vaxenburg et al. 2025, Nature, doi:10.1038/s41586-025-09029-4 | FlyBody has joint actuators (and tendon actuators coupling abdomen and tarsus DoFs), no muscles and no muscle → actuator table; legs body–coxa, coxa–femur, femur–tibia, tibia–tarsus; no trochanter | CC BY 4.0 | text |
| Lesser et al. 2024, Nature, doi:10.1038/s41586-024-07600-z | Extended Data Fig. 6, read on PMC11356479 (2026-10-16): the tergotrochanter originates on the dorsal thoracic cuticle, crosses the body–coxa joint and extends the trochanter, and it inserts on the same tendon as the sternotrochanter and the trochanter extensor (citing Azevedo et al.). Biarticular, and not named as a FlyBody actuator | PMC11356479. Crossref lists a Springer text-and-data-mining licence and no Creative Commons licence | text (2026-10-16) |
Force and geometry checked on 2026-10-18 (agent; abstract-level items re-read):
Zumstein et al. 2004 (J Exp Biol 207:3515, doi:10.1242/jeb.01181) measure the jump muscle's
peak force through the mesothoracic legs of female Canton-S flies, 101 ± 4.4 µN, reaching its
peak in 8.2 ms (abstract read 2026-10-18); latency and decay landmarks are in the paywalled
Results and were not re-read. Eldred et al. 2010 (doi:10.1016/j.bpj.2009.11.051) give
skinned-fibre active stress, not whole-muscle force or torque. No measured T2 TTM moment arm,
tendon line of action or joint torque was found; FlyMimic's muscles are left-front-leg only.
NeuroMechFly fuses trochanter and femur, so the candidate is the
coxa-trochanterfemur-pitch hinge of each middle leg, an authored assignment. The chain
therefore lacks the published actuator pairing, the T2 lever and a spike-to-force kernel.
Next: still blocked as a published co-naming. A search on 2026-10-16 found no source that names a MaleCNS or MANC muscle together with a FlyBody or NeuroMechFly actuator. Lesser et al. 2024 confirm the tergotrochanter crosses the body–coxa joint; they do not assign it to an actuator. Candidate rows (tibia T1 from Table A1; TTMn on the T2 coxa–femur actuator) would need a documented two-source join, an antagonist rule and a decision on whether FlyBody's coxa–femur joint stands for the coxa–trochanter one, which Azevedo et al. 2024 put in doubt. Each is an authored choice for a new lock, not a citation.
Synapse strength and per-type parameters
| Source | What it gives | Data, licence | Checked |
|---|---|---|---|
| Liu et al. 2022, Curr Biol, doi:10.1016/j.cub.2021.11.056 | 29 measured PN → LHN unitary EPSPs (0.4–6.6 mV) matched to hemibrain: amplitude increases linearly with synapse density (count over whole-neuron area, r² 0.77 over 135 pairs, 0.84 dendritic); no count-only statistic is reported (corrected 2026-10-13). Pair means in Table S3 (Document S1) | Elsevier open archive (supplements); Zenodo 5716323/5716321/5711439 CC BY 4.0, code and morphology only | text |
| Liu & Wilson 2013, PNAS, doi:10.1073/pnas.1220560110 | Glutamate inhibits antennal-lobe neurons through GluClα (the basis of the sign rule, measured there only) | — | abstract |
| Shiu et al. 2024, Nature, doi:10.1038/s41586-024-07763-9 | The model's source; names circuits that need tonic inhibition as its failure mode (zero baseline) | MIT | text (agent) |
| flycns 0.7.0 (PyPI) | README on PyPI, read 2026-10-16. Compiles MaleCNS v1.0 and runs Shiu's LIF plus a graded optic lobe in the form of flyvis. The README says that transferring flyvis's numbers onto the MaleCNS columns makes T4 prefer its known directions. That is flyvis's fitted model, which this project does not treat as a measurement, and it was not re-run here. The README cites 166,700 neurons from Berg et al. 2026; this model is the 165,122 traced neurons. No numerical comparison was made | MIT code. The README says the MaleCNS data are CC BY 4.0 | text (PyPI README, 2026-10-16) |
Used by validation/weight_rule.yaml v1 (#139): on 57 PN → LHN pairs density ranks the
amplitudes barely better than count (Δρ +0.015, interval −0.011 to +0.054), undecided; count
stays the rule.
