The previous stats had a large bias towards excluding fails, because of
some bugs that were present: A failing run had larger chance of
triggering a bug causing a crash, leading to that run did not end up in
statistics.

The present statistics had no crashes during the whole run.