Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

I Ran AI-Trader Through a Paper-Trading Cycle. It Passed Locally, Not in the Real Market

Description and use

All local tests passed (8 price-fetching, 12 assignment/guard, 103 backend). The paper-trading cycle produced a settled trade with 2% return under mock prices, demonstrating the workflow but not real-market profitability.

Problem

Can an AI agent actually participate in a trading workflow, or does it only produce a confident-looking buy/sell sentence?

Outcome

All local tests passed (8 price-fetching, 12 assignment/guard, 103 backend). The paper-trading cycle produced a settled trade with 2% return under mock prices, demonstrating the workflow but not real-market profitability.

Workflow

  • Set mock BTC price to 100
  • Buy 2 BTC
  • Verify cash balance 800
  • Move mock price to 110
  • Check position value 1020 and return 2%
  • Sell at 110
  • Confirm empty position and cash 1020
  • Review settled record and ranking

Evidence

The local paper-trading path completed successfully. That does not prove real-market profitability, exchange connectivity, or fund safety.

Source

The test groups produced these results: Price-fetching tests: 8 passed; Experiment assignment, event records, agent recovery, and real-time price guards: 12 passed; Other local backend tests: 103 passed.

Source

  • Record date: 2026-09-09T18:30:49.824Z