Research and evaluation
Research & working papers
What does a successful result establish?
Across coding agents, financial models and microbiome studies, I examine whether the evidence supports the claim.
These papers are working manuscripts, not peer reviewed. AI tools were used throughout the research workflows. Each project summary describes individual and collaborator contributions and the limits of the evidence.
Working manuscript · October 2026 · Not peer reviewed
Commandem
Completion Claims and Verifier Integrity in Coding Agents
A retrospective audit of what completion markers, passing checks and recorded agent behavior actually establish.
Contributions. I developed Commandem and contributed research design, implementation, experiment execution, analysis and manuscript preparation, using AI tools throughout. My collaborator helped audit, develop and test the security experiments.
Working manuscript · October 2026 · Not peer reviewed
Quantem
Auditing Timing, Costs, and Graph Value in Relational Stock Prediction
An audit of timing, transaction costs and graph evaluation in relational stock prediction, separating controlled evidence from recovered historical outputs and conditional replay.
Contributions. My contributions span research design, implementation, experiment execution, analysis and manuscript preparation, using AI tools throughout the workflow.
Collaborative working manuscript · Not peer reviewed
Oralbiome
Oral Microbiome Models of Diabetes Status: A Retrospective Audit of Cross-Study Transfer
A retrospective audit of oral-microbiome diabetes-status models, focused on cohort eligibility, leakage, calibration and cross-study transfer.
Contributions. I contributed study design, baseline and CatBoost analyses, interpretation, transfer investigations and manuscript preparation. My collaborator contributed preprocessing, exploratory analysis, feature engineering and a gradient-boosting comparison. We both contributed to methods and experiments, using AI tools throughout the workflow.