A list of mistakes that new EA builders often make, collected from threads here and elsewhere. Maybe useful for someone starting.
Testing only in "every tick" mode on generated data. Use real ticks where the broker provides them.
Ignoring commission in the tester.
Hard-coding pip size, which breaks on 3-digit yen pairs and metals.
Not checking order return codes, so failed orders go unnoticed.
Optimising many parameters on a short period, then being surprised when it fails.
No logging, so a bad live trade can't be explained afterwards.
Using local time instead of server time, or forgetting daylight saving changes.
A good warm-up exercise: write an EA that does nothing but place and manage one trade with full logging and error handling. Run it on demo for a week. It's boring, but it builds the scaffolding everything else needs.
Another: rewrite the same simple strategy twice, once quickly and once carefully, and compare the logs.
One I would add is not testing with the spread that the live account actually has. For a long time I backtested with a fixed two pip spread that I had picked at random, and when I switched to the variable spread from my broker's history, a scalping EA that looked profitable lost money in the quiet Asian hours. I now export the real spreads and run the test on those hours separately.
Another one is trusting the optimiser's best result. I once picked the top pass from a few thousand runs and it was a narrow spike in the parameter space, so a small change gave a completely different outcome. These days I look for a plateau where neighbouring values also work and I take the middle of it.
Your warm-up exercise is solid. My own was a single trade EA with a log file, and the thing I found was that my code treated a requote as a success. It is a boring project, but I reuse the scaffolding for everything now.
