🐀 ratctl v0.1.0
📦 GitHub 📄 PyPI

🔍 Verifier Code Audit

Paste a Python verifier, reward function, or grading harness to scan for reward-hacking exploits.

verifier.py
📚 About ratctl
Audits RL environment verifiers for reward-hacking exploits across 6 exploit classes: Test Tampering, Grader Manipulation, Premature Termination, Environment Hijacking, Reward Skipping, and LLM Judge Bias.

Empirical Study: Audited 112 public RL post-training environments — 61.6% vulnerability rate.
View on GitHub →
🐀

Reward-Hackability Auditor

Select a preset example or paste your own verifier code, then click "Run Security Audit" to analyze for reward-hacking vulnerabilities.