Ah, so you're referring to the first jump of the second-place skater?
Looks like that's an example of my reason no. 2 -- there was a < call so most judges reduced the GOE below 0.
As of this year a << call requires that the final GOE be negative; the < just requires that the GOE be reduced by -1 to -2 from what it would have been otherwise.
(Last year, the judges weren't shown the < symbols)
In theory a judge could say that s/he saw the slight underrotation but also saw 6 or more other aspects of the jump that were good enough to deserve +3 so s/he reduced it to +2.
More likely, the judge just thought it looked great in real time, inputted the +2, and didn't go back to check the calls and adjust the GOE after the program.