If basestations include a direction vector the triangualation with two of them would be accurate enough. But I think they're only using signal strength to project a circle around them and calculate the crossing areas as your possible locations.

This area will be bigger with only two BS (not necessary, but likely).
However, this whole thing seems strange to me, as you may be in a parking lot, with bad network connection at all, and the BS triangulation will yield to false values?