Backtesting by hand: how many trades do you collect before trusting a setup?

Manual backtesting — scrolling back, marking setups, recording results — is slow but teaches more than any automated run. The question is always sample size: twenty trades is noise, two hundred is information.

What's your standard?

• how you run a manual backtest
• the trade count you require before going live
• how you keep the test honest (no hindsight editing)

The expectancy guide tells you how to read the results.