I am within a few days of releasing WXSIM 2015 Build 2.0, which corrects a couple of bugs, slightly improves use of WXSIM-Lite data, and improves the “enhanced nocturnal decoupling” routine, which seems to be especially useful in New Zealand. Meanwhile, I did a fresh analysis of the data I’m collecting on the optimum mix of WXSIM and WXSIM-Lite. I have data from mid October of last year through now, but am concentrating on the part since the GFS changeover in January. I’m attaching a graph which shows the mean absolute errors (for minimum, maximum, and mean temperature, in Fahrenheit degrees, for my site here near Atlanta) as a function of mixture percentage (0, 40, 60, 80, and 100%). The graph posted here uses 634 forecasts from each of these mixtures. I am getting a very robust result that the best percentage is around 60. I do suspect this varies with location, and some of the more difficult customizations (i.e. if WXSIM was not all that accurate for you in the first place) might do better with 70%. In this data set, WXSIM-Lite on its own just barely edges out WXSIM alone, with the mix at least 20% better (less than 80% as much error) as WXSIM-Lite alone. These errors, by the way, are mean absolute ones, for forecast days 2 through 4. The mean error of about 2.4 degrees F (1.3 or 1.4 Celsius degrees) - with data spanning winter, spring, and summer - is about as good as any source around, so this effort has proven to be a real success, at least here.
If you are using autolearn, and have been for some time (spanning before and after beginning use of WXSIM-Lite), you may be able to see this in your own data, especially by looking at graphs. I’d be interested to see such results!
By the way, so far, I’m still not charging for the WXSIM-Lite data, but do intend to eventually. Sam’s helping me with that and has been pretty busy with work, and I’ve been busy with our first grandchild (daughter’s little girl!).
Thanks to everyone for your support of this software! 
Tom
