Final-context accuracy in multi-passage fact verification can hide whether an added passage contributes a new verdict-relevant fact or merely repeats evidence already present. We introduce a controlled same-prefix diagnostic derived from HoVer supporting-fact annotations to measure marginal evidence contribution. For each claim, a shared prefix covers one fact’s evidence while leaving another required fact absent; the next passage either repeats the covered fact or supplies the missing fact. The diagnostic contains 1,180 same-prefix pairs and an exact-repetition ladder testing whether repeated copies of one evidence sentence can change a verifier’s label. Supervised verifiers with similar clean F1 scores, from 0.6324 to 0.6528, differ sharply in eight-repeat flip rate, from 8.5% to 30.3%. Conservative pooling and natural-language-inference (NLI) scoring reduce duplicate sensitivity but give near-zero missing-fact gain, while gain-oriented systems improve gain but remain repetition-sensitive. Latent-Gated Evidence Dynamics (LGED), a compact stateful reference model, shows that low duplicate sensitivity and positive missing-fact gain can coexist. Surface controls and decision-level analyses support this diagnostic interpretation. Redundancy and reader tests extend the duplicate-sensitivity finding beyond HoVer, while prompted open-weight LLMs reveal a gap between role recognition and verdict updating.

Puns are a challenging form of figurative language that exploit phonetic similarity and semantic ambiguity to convey multiple meanings. Although large language models (LLMs) demonstrate strong language understanding capabilities, they still struggle with pun reasoning due to limited phonetic modeling and uncontrolled end-to-end generation. We propose PunGraph, a retrieval-enhanced knowledge graph framework for pun understanding. PunGraph constructs a phonetic-semantic lexical graph using the Unisyn phonetic dictionary, IPA and G2P representations, and WordNet definitions, and retrieves candidate words or senses to constrain LLM reasoning within a structured candidate space. We further introduce WebPun, a new large-scale dataset containing 5,730 annotated heterographic and homographic puns. Experiments on SemEval-2017 and WebPun show that PunGraph consistently improves the performance of small-scale LLMs and achieves competitive results against strong proprietary models. Further analysis shows that retrieval-guided phonetic and semantic constraints effectively reduce common reasoning errors in pun interpretation, highlighting the benefits of integrating structured knowledge with LLMs.