Cognitive Flow Trading
AI · Futures · Real time
Real-time futures trading research

AI that has to earn its trades

Cognitive Flow Trading is a real-time futures trading platform where language models take part in live decisions. Every idea has to survive the same evidence bar before it touches real capital. Most don't.

Evidence funnel

Most ideas should die

Ideas flowing through five gates
In testRejectedReaches live
Illustration of the process, not real counts.
Principles

How the work gets done

// 01

Evaluation is the product

A good-looking backtest is a hypothesis. Success is defined before looking at the data, then tested on markets and periods the system has never seen.

// 02

Most ideas should die

The majority of strategies tested don't hold up. Finding that before capital is at risk is the job, not a setback.

// 03

The backtest lies

Live behaviour is reconciled trade by trade against what the research predicted. Any gap is treated as a bug.

// 04

AI agents are teammates

Much of the platform was built alongside AI coding agents. The skill isn't prompting them; it's checking their work.

The path to live

Five gates, in order

  1. Frozen specThe hypothesis and its pass/fail rules are written down before any data is touched.
    Entry
  2. Out-of-sample testTested on time periods and markets held back from development.
    Many stop here
  3. Statistical checkResults have to beat what chance alone would produce.
    More stop here
  4. Live parityReal fills are compared against what the simulator predicted.
    The hardest gate
  5. Guarded deploymentRuns alongside existing systems, monitored and safeguarded, before it is trusted.
    Live
What's built

Market data to execution

Decision

AI decision layer

Language-model judgement in live trading, with structured outputs, confidence controls and full decision traceability.

Runtime

Real-time engine

A high-performance signal engine, orchestration and execution, broker integration and live monitoring.

Research

Research harness

Backtesting and replay on historical market data, so ideas are tested long before they meet a live order.