Claude Opus 4.6 accuracy on BridgeBench hallucination test drops from 83% to 68%

Develop a lightweight automated monitoring tool to track LLM hallucination metrics in production. Bridge the gap between static benchmark scores and real-world performance degradation.

Suggested repo: hallucination-watch

"Stop trusting benchmarks and start measuring real-world reliability."

Estimated effort: 60h