How do we label a mark we cannot agree about?

Our proposed FR10 inspection uses brushed stainless trim plates. We kept the physical samples, so I can trace each image back to its plate. The problem is now the judgement: one inspector calls a line part of the brushed finish, another calls it a scratch. And a separate bright band changes when the plate is presented differently. I am reluctant to hand either disagreement to the programmer as a threshold problem. Who has found a useful way to compare the samples with the people accepting the finish?