Trying to evaluate strategy for robustness. It shows good result on one particular NumWFOCycles=10 setting but much worse on another settings like NumWFOCycles=20 or NumWFOCycles=5
Can be strategy trusted if working well only on one WFO cycle number?

I've had the same issue with several strategies...

For example, in daily timeframe i've found it works well to set one year out of sample by default:

DataSplit = 85;
WFOPeriod = 252*DataSplit/(100-DataSplit);

Ideally, a strategy should perform reasonably good with different cycles, but i guess this is something one must find out by experience. I agree with the manual that most robust cycles numbers are between 5 and 10.

This is just my experience so far, surely more expert people like jcl can shed more light into this question.

Cheers.

I see few issues here.
Even when running WFO test with same cycle number Lookback setting heavily influences Test and Train period. For example:
LookBack = 500;
StartDate = 2005;
EndDate = 2016;
NumWFOCycles = 10;

result:
WFO test cycles 9 x 178 bars (36 weeks)
Training cycles 10 x 1008 bars (208 weeks)

and when change only LookBack:
LookBack = 200;
StartDate = 2005;
EndDate = 2016;
NumWFOCycles = 10;

result totally different:
WFO test cycles 9 x 198 bars (40 weeks)
Training cycles 10 x 1124 bars (232 weeks)

That's why I was not able to find a way to calculate WFO cycle number for fixed number of days by using
NumWFOCycles = (barsTotal-barsTrain)/barsTest + 1;

Is there a way to keep same train/test period regardless of LookBack and StartDay ans EndDay settings?

Also I think to stick to number of cycles between 5 and 10 is irrelevant because it's totally depends on testing period.

I know no better way to set a "fixed" train/test period than what i stated before. Obviously LookBack and test period make a impact on WFO results, that's the same with almost any strategy you can find out there.

There'll be always some factors that will escape our control, that's why i think it's not so important to set a magical number of cycles or parameters and so on. Live trading will always surprise us.

The only rule i know about WFO is that your strategy should at least perform 50% of the ideal performance of a non-wfo test.

Again, this is only my opinion. I'd really like to read the advise of more experienced people.