We dropped a simulated fruit fly brain onto a randomly chosen criminology paper and let it walk outward through reference lists. Open-access papers are corridors. Paywalled papers are walls. It explores until it runs out of things it can read.
The wiring is the real thing: MaleCNS v1.0, the complete connectome of an adult male Drosophila central nervous system — about 166,700 neurons, published September 2026 by HHMI Janelia, the University of Cambridge, the MRC LMB, and Google Research, released CC-BY. The citation graph is real too, from OpenAlex (CC0). The paywalls are extremely real.
The fly is not smart. It is not learning. It is a reactive random walker whose turns are decided by spikes propagating through that wiring. That is the point: even a bug can explore an open literature, and even a bug is stopped by a paywall.
The corpus is the Web of Science research area Criminology & Penology, adopted as-is. European Journal of Criminology and Crime Science are not in it. That is Clarivate's boundary, not ours. The authoritative version of that list is available only to Web of Science subscribers: the conventional definition of criminology is itself paywalled.
| OpenAlex status | Here | Why |
|---|---|---|
| diamond | corridor | fully OA journal, no author charges |
| gold | corridor | OA journal indexed in DOAJ |
| green | corridor | free copy in a repository |
| hybrid | gate | open license inside a toll journal |
| bronze | trapdoor — impassable | free today at the publisher's discretion, no open license |
| closed | wall | not free to read |
Excluding bronze is stricter than OpenAlex's own is_oa. Because that choice makes
paywalls look worse, the bronze-inclusive number is published too.
Most of the fly-brain demos going around do not run a control. We did: the whole pipeline was rebuilt on degree-preserving shuffles of the same connectome, and we compared how well the readout identifies which input channel was driven. One shuffle would be an anecdote, so we run several and compare against their spread.
An earlier run of this pipeline reported that 99% of seed papers were paywalled. That was wrong; the figure is 75%. Twenty-five of 100 seeds were open, but 24 of those hit the crawl's compute cap and were then excluded from the headline, leaving 75 blocked and 1 open. Hitting a compute cap says nothing about whether a paper was readable. Seed-level statistics now use every seed, and a regression test fails if that ever changes back.
Code, data, the full trace, and the negative control: github.com/crimconsortium/fruitfly. Code MIT, our data and figures CC-BY 4.0. Connectome CC-BY (HHMI Janelia / Cambridge / MRC LMB / Google Research). Metadata from OpenAlex, CC0.