Begin with evidence quality, not net profit
A large net-profit figure cannot repair a weak experiment. First check the test period, symbol, timeframe, modelling method, spread, commission, swap and the number of completed trades.
A short test or tiny sample can look excellent by chance. Treat the result as a hypothesis until it covers enough trades and more than one market regime.
- Test duration and market coverage
- Completed trade count
- Modelling and tick assumptions
- Spread, commission and swap treatment
Separate balance drawdown from equity risk
Balance drawdown only changes when trades close. Equity drawdown includes open profit and loss, so a smooth balance curve can hide large floating losses.
Look for the maximum equity drawdown in money and percentage terms. Then compare it with the balance drawdown, deposit size and the strategy's stated risk settings. A wide gap deserves investigation.
- Maximum equity drawdown
- Balance-to-equity gap
- Largest loss and losing streak
- Margin level during stress
Test whether the return is repeatable
Profit factor, expectancy and recovery factor are useful only in context. A profit factor just above one may be fragile after slippage; a high recovery factor can be dominated by one lucky sequence.
Ask how much of total profit came from the best few trades, one direction, one year or one symbol. Concentration is not automatically bad, but it makes the evidence narrower than the headline suggests.
- Average win versus average loss
- Profit concentration in top trades
- Long and short contribution
- Stagnation and time to recovery
Look for execution assumptions the tester cannot guarantee
Very short trades, tight stops and large volume jumps can be vulnerable to real spread expansion, slippage and broker constraints. A historical test may calculate an order that would not have been filled at the displayed price.
Grid or martingale behaviour may appear as increasing lot sizes, clustered entries and long floating recovery sequences. Judge the exposure pattern, not the marketing label.
- Lot-size escalation
- Overlapping positions
- Holding time and stop distance
- Broker-aware volume and margin validation
Turn the report into the next experiment
A useful review ends with new tests: unseen periods, higher costs, different symbols, delayed execution and parameter sensitivity. The objective is not to prove that an EA will win; it is to discover which assumptions make the result fail.
Kito's EA Forensic Engine reads compatible Strategy Tester HTML reports locally in the browser and separates evidence confidence from the performance score.
- Out-of-sample period
- Spread and slippage stress
- Neighbouring parameter values
- Cross-symbol or cross-regime test