This group of work covers everything required to build a representative set of candidate benchmark programs to choose from and to select a subset of them for inclusion in the benchmark suite. We would choose the subset such that it doesn't contain duplicate workloads but still covers the range of workloads represented in the full set of candidates. Upon completion we can be sure that our benchmarks make efficient use of experiment time, and that speed ups to our benchmark suite translate into speed ups in real world applications.

See the RFC for more details.