# Fly Neuron Atlas

An evidence and data snapshot of completed analyses of fly neural connectivity,
activity, and motor function, selecting reusable material from 21 research bundles
through 2026-09-16.
The detailed findings are in Japanese; table headers and runnable tools are in English.

The September 16 update adds [portable circuit operations](circuits/README.md)
and five normalized input sets, plus [research findings and limits](research/2026-09-16/README.md)
on experience-dependent heading representations and their possible use in an
[artificial subject with continuing experience](docs/research-purpose.md).
The biological findings and proposed artificial comparisons remain separate.

The September 15 implementation revision rejects missing or inconsistent analyses
and isolates voltage processing at clock resets. All 23 raw records were rerun.
The three principal bilateral DNa02 records retain their original numerical
results; two clock-reset records have documented changes. See the
[remediation record](docs/audit-remediation.md), [saved-result comparator](docs/ephys-comparison.md),
and [sensitivity generator](docs/sensitivity.md). `data/activity` remains the
original publication baseline; versioned corrected results are in `data/reanalysis`.

## Included results

- An atlas of all **815 motor-annotated neurons** in MaleCNS v1.0, with 330,888 direct incoming edges (total weight 3,224,310).
- **2,155 shared upstream candidates**, defined relative to 166 selected motor neurons in 42 muscle groups, with individual edge weight ≥10 and at least two distinct target groups.
- All direct incoming and outgoing edges for **40 selected candidates**: 39,909 incoming and 323,398 outgoing edges.
- Reanalysis tables for **23 DNa01/DNa02 voltage recordings, 33 channels**, representing 53,855 seconds of distributed voltage data. Technical eligibility after exclusions is 51,816.18 seconds; this is not a physiological quality certification.
- Documented navigation and experience-dependent response analyses, corrections, negative results, and explicit unresolved questions.
- Static extraction, boundary preservation, reconnection, cloning and rewiring for five selected sets of 6, 41, 131, 220 and 1,829 cells. These operations do not simulate firing, muscles or learning.
- A decomposition of heading-alignment statistics, intervention and context-switching evidence, and falsifiable proposals for using experience across artificial tasks. Improved artificial-subject performance has not been demonstrated.

## Quick start

```sh
python scripts/verify.py
python scripts/reproduce.py
```

Python 3.10+; standard library only. Commands are read-only, emit JSON, and exit 0
on agreement or 1 on inconsistency. Reaggregation checks edge counts, per-cell
totals, the exact shared-candidate set, threshold sensitivity, and annotation
exclusion. Saved activity correlations are read from tables by this command;
they are not recomputed from raw voltage.

For an optional voltage rerun, see [reproduction instructions](docs/reproducing.md).
Raw voltage files are fetched separately by immutable Dataverse file IDs and
checked against recorded size and MD5. They are not bundled here.

For circuit input verification and a reproducible extraction/reconnection example,
see [the circuit guide](circuits/README.md). The input sets are included; the large
collection of generated configurations is regenerated into a new output directory.

## Reading the evidence

Start with the [completed-work catalog](docs/coverage.md),
[body and activity findings](docs/findings-body-and-activity.md),
[navigation and learning findings](docs/findings-navigation-and-learning.md),
[data dictionary](docs/data-dictionary.md), and [limitations](docs/limitations.md).

An anatomical connection is not a firing event, intervention effect, or muscle
force. MaleCNS IDs identify a reconstructed male specimen; physiological data
come from other specimens, often females. Similar type names do not establish
physical identity. Unobserved edges remain unknown. This project does not claim
a complete functional decomposition of the fly nervous system, a homologous
cerebrum/cerebellum split, or demonstrated general learning in the selected cells.

Original software is MIT; original documentation and curation contributions are
CC BY 4.0. Upstream terms and credits remain applicable. See
[third-party notices](THIRD_PARTY_NOTICES.md), [provenance](data/provenance.json),
and [citation metadata](CITATION.cff). Cite the underlying datasets and papers as
well as this curation.