Entry-protocol search (2026-10-16, evening)
Three searches against the entry protocols, then an independent agent re-read the claims below in the primary text or repository record. No source is admissible: no protocol's input exists. Each row says what is missing. No response file of a candidate benchmark was opened.
| Source | Protocol | What it gives | Data, licence | Verdict | Checked |
|---|---|---|---|---|---|
| Yang et al. 2016, Cell 166:245, doi:10.1016/j.cell.2016.05.031 | 1 eye | ASAP2f voltage and GCaMP6f in L1, L2, Mi1, Tm1, Tm2, Tm3: "we observed a sign inversion between L1 and its postsynaptic cells" (Fig. 4G–H). Peak ΔF/F, pre and post in different flies | no data statement | partial: no mV, no paired column | text |
| Pang et al. 2025, Curr Biol, doi:10.1016/j.cub.2024.11.064 | 1 eye | L1/L2 voltage imaging with lamina feedback dissected; no medulla readout | Dryad 10.5061/dryad.ngf1vhj4c, CC0, 49.6 GB | not admissible | abstract, metadata |
| Kazama & Wilson 2009, Nat Neurosci 12:1136, doi:10.1038/nn.2376 | 2 electrical | Paired DM6 PN recordings, reciprocal coupling; coefficients below about 0.04 only in Fig. 7d | none | partial: no nS, no input resistance | text (author PDF) |
| Augustin, Zylbertal & Partridge 2019, eNeuro, doi:10.1523/ENEURO.0423-18.2019 | 2 electrical | Giant-fiber gap conductance "135 μS (estimated)" young, "34.5 μS" old, values "manually adjusted" to match latencies | ModelDB 245415, licence not stated | not admissible: fitted, recorded here so it is not reused | text |
| Kohn, Portes, Christenson, Abbott & Behnia 2021, Curr Biol 31:5249, doi:10.1016/j.cub.2021.09.061 | 2 modulation | Whole-cell Tm1, Tm2, Tm4, Tm9 filters in saline (n 8, 5, 6, 7) and 10 μM octopamine (n 4, 5, 4, 4) | code "GITHUB LINK TBD"; no data deposit | partial: per type in mV, values only in figures | text |
| Strother et al. 2018, PNAS 115:E102 | 2 modulation | Chlordimeform/octopamine on Mi1, Tm3, Mi4, Mi9, T4 (GCaMP); Mi4 strongest | on request; CC BY-NC-ND | partial: calcium, no gain | text (agent) |
| Apostolopoulou & Lin 2020, PNAS 117:16606, doi:10.1073/pnas.1921294117 | 2 APL/KC | APL>Ort, 2 mM histamine, the same hemisphere before and after: KC responses "dramatically increased"; APL>Ort 17 hemispheres (11 flies), flies "kept at 31 °C for 4 d" | Dataset S1 (182.3 KB xlsx) listed on PMC; scripted download refused; code GPL-3.0 | partial, most useful: both arms in one preparation; needs the dataset and a baseline arm without 31 °C | text |
| Dombrovski 2023 and 2025, Klapoetke 2017, Jang 2023, Zhao 2025 | 3 looming | Every patch recording is the giant fiber or a downstream DN; LPLC2 is driven, never recorded | — | not admissible: no LC4/LPLC2 spike recording exists in these | text (agent) |
| DANDI 001869 | 3 looming | Brain-wide two-photon imaging during visual threat, 13 flies | CC BY 4.0, 1.45 TB | not admissible: not identified cells, not spikes | metadata (agent) |
| Hengstenberg 1992, in Rhythmogenesis in Neurons and Networks (Elsner and Richter, eds), Thieme, p. 305 | 4 head | Head pitch "relative to the thorax": "-30.7° ± 6.6° SD, N = 16, n = 160, tethered flight"; mounted, not flying, upside up −25.3° ± 7.3° (N = 14); tarsal contact removed −6.9° ± 5.3° (N = 16). The zero is not defined and no free-standing value is given | author PDF on pure.mpg.de | partial: tethered, undefined zero | text |
