I remember there being a discussion somewhere about the risk of someone exploiting TCD by "bunkering", i.e. downloading a bunch of project work units at once, processing them, and waiting to upload the results until a drop in workunit supply from the project servers. If the workunit supply is significantly depleted (which definitely DOES happen occasionally in practice) then this person could dominate the project magnitude and get huge research rewards.

RAC is resistant to this exploit since it is a moving average which smooths out a researcher's output [credits/unit time] over the timescale of a couple of weeks. On the other hand, to protect TCD from the exploit we really do need stricter standards for projects (i.e. minimum workunit supply requirements) and a mechanism for rapidly sending projects to a greylist if they drop below this requirement. It seems to me that TCD should only be considered in combination with this mechanism.

There's some more in depth technical analysis which can be done in connection with this proposal. What I have in mind is quantifying any possible exploits/weaknesses in each proposed mechanism by parametrizing with respect to a finite number of variables, and maximizing research rewards with respect to those variables. I want to work on this if I have time (never a guarantee given my current job, but we'll see!).