I saw when running system testes that somehow it is not using the latest merged version.

On my machine I created custom repo with my changes and I pulled dependencies from ovirt repo for this specific version. I run the tests several times to make sure that my changes are not breaking anything. In between ovirt repo was updated with newest build which caused lago to pull it instead of my changes.

Both scenarios make lago/system tests unpredictable. I am do not know what I am testing and this leads to wrong assumptions.

I'm not sure what you mean, how the tests are unpredictable?
You can choose which suite you want to run (3.6/4.0/master) basic / hosted-engine - what exactly is unpredictable? can you elaborate?

Due to this issue I wasted time running lago on my host. We saw issues using system tests and this makes the tests unpredictable and without it in my opinion there is no value in running those tests since we do not know what is tested.

I think this is more of am ovirt-system-tests issue then a Lago issue, so not appropriate for Github.

please open in Lago github project if still relevant.
https://github.com/lago-project/lago/

You already have an option to specify a custom repo to use instead of the ones defined in the conf files.
do we have this feature documented on Lago docs?