Develop benchmarks for agent validations

[rich-iannone]

This will involve taking a standard table (with different storage types) and running against a standard set of validations. The results can be put into a table, possibly part of a vignette/article.