Evaluation refers to an agent's evaluation of the expected benefit of a particular action; meta-evaluation is the agent's evaluation of the computational cost-effectiveness of evaluation in general.

There are two difficulties with meta-evaluation.  One is the nature of the data, which by nature will consist of unique events*.  The second is the obvious self-referential nature of the problem.

Meta-evaluation is a central issue for any theory of rational decision-making, yet I have not yet seen it directly addressed here.

*This quality of nonredundancy occurs in its purest form in "mathematical data," ie the distribution of primes.