Polygenic risk scoring in forensic psychiatry: Predictive validity, evidentiary standards, and a proposed auditability framework

Corrected author preprint, version 0.2, 14 September 2026. This version supersedes the substantive claims of the earlier preprint. It is not a journal acceptance or journal publication. Polygenic risk scores summarize genetic liability but do not establish individual diagnosis, intent, credibility, or conduct. This narrative review and conceptual analysis examines psychiatric prediction studies alongside evidentiary requirements for algorithmic genomic reports. It proposes an auditability framework comprising cryptographic provenance, transparent analytical configuration, expert review, and deterministic calculation traces. The framework supports scrutiny but has not been independently validated and cannot guarantee admissibility or clinical safety. Evidence from a hospital respiratory-diagnosis vignette study illustrates possible benefits and harms of artificial intelligence assistance; it does not measure this genomic system or forensic practice. Earlier suicide-mortality projections are withdrawn because their transport, detection, treatment, and causal-effect assumptions were unvalidated. Any clinical utility must be established prospectively against usual assessment, with ancestry-appropriate validation and measurement of care changes, adverse effects, and patient outcomes. The contribution is a proposed verification standard, not evidence of clinical effectiveness. Correction notice. The earlier estimates of approximately 297 Australian and 66,900 global deaths averted per year, including the roughly one-tenth Australian framing, are withdrawn. No number of deaths averted, positive lower bound or net mortality benefit is established. The manuscript also qualifies claims concerning admissibility, auditability and individual genetic or epigenomic inference. The underlying manuscript was corrected on 13 September, with these targeted qualifications and the author release prepared on 14 September. Related methodological correction: Accuracy is not mortality. The paper is a literature-based conceptual analysis and reports no new participants, personal genomic data or primary clinical data. Research and preparation were self-funded. The author develops the AthenaWGS system and discloses a commercial and intellectual interest. AI assistance is disclosed in the author declarations. The previous version remains available as part of the version history.