I think that they have no other option than to assume that human observations are at least reliable enough to work with, even if sometimes they go a little craz

I think that they have no other option than to assume that human observations are at least reliable enough to work with, even if sometimes they go a little crazy. I think that's actually why they focus on reproducibility - that's their tool for reaching something approximating a consensus reality in the face of experiential subjectivity, human unreliability, and the strong limits on individual human knowledge. Me, I think that reproducibility is really useful, but I'm not so willing to say that if it's not reliably reproducible it must be coincidence or delusion.