Pinned Loading

1. flawed-judge flawed-judge Public

   GSM8K RL environment for verifiers where an LLM judge is the reward, measuring how far judge reward drifts from true accuracy

   Python

2. decompilation-eval decompilation-eval Public

   Rebuilds every matched function of 16 public game decompilations on stock compilers to measure how much of each agent-built byte match holds

   Python

3. apex-rubric-audit apex-rubric-audit Public

   Pre-registered audit of whether the APEX-v1 rubric penalizes a response that contradicts itself

   Python

4. llm-rubric-eval-pipeline llm-rubric-eval-pipeline Public

   LLM-as-judge evaluation pipeline behind my NeurIPS 2025 GenProCC workshop paper on prompt valence

   Jupyter Notebook