I feel like we're overcomplicating things. We should be thinking of this as a simple physics problem: What method gives the best power-to-weight ratio for a simple bicep curl?
A 20 inch long blocks split half-way down the middle by a single hinge, pointing straight down with the upper "arm" fixed and a weight attached to the "forearm" and made to curl that weight 180° up using only electricity. The more torque out compared to weight of the arm of that length using an input of X watts DC for Y seconds is the winner. (I'm not sure how many watts or long would be considered reasonable for the experiment)
Finding out the most energy efficient way to do this is the key to making knees, elbows, fingers, and probably any other joint we'd need. Because most robots would run on batteries instead of wall outlets, the weight of any inverters would have to be included as part of the arm, because even if it weren't really in the arm, you'd still have the weight in her somewhere. This will likely also make hydraulics and pneumatics impractical even if it were more practical with more limbs. I know this might seem like an oversimplification, but it seems like something that can actually be made into class experiments or a challenge just about anyone could compete in.
And I didn't really know how to explain the rule, but no "C-3PO arms", so the blocks have to rest flat against each other, not having a linear motor or cables connecting the bicep to the forearm. If it were a leg that'd be like the back of the thigh connected to a calf, which would get in the way of sitting in a chair. It's not something you could easily hide inside of a human body or clothes and would really get in the way if they were used in prosthetic limbs.