[quoting Foshay, 3 Mar 97.a] Sigh. It would be much easier to design my next math problem solving simulation if I could just suspend the law of gravity and count number of photon torpedoes used instead of having to figure out how to measure some real problem-solving skill.
I wonder here if you are trying too hard. For some specific skills, you need very context-specific training. But for something like maths problem-solving, you can't expect all the contexts you need, so you want far transfer. Why, then, NOT use photon torpedos? Well, OK, something a little less 11-18 male demographic, i.e., raising Vulgarian Mud Ponies, or producing Malbanian ZoomWands.
An idea I've been toying with is that you may need two vastly different contexts to facilitate abstraction. As in the analogy research of Gick & Holyoak, when it took presentations of two vastly different examples of a solution before transfer significantly improved. Thus Mud Ponies AND ZoomWands.
I'd like to suggest that you may even facilitate far transfer by fantastic settings rather than concrete ones.