There is a growing recognition of the importance of considering findings in relation to activity-level propositions to best advise the fact-finder of the weight of support that the evidence provides in a criminal case. However, many practitioners perceive a shortage of high quality, relevant data for supporting such activity level evaluations (ALE). This perception may be driven not only from limitations in study design but also from uneven access to the underlying literature across the discipline. Robust and trustworthy ALE assessments depend on the systematic identification and validation of appropriate data sources.

A major time component in performing activity level evaluations (ALE) is the extensive, case specific literature review required to populate Bayesian networks with fit for purpose parameters. Previous efforts have compiled publications on the transfer, persistence, prevalence, and recovery (TPPR) of forensic traces to facilitate this. However, these databases were designed for manual interrogation and lack the metadata and interoperability needed for automated workflows. Moreover, the overall breadth and depth of the literature have not been described, leaving practitioners uncertain about which forensic questions are sufficiently supported by empirical data.

This presentation showcases the use of an AI-supported analysis pipeline to collate information on TPPR relevant publications from open access repositories and forensic specific data sources. Using only publicly available content, the studies were then scored by their semantic similarity. These similarities were visualised to reveal clusters of dense empirical coverage versus sparsely studied forensic traces.

Our findings indicate that studies can be clustered by their content into high-level groups and the TPPR literature contains data, though uneven, for many commonly recovered forensic traces. The AI‑supported approach not only clarifies where data are abundant enough to support robust ALEs but also highlights gaps that warrant targeted experimental research.

This pipeline describes a way of rapid triaging TPPR literature to those studies most pertinent to a given case scenario. The collation of the data also acts as a precursor to a knowledge base appropriate to integration into an automated ALE analysis pipeline. Understanding the availability of underlying data may encourage forensic scientists that evidence‑based activity‑level evaluations are possible in some circumstances, ultimately strengthening the scientific underpinnings of courtroom testimony.

Authors

• Janet Stacey (The University of Auckland/ PHF Science, New Zealand)
• James Curran (The University of Auckland, New Zealand)
• Sally Ann Harbison (The University of Auckland/ PHF Science, New Zealand)