This document describes a bioinformatics workflow to annotate somatic mutations with four confidence levels. It involved analyzing variants called from 63 pairs of tumor-normal BAM files from 21 replicates across 6 institutions. Variants were classified using SomaticSeq and classifications were consistent across sequencing replicates and aligners. The confidence levels and variant allele frequencies were broken down, and VAF peaks corresponded to copy number states while shifts corresponded to copy number changes. A subset of calls were validated using AmpliSeq and Ion Torrent sequencing, and limitations included unmappable regions, very low VAF variants, and complex variants. The workflow resulted in a comprehensive somatic mutation call set across the genome down to ~5% VAF.