| Ispizua et al. 2026, bioRxiv doi:10.64898/2026.05.03.722293 (PMC13174342) | 4 head | 13 male and 9 female flies, keypoints Antenna_Base, EyeL, EyeR, Scutellum, WingL_base, WingR_base, T1L/R_ThxCx (fly50.json); a stationary.yaml dataset config |
preprint CC BY-NC 4.0; repository has no licence; the data link in the text is an empty placeholder | partial: a head and a thorax frame are definable; needs data and a licence from the authors, then a measurement written before it is made | text, GitHub API |
| Lobato-Rios et al. 2022 µCT, Harvard Dataverse doi:10.7910/DVN/PEOVAV | 4 head | One fixed, resin-embedded female, 616 MB | CC0 | partial: n = 1, posture after fixation | metadata (agent) |
FlyGym 2.1.0 (flygym/utils/api1to2.py, assets/model/flybody/actuators.yaml) |
5 muscles | NeuroMechFly's Femur becomes trochanterfemur; actuators *_coxa-*_trochanterfemur-pitch/roll |
Apache-2.0 | partial: names the joint as coxa–trochanter with trochanter and femur fused, which Azevedo et al. 2024 dispute; no muscle named | data (installed package) |
| Fenwick et al. 2026, Biophys Rep, doi:10.1016/j.bpr.2026.100268 | 5 muscles | TDT myofibrils, females 2–3 d, 17 °C, n 27–31: "19.8 ± 10.5 mN/mm²", activation "8.2 ± 4.0 s⁻¹", slow phase "75.6 ± 21.1 ms", relaxation k_REL −19.7 ± 9.6 s⁻¹ (w1118) | Mendeley 10.17632/fvx7g8g3jp.1, CC BY 4.0 | partial: chemically activated myofibrils, no twitch, no actuator | text |
| Azevedo et al. 2020, eLife, doi:10.7554/eLife.56754 | 5 muscles (#19 tibia) | Tibia flexor motor neurons: slow <0.1 µN per spike, fast about 10 µN | Dryad 10.5061/dryad.76hdr7stb, CC0, 48.8 GB | partial for tibia rows; not the TTM | text (agent) |
| Matsuo et al. 2014, Front Physiol 5:179, doi:10.3389/fphys.2014.00179 | 6 JO | Females, GCaMP3, 40–800 Hz: zone D prefers "middle-range frequencies around 100–200 Hz" (N = 7); zones A and B peak at high and low frequencies | no deposit | partial: calcium, female, figures only | text |
| Ishikawa et al. 2017; Patella & Wilson 2018 | 6 JO | JO-A by driver line; pan-JON best frequencies 4–600 Hz, types shared across zones | none / on request | partial | text (agent) |
| Baker et al. 2022, Curr Biol, doi:10.1016/j.cub.2022.06.019 | 6 M4 candidate | 24 WED/VLP types found by a split-GAL4 calcium screen, then identified in FlyWire; sine, pulse, IPI and carrier tuning in virgin females | Princeton DataSpace 10.34770/7zem-x425, CC BY-NC-ND 4.0, 3 files, about 2.3 MB; not opened | identity audit: 12/28 complete (AVLP_pr36 excluded), 14 partial, 2 absent; still needs sex/input/readout evidence and a prospective partition/lock | text, source metadata and CAVE v783 identity audit (2026-10-17) |
| Deutsch et al. 2019, Curr Biol, doi:10.1016/j.cub.2019.08.008 | 6 M4 candidate | dsx+ pC1, pC2l, pC2m, pMN1, pMN2 tuning to IPI and sine, both sexes | Table S1 only; raw data on request | candidate, partial | text (agent) |
What would unblock a front is now a request, not a search: LC4/LPLC2 recordings (protocol 3), the Ispizua keypoints and their licence (4), Dataset S1 of Apostolopoulou & Lin 2020 (2) and the Kohn et al. 2021 traces (2).