Algo-trading

Dyson does not contain most of its own trading logic. It imports a separate Rust workspace — algo-trading — which holds the reusable trading machinery: signals, actors, order management, backtesting, and market-data formats.

In the dyson repository it lives at external/algo-trading.

If you are new to Dyson, the single most useful thing to learn first is which repo owns what. Most confusion comes from looking for something in the wrong one.

Which repo owns what

flowchart TB
    subgraph D["dyson — how a strategy is deployed and run"]
        D1["Lifecycle, commands, health"]
        D2["MongoDB config and versioning"]
        D3["NATS transport"]
        D4["Kuru and Lighter live adapters"]
        D5["Concrete strategies and operator surfaces"]
    end
    subgraph A["algo-trading — how trading decisions are made"]
        A1["Signals and fair value"]
        A2["Actors and order intents"]
        A3["Order manager and execution traits"]
        A4["Market-data formats and replay"]
        A5["Simulator and backtesting"]
    end

    D --> A

    classDef own fill:#2e7d32,stroke:#1b5e20,color:#fff
    class D,A own

I want to change…	Go to
What a strategy quotes, and when	algo-trading — actors, strategy
A new market signal	algo-trading — valuation
How orders are tracked and diffed	algo-trading — execution
How a backtest fills orders	algo-trading — simulator
Parsing or archiving a venue feed	algo-trading — market_data*
Start, pause, drain, or shutdown behavior	dyson — crates/runtime
Talking to Kuru or Lighter live	dyson — crates/venue-adapters
Strategy configuration or deployments	dyson — crates/database
A specific strategy's own policy	dyson — strategies/<name>

The rule of thumb: algo-trading is venue-agnostic and process-agnostic. It does not know about Kubernetes, MongoDB, NATS, or Kuru. Anything that names one of those belongs in Dyson.

Why the split exists

The same trading logic has to run in three very different settings:

Setting	Data source	Execution
Live trading	NATS live packets	Real venue adapters
Historical replay	Archived market-data files	Simulator
Signal research	Archived files	None — rows out

Keeping decisions venue-agnostic is what lets one strategy run in all three without a rewrite. Dyson supplies the live setting; algo-trading supplies the decision logic and the other two settings.

Where to start

Goal	Read
Learn the vocabulary	Core concepts
Add a signal or change quoting	Signals and actors
Find where some code lives	Crate map
See how it plugs into Dyson	Runtime and strategies