Any results that are reported to constitute a blinded, independent validation
of a statistical model (or mathematical classifier or predictor) must be accompanied by a detailed explanation that includes: 1) specification
of the exact «locked down» form
of the model, including all
data processing steps, algorithm for calculating the model output, and any cutpoints that might be applied to the model output for final classification, 2) date on which the model or predictor was fully locked down in exactly the form described, 3) name
of the individual (s) who maintained the blinded
data and
oversaw the evaluation (e.g., honest broker), 4) statement
of assurance that no modifications, additions, or exclusion were made to the validation
data set from the point at which the model was locked down and that neither the validation
data nor any
subset of it had ever been used to assess or refine the model being tested