Why would this happen?
Is there anything you can do about it without killing the history?
I’m wondering if selecting both the base surfaces and doing a Dir > UReverse or VReverse (depending on whether the U or V direction edges were used for the blend) might result in flipping the blend back without killing the history…
Thanks for the tip, I did not know about that command before (but dear god, do I wish that Rhino would write the UV direction names instead of just color coding them, especially since although I’m not color blind, the red/green color blindness is apparently the most common).
Unfortunately, it didn’t seem to do much to improve the result of the command (recreated it as well, just in case):
Can you post your example (just the 2 surfaces, plus the history-enabled blend) for people to experiment with?
Oh. My. God.
If I select the right surface first, I get a blend surface oriented in one direction, and if I select the left surface furst, I get a blend surface oriented in the other direction… 
Interesting - the Blend surface does not seem to be history-related to the original surface UV directions or their normals…
Surfaces constructed with their edges running in the “counter-clockwise” direction generally have their normals facing upward. In playing with it, it also appears where you click on the edges that changes the direction loop.
Clicking on the upper-left then upper-right results in a up-facing normal, as does lower-right then lower-left. The other two choices result in a downward facing normal.
It seems fairly arbitrary actually, like the “auto guess” function in FlowAlongSrf - no way to know where to click in advance that I can find.