What this is

The Trading Fly follows one fly-brain trader, "the fly", through rounds of a market. The show tracks a single account at a time: its bankroll, its ledger, and a receipt for every decision. Seed capital, sponsor credits and referrals are tracked separately and never counted as trading gains.

@thetradingfly · X · Instagram · TikTok

Two data sources

• Replay · example run. An embedded, scripted example. Prices are invented, decisions come from two scripted rules labelled fixture, and both accounts are paper. Nothing in the replay is neural activity. It exists so the page works offline and so the replay controls and receipts can be tried without an engine.
• Local engine. A read-only loopback adapter serves one paper account, "the fly", recorded by an experimental spiking model whose connectivity comes from the FlyEM MaleCNS v1.0 connectome. The market chip says which world the prices come from: Synthetic prices (the original recorded demo, a seeded random walk) or Live reference (public Coinbase Exchange ETH-USD last-trade prices, quoted in fiat USD). A live reference price is a centralized-exchange reference, not an executable quote, and USD here is a paper ledger unit, not USDC. Funds are always paper; fills are paper fills at the next recorded price. The header chips and every receipt show the backend's own metadata (brain kind and version, biological connectivity, feed source, execution mode); that metadata is authoritative, not this text.
• Capture status. Reference prices only advance while a separate capture process (a bounded command-line run, never this page) is observing the venue. The capture chip reports what the backend says: Live updating only when that process is alive and the newest observation is within the engine's freshness limit; otherwise stopped or stale, meaning the price on screen is the last recorded observation. This page only reads; it never triggers a capture, an order or any network request beyond the loopback backend.

What the model is and is not

The engine's model is a fixed-weight spiking network driven by an encoder that maps market features onto input neurons and a decoder that reads a trade proposal from descending-neuron activity. Encoder and decoder are experimental engineering choices, the model does no learning, and the engine reports forecasting_validated: false. Nothing here claims the model understands markets, and no profitability is claimed or implied. A risk check sits between decoder and execution and can reject any proposal; the verdict, and any settlement cancellation, is printed on every receipt.

Mood on screen

The fly's expression is a presentation layer computed by this page from recorded evidence: the sign of the round's fair P&L change, a rejected proposal, or an order cancelled at settlement. It is not a readout of the network and it does not claim the model feels anything. The current automatic mood mapper is a prototype; the final editorial grammar for expressions is not decided by this code.

Spike counts and scores

Receipts show the engine's raw numbers: action, scores, spike counts, stimulus and state hashes, network id. They are shown as numbers with the engine's own note, and no intention is inferred from them.

Can you beat a fly?

Take the fly's trade and Countertrade the fly are optional links to Fomo, an external venue. They appear only while the fly has a current, explicit buy or sell proposal on a market with a hand-checked route, and only while the venue capture is live-updating. A held round is no signal, a sell is a sell, and a pending order left over from a stopped capture is old evidence, so the panel then says "Trading calls paused". Opening a link places nothing, prompts no wallet and is not treated as a trade; it is a referral link and we may earn fees.

Practice calls are two free games, stored only in this browser, with points of no monetary value. The market call is your up/down/flat direction for the instrument's price from this round to the next, compared with the fly's recorded proposal for the same interval (buy read as up, sell read as down; a held round is no call and is not scored). The account call is older and different: it scores whether the fly's account trading P&L closes the round up, down or flat, not the market. Both run on the replay only, because the engine publishes neither a next-round cutoff nor a per-round price series. No ranking, no prizes.

Inspiration and independence

The idea of fly-brain traders on a public scoreboard is inspired by the Stonkfly concept by Alex Wormuth. The Trading Fly is an independent implementation with its own code, design and data model, and is not affiliated with that project.