Formal outputs: preprints and completed studies, each with a fixed version, a date, and the materials behind it.
Preprint
Source Selection, Conflict Retention, and Novel-Case Classification in AI-Mediated Interpretive Pipelines
An empirical study of what happens before an AI answer reaches the user: which sources survive selection, which disagreements remain visible, and whether materially new cases can escape existing classifications.
Preprint
Runtime Action Spaces, Forensic Readiness, and the Auditability of Agentic AI
A study of how AI agents acquire real-world capability through connectors, skills, permissions, tools, and execution environments — and why auditing them requires reconstructing the runtime that made an action possible.
Study
LoRA-SFT / LoRA-DPO Evaluation Study
A completed experimental study of small factual updates in LLMs, comparing LoRA-SFT and LoRA-DPO across update insertion, hallucination-like side effects, Mean BS scores, and general capability retention.
