Open the lab
  1. Field
  2. Line-up
  3. Raster
  4. Circuit
  5. Strands
  6. One cell
  7. Dust

Zebrafish fry · 6 dpf · two-photon · live

Put the market
under a microscope.

A zebrafish fry, the larva of Danio rerio, is 4 mm long with a see-through head, so you can watch its neurons light up. FRYBRAIN models eleven of them, one per asset on Robinhood Chain, firing on real market data. Fast cells buy. Silent cells sell. Paper only.

Open the labor scroll and watch it change shape

01 · Line-up

Eleven cells.
One per asset.

Each asset is a real traced neuron, shown where it sits in a 6-day-old fish: ETH is the Mauthner cell, the fastest escape neuron; SPY is in the optic tectum, which maps the whole visual field. The pairing of asset and region is our metaphor, matched on what each region does; the neurons and positions are data.

neurons: Kunst et al., Neuron 2019 and Hildebrand et al., Nature 2017, via NeuroMorpho.org · brain imaging: Ahrens et al., Nat. Methods 2013

brightest now — · quietest —

02 · Spikes

Every tick
is a spike.

Swap flow, price moves and news become current. Cross the threshold and the cell fires.

— spikes in the last 6 s of model time

03 · Circuit

Fire together,
wire together.

Assets that light up in the same cycle grow synapses. The arcs are the wiring it has actually learned.

strongest pair —

04 · Strands

Four books,
one honest test.

The strands are four real paper books: the circuit, the same circuit blind, a coin flip, and plain hold.

Fry —Dark —Random —Hold —

05 · One cell

Read, inject,
fire, learn.

  1. Read on-chain flow, the tape and the news
  2. Inject it as current into eleven cells
  3. Fire six seconds of model time, every minute
  4. Learn from wiring and from outcomes
See it live in the lab

06 · The math

Five equations,
all published.

Larval zebrafish weigh evidence for seconds before they commit to a move, and a leaky integrator explains how (Bahl & Engert, Nat. Neurosci. 23:94, 2020). FRYBRAIN runs the same kind of unit on market evidence. Below are the exact rules, each from the literature.

  1. Membrane leaky integrate-and-fire

    τm dV/dt = −(V − Vrest) + μi + σ√(2τm) ξ(t)

    V ≥ Vth → spike, V ← Vreset, 2 ms refractory

    τm 20 ms (larval tectum: 21 ± 7 ms) · Vrest −67 mV (larval neurons: −67 ± 5 mV) · Vth −40.6 mV = rest + rheobase × input resistance · σ 6.3 mV

    Lapicque 1907; Gerstner et al., Neuronal Dynamics, CUP 2014; Grama & Engert, Front. Neural Circuits 2012; Hamling, Harmon & Schoppik, eNeuro 10(6), 2023, Table 1

  2. Input the market as current

    μi = μ0 + μg di(1 + 0.8 Li)

    di = 0.45 flow + 0.35 tape + 0.20 news

    spike of cell j → Vk += Wjk − winh for every other cell k

    μ0 19.2 mV (below threshold) · μg 14.8 mV · winh 0.62 mV

    Brunel, J. Comput. Neurosci. 8:183 (2000)

  3. Wiring Hebbian plasticity

    Wjk ← (1 − λ) Wjk + η rjrk

    r = clip((ν − 6 Hz) / 20 Hz, 0, 1.5)

    η 0.088 mV · λ 0.015 per cycle · W capped at 2.8 mV

    Hebb 1949; Gerstner & Kistler, Biol. Cybern. 87:404 (2002)

  4. Outcomes three-factor rule

    Li ← clip(Li + ηL tanh(R / 1%), −1, 1)

    R = 15 min return after a buy · ηL 0.35 · 6 h half-life

    Frémaux & Gerstner, Front. Neural Circuits 9:85 (2016), simplified to rates

  5. Readout calcium glow

    ΔF/F(t) = Σspikes A(e−t/τCa − e−t/τrise)

    GCaMP6s, one spike: peak 23 % at 179 ms, half-decay 550 ms · τCa 794 ms, τrise 66 ms · saturates at 1680 %

    Chen et al., Nature 499:295 (2013), Supplementary Tables 1 and 3; Vogelstein et al., J. Neurophysiol. 104:3691 (2010)

  6. Measured: membrane time constant, resting potential, threshold (estimated from rheobase and input resistance), GCaMP6s kinetics. Chosen by us: drive gains, noise, inhibition, wiring and learning rates, tuned so a quiet market sits just below threshold. Whether any of it says something about markets is ours to test: that is what the four books are for.

07 · Dust

A model of a fish,
not a fish.

The shapes and positions are traced neurons from published larval zebrafish data; the activity is a simulation of eleven point neurons with computed glow. No animal is recorded. Paper only: no wallet, no real orders, no user funds. Not affiliated with Robinhood or Uniswap. Not investment advice.

Open the lab