05 / AI PROJECT STARTER

Give Codex or Claude the data context directly.

This project starter turns the full backtesting guide into an actionable prompt. It tells an AI tool how to inspect files, join markets, simulate fills, avoid look-ahead bias, and build a reproducible first version.

How to use it

  1. Copy or download the Markdown file.
  2. Open Codex, Claude Code, or another coding tool.
  3. Paste the full file and provide your local data directory.
  4. Ask the AI to inventory and validate the data before writing a strategy.

What the prompt includes

Data boundaries

It separates order-book snapshots from trade-by-trade data and states what the dataset can and cannot prove.

Files and schema

It covers live, hourly, and daily layers plus the fields and joins for ticks, depth, outcomes, and seen_markets.

Fill simulation

It defines ASK-side buys and BID-side sells, with explicit latency, partial fills, fees, and slippage.

Quality and leakage checks

It requires checks for window boundaries, gaps, malformed books, outcome labels, and look-ahead bias.

Provide these alongside it

For a smooth start, also provide the data directory, a small date range, the strategy you want to test, and your preferred Python version.

The website provides real 5m and 15m samples. The complete historical archive is free on Telegram; use the samples to validate inventory and schema first.

If you do not have a strategy yet, ask the AI to build only the inventory, schema checks, fill simulator, and baseline report first.

Download the free samples →Read the backtesting assumptions →