Paper 1 · Measure · July 2026
Validating an AI Grader for Community Corrections Supervision Visits
Can a frontier AI model grade recorded supervision visits against a 43-item, standards-based rubric as reliably as expert human auditors?
RePath Labs Evidence Series
Research papers on AI and supervision quality
Working papers from RePath Labs on measuring and coaching what happens inside community corrections supervision visits, on choosing the right AI for each task, and on what routine supervision data says about outcomes. Full text on this site, permanent DOIs, methods and limitations stated. Free to read, no email required.
The series so far
Five papers: expert feedback on every visit, the right AI for the job, and an early signal in check-ins
Paper 1 tests measurement, Paper 2 tests coaching, and Paper 3 tests both against an expert-agreed standard. Paper 4 is the first study we know of to test decision models, a new class of AI, against an expert-agreed standard in a professional field. Paper 5 looks at RePath's routine check-in data as an early indicator of how pretrial cases end. See how they fit together →
Paper 2 · Coach · September 2026
Validating AI-Generated Officer Coaching for Community Corrections Supervision Visits
Can AI turn a supervision-visit transcript and its scores into officer coaching that an independent expert rates as highly as expert human coaching?
Paper 3 · Standardize · September 2026
Establishing and Meeting a Reference Standard for AI Grading of Community Corrections Supervision Visits
Can three experienced practitioners agree, item by item, on what a supervision visit demonstrates, and can an AI grading workflow reproduce that agreement?
Paper 4 · Compare · October 2026 · Submitted for publication
Decision Models in a System Built for Language Models
How well does a new class of AI, decision models, fit into a grading system for supervision visits that was built with expert human effort and frontier large language models?
Paper 5 · Signal · October 2026 · Submitted for publication
Mobile Check-In Completion as an Early Indicator of Case Outcomes in Pretrial Supervision
Does the share of scheduled app check-ins a participant completes in the first 30 days of pretrial supervision say anything about how the case ends?
How we publish
Our publication standards
Labeled honestly
Every paper is a working paper (preprint). A DOI makes it permanent and citable; it is not peer review, and we never call it that.
Methods and limits in full
Design, data handling, statistics, conflicts of interest and every limitation are in the paper, not in a footnote.
Versioned, never silently changed
Corrections ship as new versions with a stated change. Earlier versions stay available through the DOI record.
Privacy first
Visit transcripts are de-identified before research use and are never published. Proprietary prompts are withheld and described.
Questions about methods or materials: tedgreen@ehawksolutions.com. Published literature these studies build on: Chadwick, N., DeWolf, A., & Serin, R. (2015). Effectively training community supervision officers: A meta-analytic review of the impact on offender outcome. Criminal Justice and Behavior, 42(10), 977-989.; Labrecque, R. M., Viglione, J., & Caudy, M. (2023). The impact of community supervision officer training programs on officer and client outcomes: A systematic review and meta-analysis. Justice Quarterly, 40(4), 587-611.