Stonkfly is an open-source simulation that points a reconstructed male fruit-fly nervous system at cryptocurrency-market data. It turns public BTC-USDC prices into an RGB candlestick chart, sends the resulting visual signals through the simulated connectome, and produces buy, sell or hold proposals. The project defaults to paper trading with a $100 simulated balance—and profitable learning has not been demonstrated.
A virtual fly brain faces a crypto chart
Stonkfly uses the MaleCNS v1.0 graph, a digital wiring model of a male fruit fly’s nervous system. That distinction matters: this is a structural simulation, not a biological fly, a recording of consciousness or proof that an insect is trading cryptocurrency.
The model’s documented scale is substantial: 166,700 simulated neurons, 25,582,938 directed connections and 124,177,617 synaptic contacts. Those figures describe the graph’s architecture. They do not turn it into a validated full replica of living physiology.
The output is also narrower than the science-fiction version of the idea. A fixed neural readout maps simulated neural activity to three possible proposals: buy, sell or hold. No large language model chooses the trades.
The market arrives as pixels
The simulation does not receive a stream of raw price numbers as its neural input. Public Coinbase prices are rendered locally as an RGB candlestick chart, then presented to simulated visual inputs.
A reported display description puts the chart at 320 × 180 pixels across the simulated fly’s visual field, with an intersecting central region. The implementation separately lists 3,335 brightness inputs and 811 R8 color inputs. Those are different descriptions of the interface, not a demonstrated one-to-one mapping between every display pixel and every neural input.
That conversion is the project’s most interesting trick. The virtual nervous system is not being handed a tidy table of prices and told to calculate a forecast; it is being given a visual representation and asked to generate an action proposal from its simulated neural activity.
Reinforcement without biological pain
Portfolio gains and losses feed back into the simulation as engineered reinforcement signals. Positive portfolio P&L, including unrealized changes and fees, stimulates 15 identified PAM11 dopamine cells. Negative P&L stimulates two PPL101 aversive dopamine cells.
Those signals are part of the software design. They are not modeled pain or pleasure, and a gain does not prove that the last proposal caused the result. The feedback mechanism gives the simulation a way to respond to outcomes; it does not by itself establish learning, understanding or market prediction.
Paper trading first
The default configuration uses real public BTC-USDC data but executes paper trades against a simulated $100 balance. That makes the standard setup a software experiment rather than evidence of financial performance.
The project also documents a path toward cryptocurrency-account operation through Coinbase Advanced, but the existence of an order pathway is a different claim from showing that the system can trade profitably. A program can generate an order proposal—or even connect to a trading interface—without having a validated strategy.
| Dimension | Stonkfly detail | Scope or condition |
| Neural model | MaleCNS v1.0 | Simulated male fruit-fly connectome graph |
| Graph size | 166,700 neurons; 25,582,938 directed connections; 124,177,617 synaptic contacts | Structural model counts |
| Market input | RGB candlestick chart | Rendered from public BTC-USDC prices |
| Trading output | Buy, sell or hold | Produced by a fixed neural readout |
| Default operation | Paper trades with a $100 simulated balance | No real-money result follows from the default mode |
| Reinforcement signals | 15 PAM11 cells and two PPL101 cells | Engineered feedback signals, not biological pain or pleasure |
Why this is not a proven trading strategy
The important result is not that a fly brain has beaten the market. It has not. Stonkfly’s documentation says profitable learning has not been demonstrated, and the project does not establish genuine trading ability, predictive skill or investment suitability.
That boundary is easy to lose when the architecture sounds so unusual. A large connectome, a colorful market chart and dopamine-linked feedback make a compelling experiment. They do not replace a reproducible evaluation of returns, risk and generalization across market conditions.
For readers, the useful takeaway is straightforward: Stonkfly is a fascinating demonstration of how a mapped nervous-system model can be connected to an application outside biology. It is not a validated crypto strategy or a reason to risk money.
Who can run the simulation?
Stonkfly supports macOS and Linux. Its documented software requirements include Python 3.11 and a C++17 compiler, with 16 GB of RAM recommended.
Those requirements frame the project as a demanding open-source experiment for technically minded readers. The real story is the bridge it builds between a connectome, visual encoding and an action interface—not a promise of easy profits.
A virtual fly can be pointed at a crypto chart. Stonkfly shows how that connection might be engineered; it still does not show that the simulated nervous system has learned to trade successfully.
