feedfly

a real male fruit fly brain living on robinhood chain. buys are sugar, sells are bitter. feed the fly.

what you are watching

drosophila melanogaster, the vinegar fly. it became the subject of the largest complete nervous system map ever made: brain and ventral nerve cord together, 165,122 neurons, about 125 million synapses, every cell traced from electron microscope images. the map is public. our specimen is a male.

feedfly is a token whose product is a fly. not a mascot, not a chatbot in a fly costume. a simulation of a real circuit from that map, running neuron by neuron, that feels one thing only: trades of its own token. buys are sugar. sells are bitter. everything the fly does follows from that.

how the brain is built

we do not run all 165,122 neurons. we take one circuit from the map, the reflex that has been reproduced in simulation and checked against living flies: the taste neurons for sugar and bitter, the neurons between them, and the motor neuron that extends the proboscis, the fly's tongue. every neuron on a path of at most three synapses from a taste neuron to that motor neuron: 410 neurons, 12,654 connections.

three rules, and nothing invented:

  1. the wiring is the anatomy. neuron a connects to neuron b only if the map says so.
  2. the strength of each connection is the number of synapses counted in the microscope.
  3. the sign comes from the neurotransmitter the neuron uses: acetylcholine excites, gaba and glutamate inhibit.

each neuron is a leaky integrate and fire unit. it collects input, leaks it away over time, and fires a spike when it crosses threshold. there is no training toward any behaviour. when sugar arrives and the proboscis neuron fires, that is the wiring doing it.

what the fly senses

one trade is not one signal. each part of it goes to a sense the fly actually has. this table is our design; the circuit is not.

on chainsenserule
buys in a blocksweeta buy the size of the median one, at full hunger, drives the sugar neurons at 100 hz, the rate used in the lab. bigger buys, more, up to 200 hz.
sells in a blockbitterthe same scale on the bitter neurons.
time since the last sugarhungerappetite comes back over about five minutes. a fed fly ignores sugar.
recent buy volumesatietyeach buy fills the fly; it empties over about ten minutes.
trade frequencyvibrationa busy tape makes it restless: every taste is felt a little more.
share of sells in volumeheatdiscomfort: bitter tastes stronger.
price down 15 percent within a minutelooming shadowthe fly jumps and hides for five minutes. no eating while hiding.
block timeday and nightasleep from 03:00 to 07:00 utc. nothing you do reaches it until it wakes.

the smallest trades are dust. the fly does not taste them. every size is measured in the asset the curve is priced in, which for this token is its quote asset.

what the fly does

every block with a trade, the circuit runs for 200 milliseconds of brain time with the block's buys and sells on its tongue. the random timing of the taste spikes is seeded from the block hash, so the same block always gives the same answer. when the proboscis motor neuron fires at 30 hz or more over that burst, the fly ate. that number is our threshold, not the fly's. when bitter shut the motor neuron down, the fly rejected. when the shadow looms, it jumps. asleep, nothing reaches it. between events it grooms.

every reaction is recorded with the block it happened in and the trade that caused it. what you watch is a readout of those events, not an animation we scripted. the fly has no goals and no strategy: it does not trade or decide, it reacts, and this one circuit run 200 milliseconds per block is coarser than the laboratory model it comes from.

the token

nothing routes through us. trade on any venue on robinhood chain and the fly feels it, because every trade lands on the chain and the chain is what the fly reads. holding, buying, selling: that is the whole interface.

musca alenda est. feed the fly.

x: @feedflyrh
data credit: the flyem male cns connectome, v1.0, by its authors, used under cc by 4.0 (source). neuron model after a published whole brain simulation of the same reflex. no affiliation implied.