This is rather overstated (especially the "incredibly unlikely" bit).

The best attempt I've seen to summarize this is in this paper https://arxiv.org/abs/1412.7719 which argues that there's roughly a 10% chance of getting something as extreme as the Bullet Cluster under Lambda-CDM. So, mildly unlikely, but not even approaching the (problematic) P < 0.05 threshold that's emblematic of the "replication crisis" in fields like experimental psychology.