Reproduce “Decision Models as Policy Gates for Coding Agents: Explicit Policy Boundaries Improve Detection and Reduce False Alarms” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
James P. Beno
The authors test a decision-model gate for coding-agent tool calls. Rule examples define included and excluded actions; deterministic policy code combines probabilities into allow, block, or ask decisions. On a synthetic tool-only holdout, Jev 1.14’s recall rose from 89.3% to 98.2%, and false-positive rate fell from 2.49% to 0.83%. Clean-action interruptions fell from 30.6% to 4.8% across 252 actions, not production traffic.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.