I am not in a position right now to check that particular test vector out but i wonder if the payload itself is in the correct format.
If you are getting the location of the GW returned then that suggests that the Skyhook lookup failed and Collos is resorting to location based on the next best thing and with a reception by only one GW that is the location of the GW.
It is interesting that the query works fine from Collos directly. I will try to take a look myself in the morning. Are you performing a simulated uplink or is this actually the payload you see in TTN console?