Step 1. Write the hypothesis before opening the tester
A useful hypothesis is specific: define market, entry, exit, period and failure condition in advance. “We will tune indicators until the chart looks good” is not a hypothesis; it is a path to overfitting.
Step 2. Separate research from validation data
Do not make decisions using the same period where parameters were selected. Keep an out-of-sample period untouched. Better still, use walk-forward validation: select settings on one window and apply them to the next.
Step 3. Include real trading costs
Commission, spread, swap, delay and contract size belong in the test, not just in the live launch checklist. A few ticks can remove an intraday edge; for D1 systems, correct rollover and symbol availability may matter more.
Step 4. Avoid a single perfect parameter
A stable idea usually survives across an area of similar parameters. If the result exists only at one exact RSI, stop size and entry time, treat it as an overfitting warning. In MT5, inspect the optimisation surface, not only the top pass.
Step 5. Translate rules into MQL5 literally
Document decision timing, price source, multi-symbol order, volume rounding and order-error behaviour. Compare the research trade log with MT5 over a short sample. Any mismatch deserves an explanation.
Step 6. Measure the path, not only the return
Record intraday drawdown, worst months, drawdown duration, instrument concentration and cost sensitivity. For a prop environment, these often matter more than net profit or a single Sharpe ratio.
| Status | Meaning |
|---|---|
| Idea | A clear hypothesis exists; it is not tested yet. |
| Research | Data and IS/OOS separation are defined. |
| MT5 validation | Logic is implemented and reconciled with research. |
| Forward | Observed on demo or controlled small risk. |
| Archive | The strategy degraded or failed validation. |
For each strategy: idea, market, timeframe, data period, cost assumptions, IS/OOS results, known risks, code version and current status. A commercial EA need not reveal its source, but its methodology and limits should be clear.
Conclusion
The advantage of MT5 research is that it can become executable. Make each article a working object: not a promise, but a transparent route from question to controlled implementation.