The present examples provide technologies for estimating code failure proneness probabilities for a code set and/or the files that make up the set. The code set being evaluated is typically comprised of binary and/or source files that embody the software for which the estimates are desired. The estimates are typically based on a set of selected code metrics, the code metrics typically selected based on corresponding failures of a previous version of the software. A historically variant metric feedback factor may also be calculated and code metric values classified relative to a baseline code set embodying the previous version of the software.