Popular repositories Loading

1. foldguard foldguardPublic

   Catch protocol-level eval leaks at runtime: checkpoint-on-test-fold, calibrator-on-eval-labels, threshold-on-scored-data.

   Python 1

2. envdoctor envdoctorPublic

   Adversarial QA for LLM-RL environments: find out what reward an empty answer earns. Model-free, zero API cost.

   Python 1

3. hackbait hackbaitPublic

   Honeypot environments that measure reward hacking in LLMs. Structural verdicts, a control arm, and a zero-cost validation gate.

   Python 1