While i am knowledgeable in surface texture, i do not have a good answer for how to "uniformly" create the test samples you desire. I would think that any scratch that can be detected by fingernail would need to be blended smooth.
Using the profilometer can give you varied results depending on what type and how it is used. Looking at Ra would not tell you 1 scratch is bad as it is and average reading, typically over 5 sample lengths. RzMax would give you max depth, but in my opinion checking the parts with a profilometer is overkill and non-value added.