research trees are not clear for me, also why test is not stopping when it fails?
Benchmarks are supposed to fail. Each failed benchmark still provides you with telemetry, which you can use to upgrade, and get better runs.
They should be run continuously, until you've upgraded your CPU enough that they pass. They are not supposed to be one-shotted.
You use the telemetry gained from each run, to upgrade your components in the research tree.