I know a member of the Lehman Prop desk that sponsored the project, hence I have some idea of the inner workings. I thought it was interesting back in '03-'04, since I was working on a similar environment at the same time (actually conceiving, I didn't start working on it until I left the ibank). If I were to give away any specifics, here it is, the key of the PLAT project is not the various agents (they are just toys), the value (or "holy grail") is the PLAT environment itself. It enables people to experiment with different algorithms very quickly, rather than having to trade blind. Naturally the results of actual trading need to be feedback into adjusting the simulation environment itself. Building a good market simulator is no small task by itself. Then you can experiment to your hearts content with different strategies (arbing, volatility spread, phase of the moon, anything and everything). I consider almost all the commercial simulators I have see almost "jokes", since they don't reflect properly on the marketplace itself. Andrew Lo's group at MIT have done some very good work, and Morgan Staley's Market Microstructure group, have made enormous headway into how the market works, and well. "IF you can simulate it, you can beat it".