Mare Nostrum

A price line and its moving average on a dark grid

A Rule-Based Paper Trading Project in Claude

This is part 1 of a short series about a small project I run in Claude: a rule-based, paper-first watch over a handful of US ETFs. I turned it into a starter pack so a friend can copy it.

This is part 1 of a short series about a small project I run in Claude: a rule-based, paper-first watch over a handful of US ETFs. I turned it into a starter pack so a friend can copy it.

Not financial advice. This is a research and paper-trading project. Any numbers here are examples from past data, not promises, and the rules come with no guarantee. Paper trade first and decide for yourself.

What it does

The project watches a fixed list of US ETFs, computes signals from daily bars, keeps paper-trading books, and reports every weekday. That is all. It is a checker and a bookkeeper, not a trader.

The most important rule is the first one: it never places orders. No task, skill or script places, changes or cancels an order, on paper or live. If a signal says buy, I place the order by hand in my broker app and then tell Claude about the fill so it can be logged and checked against the rules.

The rules I keep

  • Report only. Claude reads and reports. I enter every order myself.
  • No order without my explicit words. “Looks safe” or “it is only paper” is not approval.
  • Paper first. Several weeks on paper with every signal logged and no rule breaks before any live step, and any live trial starts small.
  • Never fill numbers from memory. If a data call fails, the report says so on the first line and leaves the affected rows alone.
  • Check every backtest against the settings used. Costs, slippage, balance, position size, date range and symbols. If the tool ignored costs, add them by hand.
  • One connector environment at a time. The broker connector serves paper or live, never both. The daily tasks run on paper.

The pieces

The starter pack has six parts: project instructions, two scheduled tasks, three skills, the strategy rule files, a spec for a dashboard I call the Desk, and a few small Python helpers that use only the standard library.

In the rest of the series:

  1. This overview.
  2. The two daily tasks: Close Loop and Open Check.
  3. The stock books: SWE breakout and TREND-HOLD.
  4. The optional option-spread books, and why they are off by default.
  5. How to set it up yourself.

The starter pack is a public GitHub repo: ierturk/trading-project-template.