UAH calculates anomalies for 27 regions, of which some are subsets of a larger region.
Their dataset goes back 36.3 years. The period starting in December 1978, going to March 2015, consists of 11,772 data points.
All those data points get changed in the new dataset, v6.0 beta, from version 5.6.
Of those changes, 29 have a value greater than 1.0 C or less than -1.0 C.
Of those changes, 410 are greater than 0.5 C or less than -0.5 C -- that's 3.5% of their data points.
I find that pretty remarkable. And yet the usual suspects will swallow this without questioning it at all, and be thankful for it.