Set File Discipline: What to Change, What to Leave Alone

A set file is not a suggestion. It is the specific combination of parameters that produced the results you were shown, and every change you make to it separates your account from that record. Most people who say an EA underperformed are running settings the seller never tested.

The default first move is to load the provided set file and leave it alone. This sounds obvious and almost nobody does it, because every EA has one input that looks obviously wrong — a stop that seems too wide, a filter that seems too strict, a lot rule that seems too timid. That input is usually the reason the strategy survived a period you never saw in the results you skimmed.

Position sizing is yours; it depends on your balance and what drawdown you can hold, not on the strategy's logic. Everything else is the strategy. Change the lot or the risk percentage freely, and treat the entry, exit, filter and timing parameters as fixed unless you have a reason you can state in a sentence.

If you adjust three parameters and results improve, you do not know which one helped, whether two helped and one hurt, or whether the market simply changed. You have spent weeks and acquired no information. Change one input, run it long enough to see a meaningful number of trades, and write down what happened before touching the next one.

Save the seller's set file untouched under its original name and never overwrite it. Every derived version gets its own name and a line saying what changed and why. Six months later, when something is behaving oddly, the only way to answer "what did I actually change?" is to have written it down at the time. Memory is not a record.

Symbol suffixes, contract sizes, stop levels, spread profiles and swap rates all vary between brokers. A set file tuned on one broker's XAUUSD may need a different stop distance on another's XAUUSD.m. This is not the EA failing; it is a parameter that was never broker-independent to begin with. Check the symbol specification before assuming the logic is wrong.

Running the optimiser until the equity curve looks beautiful produces a set file fitted to the exact past you tested on, and to nothing else. The more parameters you optimise and the shorter the window, the more certain this is. If you optimise at all, hold back a period the optimiser never sees and check the result there. A configuration that only works on the data that produced it is not a configuration.

Our three best-selling EAs: GOLD NEURON (AI) https://www.mql5.com/en/market/product/187329 · ATLAS PORTFOLIO https://www.mql5.com/en/market/product/182751