Skip to main content

Technology

A hot path and a cold path

Market data is read once and then splits in two. The hot path goes straight from that data to a trading decision. The cold path records everything that arrived, and is built so it can never slow the hot path down.

Hot path
Market data to decision, nothing in the way
Cold path
Records everything, slows nothing
Research
The same code tests and trades
Stack
Rust for execution, Python for research

Two paths from one stream

Market data enters once, then the work separates. Prediction, decision, and execution stay on the direct route; recording happens independently and entirely outside it.

One incoming stream

Market data

Hot path

Prediction → decision → execution

The shortest route to action.

Cold path

An independent record of every event

Kept off the trading path entirely.

What the split protects

Speed

The hot path avoids anything whose timing can't be predicted.

Evidence

Every event that arrives is recorded, so we can reconstruct exactly what the system saw, when it saw it, and which version of which model acted on it.

Consistency

Research and live trading share the same code, so what was tested is what runs.

Research at scale

Models are trained in the cloud against years of recorded market activity, with every experiment tracked and every model versioned before it can be used.

Controls before capital

Trading controls are built before capital depends on them, and they remain active throughout the path from decision to reconciliation.

  • Position and loss limits
  • Duplicate-order prevention
  • Stale-data detection
  • Daily reconciliation
  • Automatic halt on a failed check
  • Manual kill switch

Discuss data and engineering

Contact Monolithify about market-data systems, research infrastructure, and engineering work connected to the trading platform.

Reach us at hi [at] monolithify.com or visit monolithify.com.