When News Becomes Theory: Testing Evidence-Preserving Cross-Domain Hypothesis Generation in AI-Assisted Scientific Research

News reports, scientific announcements, public records, and institutional statements increasingly enter language-model-assisted research pipelines as candidate evidence for novel hypotheses. Their apparent compatibility across domains can invite a sequence of unjustified promotions: a reported claim becomes an accepted observation, an analogy becomes a mechanistic explanation, and a proposed hypothesis becomes a conclusion. Existing research on literature-based discovery, scientific reasoning, provenance, causal transportability, factual grounding, and research quality already supplies substantial safeguards. This conceptual paper asks whether an additional Field-oriented evidence-qualification representation offers incremental value beyond a competent, information-matched conventional workflow. It introduces the Cross-Domain Evidence Admissibility Hypothesis, defines a minimal graph of claims, warrants, transformations, scope constraints, and challenges, and specifies a prospective comparative protocol emphasizing unsupported-claim promotion, legitimate hypothesis coverage, calibration, abstention, cost, and independent adjudication. The central comparison explicitly permits observational equivalence and the decision that no Field abstraction is necessary. The framework is not empirically tested; no improvement, causal effect, ontology, or scientific priority is established. The proposed contribution is a falsifiable research contract for distinguishing the generation of scientifically interesting questions from warranted assertions about the world.