2 stage OOS approach is it good idea?

Hello,
Recently I switched to what I call 2 stage OOS test:
Here is an Example: For H1
I’m running initial random generation for period 1.1.2011 – 31.12.2017
where OOS is  lets say 2015-2017

After I have enough strategies generated in Databank, I will retest with second OOS test:
which will look at 2004-2011 as OOS as well as 1.12.2017 – 2018…
and if I get good results for those OOS periods i would consider these strategies for further review.

Is this approach would be viable? Or strategy will have a good chance to degrade after 2018 since my in sample period was only 2015?

have any one tried this approach? Any suggestions?

It is a good approach for reason to view that last years performance in more detail. Otherwise just for robustness purpose it makes no difference since it would pass or fail what ever it is called the main thing is that it is OS and I suppose the longer OS the better. But since SQ is going thru so many strategies there is will always be a huge amount that will pass all OS and in order to find out if it is robust or not the other tools have to be used.