Walk-forward evaluation as operator rehearsal

A walk-forward run completes the evaluation

A system evaluation is treated as incomplete if it shows only in-sample historical results and omits a substantial walk-forward run on previously unused data. In-sample testing is development and tuning on historical data the designer has already seen. Walk-forward evaluation applies a finished rule set to data withheld from design and books every entry, exit and abstention under stated costs, size limits and session hours.

Design used a six-month historical window. The same rules were then left unchanged and logged forward. That out-of-sample testing checks whether the market concept still produced a usable signal stream.

Session-close logging and stated constraints

The worked walk-forward booked signals after each weekday cash session closed, between 0930 and 1600 Eastern, under pre-set commission, slippage, contract-size and starting-balance constraints. Session-close logging records each day's signals only after the cash session ends so the evaluation uses completed bars rather than unfinished prices.

Robustness testing of path and funding

Robustness testing stresses the same procedure across holdouts, cost assumptions, starting-capital levels and adverse equity paths to see whether it remains intact. The robustness review asks whether an operator who began after an equity peak could continue through an 18.7 percent decline lasting 57 trades, rather than judging the path only by a later new high.

The same decline is shown as a 36.3 percent loss if the futures margin account started at 10000 instead of at the prior equity high, so funding level is part of the evaluation.

One robustness recipe freezes the original build parameters and withholds the first 15 percent of the series so the unused remainder can serve as an out-of-sample stress test.

Process-fitness and override

Process-fitness is whether the operator can take every signal without override through a sub-50-percent win rate, long losing streaks and multi-week drawdowns. The evaluation includes whether a near-45-percent win rate and as many as nine consecutive losses would cause the operator to abandon or rewrite signals.

If the operator cannot take every signal that conflicts with prior market beliefs, the evaluation treats delegation to a third party who will execute without override as a process choice.

Hours versus months for a similar conclusion

An automated walk-forward optimizer is described as reaching a similar viability conclusion in hours that a manual, session-by-session log took more than seven months to produce.