My point is only that even when the software is correct, the reproducible build rate will either be 100% or we've got a representative sample set.

Until then, those issues will continue to be lumped together with all the software reasons it wasn't reproducible.