Reconstructed anatomy
Versioned cell identities, selected directed neuron pairs and anatomical contact counts.
Does not determine conductance or receptor effects.
Research software · working paper 01
Inside the model.
Inside the model.
Beyond the wiring.
An inspectable platform for connectome-constrained fly simulation: its anatomy, explicit assumptions, reproducible software checks, negative results and the experiments still needed to validate it.
01 / Abstract
What the connectome tells us.
And what it does not.
Synapse-resolution wiring constrains which cells can interact. It does not specify their physiology, behavior, drug response or subjective experience. NeuroCause makes that distinction inspectable rather than hiding it inside a simulation.
The current running model joins a selected female FAFB/FlyWire brain graph to a selected male MaleCNS nerve-cord graph through an explicitly modelled population-rate interface. Its body, sensory encoders and neural dynamics are model components. The paper documents the sources and choices, the checks that pass, the phenomena that do not reproduce and the empirical gates for stronger claims.
Running brain7,270selected neurons
Nerve cord1,045selected neurons
Development explorers5not all in public v2.4.0
External validationPendingno animal replacement claim
Measured wiring is a constraint.
Function is a model to test.
Function is a model to test.
02 / Architecture
One explicit bridge.
No invented cross-animal synapses.
The active closed loop uses two biological specimens. Published neuron identities and contact counts are kept distinct from the engineered bridge and virtual body. Selecting another anatomy explorer does not silently substitute it into the running circuit.
Female BANC and male MaleCNS each contain brain and cord from one animal, but the active runtime does not yet use a single-specimen brain-to-body model. MANC is a different male individual. The optic-lobe v1.1 explorer is an older overlapping release of the same MaleCNS male; its connections are not added to the newer graph.
Three evidential layers.
Do not collapse them.
Do not collapse them.
03 / Evidence
Anatomy, simulation, experience.
The paper distinguishes published reconstructions from computed states and from the unresolved question of experience. A numerical output is not automatically a biological measurement.
Simulated dynamics
Seeded membrane states, spikes, behavior and responses to interventions.
Not yet calibrated to withheld recordings or another animal.Subjective experience
No direct observable in the current software establishes feeling or pain.
Digital sentience remains a research question, not a result.Internal software tests passed for specimen identity, graph validity, reproducibility, sensory routing and exports. The documented model does not reproduce habituation, associative learning or thermal preference in the cited protocols. These failures are reported rather than hidden. See the benchmark evidence and the full paper for methods and references.
Scientific claims need
corresponding tests.
corresponding tests.
04 / Claim boundaries
A research prototype.
Not a replacement animal.
Whole animal
The running brain and cord come from different animals. The virtual eye, muscles and body were not measured from either reconstructed individual.
Pharmacology
Generic gain or blockade controls are circuit interventions. Predicting a named compound requires targets, concentration-response curves, uptake, metabolism and blinded validation.
Sentience
NeuroCause does not report measured feelings, pain or a sentience score. Whether a digital system could have subjective experience is an open research and ethics question.
The road from a working
model to a tested one.
model to a tested one.
05 / Prospective tests
What evidence must come next?
- 01One specimen, one native graph.
Audit brain-to-cord coverage, missing edges and version differences before a same-animal runtime is claimed.
- 02Calibrate physiology.
Fit identifiable sensory, cellular and motor parameters against independent recordings, with uncertainty.
- 03Predict held-out data.
Freeze code and parameters, preregister endpoints and compare with simpler baselines in blinded neural and behavioral experiments.
- 04Assess sentience separately.
Define competing theories and discriminating tests with independent ethical review; prediction alone cannot establish experience.
Read · scrutinize · cite.
06 / Full paper
The methods, in full.
The five-page PDF includes the dataset table, architecture, internal results, limitations, prospective validation criteria and linked primary literature. It is a public working paper, not a peer-reviewed publication.
Suggested citation
NeuroCause Research Software. NeuroCause: an inspectable platform for connectome-constrained fly simulation. Methods and validation-status working paper, version 0.3 (28 September 2026). neuro-cause.com/paper.html.
PDF integrity · SHA-256 f8fc4e69a70c7c74d70726b3f1e26b9aa9c8f7228af0c9f2a1bafcb45e3fb143
The public v2.4.0 source snapshot does not yet include every exploratory anatomy bundle discussed in this manuscript. Independent reproduction of those newer counts requires a subsequent public importer and asset release. Publisher data retain their own licences; the paper does not imply endorsement by the source institutions.