The SpreadAtlas lab

An open notebook of six weeks spent trying to make automated crypto trading pay. It did not. Here is everything we measured, including the parts that embarrassed us.

• Six weeks of letting AI build a trading system

  The whole story in order, with the numbers: −44% in the first week, ₩25.2 million of paper losses across 918 exits, and a rule-perfect server that still lost.

• Fees eat the edge

  26,154 indicator setups by timeframe. Only 2.5% of five-minute setups made money after costs.

• Our backtest said 75% a year. Survivorship bias said 41%.

  What changes when you pick coins the way you could have picked them at the time.

• How long does it take to prove a strategy works?

  About 34 months of live trades for a typical trend rule. Two weeks prove nothing either way.

• Two weeks of a rule-perfect trading server

  The full ledger: five trades, five losses, −3.93%, and what the audit found.

• Strategy graveyard

  Every approach we tested, how we tested it, and why it failed.

Trading research is paused until better AI models are available. When it restarts, results will be published here first, good or bad.