Selected code

Systems built to answer hard questions.

These repositories turn research questions into inspectable experiments, evaluation tools, and reusable systems.

P.012026

CFR without unbiasedness

An experimental C++ codebase for studying regret minimization with nonstandard estimators.

P.022026

LLM exploitability audit

A Python toolkit for exact worst-case audits of language-model policies.

P.032026

Safe observation capacity

Research code for active de-censoring in imperfect-information games.

P.042026

BATR research toolkit

A Python and Rust toolkit for model-guided public-belief search.