“ContextFidelity-Bench: A Three-Paradigm Framework for Evaluating Multi-Turn Hallucination Behavior in Language Models”, TBench, vol. 6, Sep. 2026, doi: 10.66834/0azb3w20.