Understanding the five AI engines

The AI layer in Horse Racing Trader

HRT is built around five distinct AI engines, each designed to address a different analytical dimension of the trading environment. They use Anthropic's Claude models — primarily Claude Opus for deep analysis and Claude Haiku for continuous, low-latency monitoring.

None of these engines tell you what to trade. They provide structured, reasoned analysis of market microstructure and your own trading patterns. What you do with that analysis is entirely your decision.

1. Live Trade Thesis + Rolling Narrative

Model: Claude Opus

When it fires: When the microstructure monitoring layer detects a significant signal — an unusual liquidity event, a sharp price move, a structural pattern that matches a known trading scenario.

This is the most reactive of the engines. When conditions warrant it, it produces a Trade Thesis — a concise, structured read of what the market microstructure appears to be doing and why. Alongside the thesis, a Rolling Narrative updates as the market evolves, providing running commentary on whether the initial thesis is holding, breaking down, or being replaced by a new pattern.

Use this engine's output as a structured prompt for your own thinking. Does the thesis match what you are seeing? Does the narrative describe price action you recognise?

2. Pre-Race Brief

Model: Claude Opus

When it fires: Automatically, approximately 3 minutes before each scheduled off.

The Pre-Race Brief is a comprehensive structural snapshot of the market taken at the point when most professional pre-race traders are forming their final views. It covers seven structured sections: Race Profile, Liquidity, Volatility, Trading Environment, Watch List, Risk Flags, and Verdict.

The Verdict section provides a confidence-rated summary of the trading environment. It is not a selection — it is a characterisation of the market's structural quality for trading purposes.

3. Liquidity Interpreter

Model: Claude Haiku

When it fires: Continuously, throughout the pre-race window.

Haiku's speed makes it ideal for continuous monitoring. The Liquidity Interpreter reads the order book in real time and produces short, plain-English notes about liquidity conditions: where depth is accumulating, where it is thin, whether the book is balanced or skewed, and whether the spread is widening or tightening.

These notes update regularly and are designed to be glanceable — short observations that give you a continuous feel for market health without requiring you to interpret raw numbers.

4. Trading DNA

Model: Claude Opus (full version) / Claude Haiku (preview version)

When it fires: On demand from /trading/dna. Unlocks progressively with your trade history.

Trading DNA is a personal performance analysis engine. It reads your anonymised trade history and produces a structured account of your trading patterns — your tendencies, your strengths, the risk patterns in your behaviour, and how your performance varies across different market conditions.

• Preview unlocks after 20 trades (uses Haiku for a lighter analysis).
• Full unlocks after 50 trades (uses Opus for a comprehensive multi-section report).

The value of Trading DNA compounds over time as the sample of your trading grows. See the dedicated article for full details.

5. Value Detector

Model: Claude Opus

When it fires: On demand, from the Value Detector tab on /trading/signals. Subject to a 30-second rate limit between requests.

The Value Detector asks Opus to read the current order book for a selected runner and assess whether the price reflects the structural balance of supply and demand. It produces one of four assessments: POTENTIALLY_UNDERVALUED, POTENTIALLY_OVERVALUED, FAIRLY_PRICED, or INSUFFICIENT_SIGNAL, along with a confidence rating and structured reasoning.

This engine contains no form data, no ratings, and no external information. Its assessment is based purely on the order book at the moment of analysis. It is structural microstructure commentary, not a recommendation to back or lay the runner.