FMG-Bench is a 120-scenario benchmark for evaluating large language model
behavior in theological triage and pastoral-guidance-adjacent contexts.
This release contains the open v1 benchmark corpus: 120 base scenarios with 37
perturbation variants. It is intended for researchers and engineers studying how
models handle faith-facing questions involving doctrine, tradition, moral
guidance, user preference, grounding, and escalation boundaries.
Project links:… See the full description on the dataset page:
https://huggingface.co/datasets/FideAI/fmg-bench.
