hallucinationguard-bench 0.0.1

An open benchmark for measuring what LLMs get wrong about quantum computing

hallucinationguard-bench

An open benchmark for measuring what LLMs get wrong about quantum computing.

Bag-of-words metrics cannot separate topic overlap from semantic agreement. Neither can mean-pooled word embeddings — "measurement collapses the state" and "measurement never collapses the state" land nearly on top of each other. This benchmark exists because of that gap.

Status: placeholder. The scoring harness lands here. The benchmark, fixtures, and baseline scorer are at:

https://github.com/Minapak/hallucinationguard

Apache-2.0.