A mouse brain in a jar, trading stock tokens on Robinhood Chain. Its neurons, their shapes and every connection between them are real, taken from one mouse's visual cortex as mapped in MICrONS, Nature 2025.
Opening the jar.
72 of the neurons, one per row. Every dot is a spike. A cursor sweeps left to right, writing each new tenth of a second over the oldest column. The top rows are voters, and they turn orange while their stock is being heard.
$10,000.00
Paper money. It started with $10,000.
| stock | shares | value | since bought |
|---|---|---|---|
| Nothing held yet. | |||
| stock | price | wiring | full rise | full fall | last call |
|---|
Wiring counts synapses from a stock's input cells to its output cells, directly and through one excitatory cell. Full rise and full fall are how many times the usual output a 0.5% move produced when we ran the simulation for a minute each.
Real. 1,627 neurons from the MICrONS cubic millimetre of mouse visual cortex: their reconstructed shapes, cell types, positions, and the 52,181 connections between them. The brain around them is the adult mouse outline from the DevCCF atlas. Prices are Chainlink feeds on Robinhood Chain.
Ours. The neuron model: leaky integrate-and-fire, with constants from a whole-brain fly model, and each connection made stronger than life because we only see a few dozen of each cell's thousands of inputs. Which cells hear prices, which ones vote, and the rule. The cube's place in the brain, which is approximate: the two datasets come from different mice and are not registered to each other. And the money, which is paper. There is no wallet and no real order.