Thought I'd share with you the research paper I wrote based on my work on Foldit, Designing and Evaluating Player Learning in Human Computation Games. The first half is about the theory I've extracted, about the design considerations important for learning games, and the second half is about the mistakes we (I) made in adjusting the design. Some statistics are made public there.
Here's the abstract:
"We present a new approach to the design of effective training systems for participants in human computation projects. When the computation involved is not only too complex for computers to solve, but too specialized even for an untrained human to handle, there must be an effective and enjoyable way for would-be participants to learn the skills necessary to contribute meaningful results to the project. Our work on the protein structure prediction game Foldit indicates that player learning is most strongly affected by a small set of key design components, whose effect on player understanding and ability can be evaluated through both formal and informal data collection."
Feedback welcome. :) Let me know if you have any questions.