Advisor(s)

Elizabeth Gardner

Committee Member(s)

Brandon Best
Shelly McGrath

Document Type

Thesis

Date of Award

6-2-2026

Degree Name

Master of Science in Forensic Science (MSFS)

School

College of Arts and Sciences

Department

Criminal Justice

Abstract

The science of firearm and tool mark identification is based on the premise that no two tools produce the same microscopic detail and the tool that made the marks can be distinguished from all others. Based on the AFTE Theory of Identification, firearms comparisons can be classified as Identification, Elimination, Inconclusive, or Unsuitable for examination, with inconclusive indicating insufficient individual characteristics to support identification or exclusion. Error rates in firearm examinations are typically calculated by counting how often examiners make false-positive and false-negative conclusions relative to known ground truth, but inconclusive decisions complicate these calculations because they neither support nor contradict a shared source. Different counting rules for inconclusive results have been used in various large-scale black-box studies, and signal detection theory (SDT) offers a framework for understanding how these rules relate to error rates. Traditionally, firearm and toolmark examination and comparisons have relied on the comparison microscope, but recent work has turned to 3D Virtual Comparison Microscopy (VCM) to generate quantitative similarity metrics from digital bullet and cartridge case models. A key question is how VCM and similarity scores influence examiner decisions and estimated error rates when distinguishing same-source from different-source comparisons. This study applies SDT to similarity scores generated by the Evofinder system (ScannBI technology, Leeds Forensic Systems) to a previous ten-consecutively manufactured barrel study that included damaged bullets. A total of 432 bullets from the Best and Gardner 2022 study, including 40 damaged bullets from ten consecutively manufactured Thompson/Center Arms G2 Contender barrels, were scanned into Evofinder, generating 3D topographic images and 93,096 pairwise similarity scores between 0.115 and 1 were generated. The study examines how score distributions for same-source (known match) and different-source (known non-match) pairs overlap and how the bounds of their distribution are affected by surface quality, bullet damage, and barrel manufacturing similarity. Scores were plotted by ground truth and evaluated to identify potential decision thresholds and outliers. This study aims to provide a more nuanced, SDT-based understanding of sensitivity, specificity, and potential error rates for score-based firearm identification using current 3D automated ballistic identification systems.

Keywords

Error rates;Evofinder;Signal Detection Theory;Similarity Scores;Virtual Comparison Microscopy

Download

Share

COinS