I tried a better approach to the USHCN code. Previously I was averaging all monthly temperatures from all stations in a given year together. The problem with this is that some months have more station data than others, and is made worse by the fact that the USHCN adjusted data had extra manufactured temperatures in May. This caused the warmer temperatures in May to get weighted more heavily in the adjusted data than in the raw data.
The new approach is to average the months individually across all stations, and then take the average of all monthly averages in that year. This eliminates the issues caused by a larger number of stations with May data in the USHCN adjusted data set, than in the raw data set.
The result is a much smaller spike in 2014, similar to what Anthony plotted. But it doesn’t change the fact that using my approach the adjustments go up rapidly after 1995, whereas Anthony’s tails off flat. That discrepancy is the $64,000 question.