About

In September 2026 the male Drosophila central nervous system connectome — MaleCNS v1.0, 166,000+ neurons, from Google Research and HHMI Janelia — landed in public. Within weeks people had wired it into Doom, Mario 64, Minecraft, desktop pets, trading bots and a horror game. The demos were extraordinary. The claims around them were not always careful.

The Learning Fly indexes those projects and rates each one on a single axis: how much of the fly's biology it actually simulates. Four levels, from a connectome used as fixed wiring up to behavior validated against real flies. The scale is here, and every level published carries the evidence that earned it.

What this is not

• Not a ranking. L0 is not a failing grade. Some of the best work here is honestly, deliberately L0.
• Not a replication. Ratings are read from source and documentation — see how they are made, below. Nobody here has re-run these projects.
• Not a competitor to awesome-fly. That list is the index; this site adds the one thing a README cannot carry — a rating with evidence behind it.

Where the data comes from

Project names, repositories, authors and descriptions are synced from the community lists below, last pulled on 2026-09-15. When two lists carry the same project the first one supplies the prose and the entry records both.

• cobanov/awesome-fly — 90 projects, entries reviewed September 12, 2026 · CC0 1.0
• townie/awesome-fruit-fly — 51 projects · see repository

Ratings are not synced. They live in one file in this site's own repository, reviewable in a diff, and of 141 indexed projects 96 carry a level.

How the ratings are made

Plainly, because this site asks projects to be precise about their own methods: the ratings are produced by reading source code with an AI. Clone the repository, find the line where a weight is written, work out what changes it, and decide. Every project that runs a model has been read this way, and the consequential claims were then read a second time by an independent pass that set out to overturn them — which it did, five times.

What that is not: a fly neuroscientist's verdict, or a reproduction. Nobody here has re-run these projects or measured a real fly.

Which is exactly why the evidence rule matters. Every claim carries a link pinned to a commit rather than a branch, so it cannot quietly start pointing at different code, and a script checks that all of them still resolve before anything ships. The rating is an argument with its working shown. Judge it on the links.

A human reading that contradicts one of these outranks it, and the dispute link is on every project page. Corrections from a project's own authors go to the front of the queue.

17 repositories that appear upstream are deliberately left out, each with a written reason — token launches wearing a connectome, projects only inspired by fly biology rather than running it, duplicate landing pages, and social posts with no source to read. That file is in the repository too.

Corrections

If an entry misreads your project, say so — the form is here. Corrections from authors are given priority over everything else in the queue.