From 125 million to 311.8 million:
what the full dataset actually contains.

After downloading the data, the first question is often its scope rather than its algorithm.

One connectome, different counting scopes.

Google Research reports more than 166,000 neurons and about 125 million connections. The MaleCNS v1.0 syn-partners bulk table contains 311,833,243 rows, including many segmentation fragments.

The site therefore labels synapse records and body IDs including fragments separately. They describe the input table, not the paper's neuroscience counts.

Give every record a reproducible encoding.

Endpoint body IDs and coordinates become fixed 40-byte little-endian records in original order. The canonical file is 12,473,329,720 bytes, split into 4,096-record Merkle leaves.

body_pre   u64     body_post  u64
x_pre      u32     x_post     u32
y_pre      u32     y_post     u32
z_pre      u32     z_post     u32

Total: 40 bytes / record

Check the source and the conversion.

The 6,777,179,098-byte source matches the official object MD5. Python and Rust agree on the full-file root. First, middle and final records were also compared with Arrow rows.

Reproducible computation starts with a reproducible interpretation of data.

These checks establish data identity, not biological model validity. The project computes on a static connectome; it does not reproduce fly behavior.

Inspect a real synapse.

In the workstation, record 155916621 shows body 516559 → 29742 and its coordinates directly from the loaded data.

Source: HHMI Janelia official download page. Data is licensed CC-BY 4.0.