> Paper2Agent
A paper is a fixed account of an analysis: you can read what I found, but not ask it anything. These pages connect an AI agent to the code behind my papers, so you can re-run the models, change a specification, and see what moves — in conversation, without installing anything.
The numbers come from the original analysis code, executed on request. The agent chooses which analysis to run and explains the output; it does not compute the results itself, and it cannot quietly substitute a remembered number for a real one.
Available agents
More of my papers will appear here as their analysis code is packaged the same way.
The method
These agents are built with Paper2Agent (Miao, Davis, Zhang, Pritchard & Zou, arXiv:2509.06917), a system that reads a paper and its repository and wraps the existing analysis code as tools an AI agent can call, then has separate agents verify each tool against results from running that code directly. Its authors anticipate an “agent availability” section in papers, alongside data and code availability. That seems right to me, and this is my attempt at it.