Upstream of likelihood ratio: A diagnostic framework for quantifying interpretive variability in forensic DNA mixture...

The reliability of DNA likelihood ratios (LR) depends on the integrity of the data input into probabilistic genotyping software (PGS). While downstream LR variance has been extensively studied, the interpretive variance introduced during the analysis of raw electropherogram (EPG), the stage upstream of PGS, remains less characterized. This study measures analyst performance in converting raw EPG signals into processed input files and determining the number of contributors (NoC) based on the available signal. Four analysts from the National Institute of Standards and Technology, the Federal Bureau of Investigation, and the New Zealand Institute for Public Health and Forensic Science independently interpreted a dataset of 375 profiles (one- to five-person mixtures). Using a multi-resolution analytical framework, we evaluated the propagation of interpretive variance across three critical stages: artifact removal, stutter filtering, and NoC estimation. To measure accuracy and isolate analyst's interpretive logic from signal loss (biological signals below detection thresholds), we introduced the novel concept of findable ground truth (fGT), a benchmark that evaluates performance based on the classification of observed signals. To measure reproducibility, concordance rates were quantified via inter-analyst pairwise comparisons. Results demonstrate that (1) there is a high level of concordance in peak classification, both relative to the fGT and among analysts, and (2) discordant outputs stem from either subjective decisions or inherent limitations in the DNA measurement process. This analysis emphasizes that the accuracy and reproducibility of the data input into PGS is as critical to the reliability of DNA profile interpretation as the final LR value.