AI & Automated Trading
What a model can and cannot do with a chart, why its backtest lies, and what changed when brokers opened their doors to agents.
Placed after evidence because half of this track is applied statistics wearing a new coat — the reason an AI backtest overstates itself is the reason any backtest does, plus one failure mode that is genuinely new. It is also the only track whose subject includes the product teaching it, so every claim here is checkable against code in this repository.
- 1What a model actually does with a chart
Separate describing an image from measuring one.
Checkpoint: Spot the trap - 2The look-ahead problem lives in the weights
Understand why an AI backtest can be contaminated before you write a line of it.
Checkpoint: Sequence - 3MCP, and what a broker opening one means
Read the agentic-trading announcements without the marketing.
Checkpoint: Sequence - 4Read-only agents versus execution agentsnot written yet
Price the difference between an agent that suggests and one that fills.
Checkpoint: Call the grade - 5Why asking a model for a setup failsnot written yet
Diagnose the failure in the most common prompt in retail trading.
Checkpoint: Spot the trap - 6Backtesting an agent without fooling yourselfnot written yet
Design a test an agent cannot pass by remembering.
Checkpoint: Sequence - 7Evaluating an AI trading toolnot written yet
Six questions that break most of them, including ours.
Checkpoint: Sequence - 8Where AI genuinely helpsnot written yet
Find the tasks where a model beats you, and notice entries are not among them.
Checkpoint: Mark the chart
Complete all 3 written lessons in this track with an average checkpoint score of 70 or better and you can generate a shareable record of completion. You are at 0/3.
The concepts here are applied in the Trade Playbook, and the engine that grades them is taken apart in how AI chart analysis works.