Expert Advisor (EA)
Test the implementation as well as the idea
Check the specific platform and tester version, tick or bar modelling, spread, commission, financing, order rules and available history. Simulated intrabar paths and missing data can change results. A claimed modelling-quality percentage does not establish realistic execution.
Operational safeguards
Test rejected orders, disconnects, duplicate signals, restart state and position limits. Record the EA version and inputs so results can be reproduced. Performance-dependent sizing, including martingale or grid variants, needs explicit stress testing; no finite account can support unlimited escalation.
Unseen data remains essential
Keep final validation data separate from optimisation. Demo or small-scale forward observation can expose implementation problems, but neither a demo nor a historical track record guarantees future returns. Capital requirements cannot be inferred from a fixed multiple of the worst historical loss.
FAQ
Does a high modelling-quality score prove an EA is profitable?
No. It does not validate the rule, costs, executable path or performance on unseen data.
What should happen after an EA restarts?
It should reconcile orders and positions, restore its documented state and prevent duplicate entries before resuming.