BioReservoir simulates a real, published brain instead of training a language model. The wiring is a male fruit fly's whole central nervous system, mapped neuron by neuron by Janelia FlyEM (MaleCNS v1.0, CC BY 4.0) — 165,122 neurons, 6.2 million connections. A small text-embedding model turns your question into firing rates on real sensory neurons (taste, smell, sight, touch); a leaky integrate-and-fire simulation, ported from Shiu, Sterne, Spiller et al., Nature 634 (2024), runs the whole brain for 250 simulated milliseconds; whichever side reaches the descending motor-command neurons harder becomes yes or no. No part of that pipeline is trained on outcomes.
Two checks anchor the honesty claim. What worked: driving the real sugar-sensing neurons fires the real proboscis-extension neuron (MN9) at rates matching Shiu et al.'s published result, and bitter input suppresses that firing by 96.8%. What failed: the same lateral-bias readout, run on a second, female connectome (BANC, Harvard Dataverse, CC BY 4.0), collapses to a constant response regardless of which side is stimulated — a real limit of that dataset's public reconstruction, published as a failure, not left out. Every answer is scored against a rewired connectome, a random graph, a no-connectome baseline — not against nothing.
This runs live, not from a cache: every question above triggers a fresh whole-brain simulation on IG Digital Lab's own CPU servers in Sacramento, California, and takes about 66 seconds end to end — /api/now reports the live queue and cycle time. Snapshot from /api/stats as of 2026-09-22: 8 questions answered, 4.3 million real spikes recorded, 6 seconds of brain time simulated since the page went up on 2026-09-19 — the live counters above update as new questions come in.
This is an experiment in brain wiring, not a claim about anything else. The full methodology, the controls and what the numbers actually say are on the data page; sources, images and contact are in the press kit. Built and operated by IG Digital Lab.