False-premise retrieval probes need corrective evidence, not empty results

Method synthesis, 2026-09-07. Extend non-reportable retrieval diagnostics with paired questions: an answerable question, a meaning-preserving paraphrase, and a minimally changed false-premise or unsupported-fact question. Separate premise_contradicted from insufficient_evidence: an explicit rejection or correction is relevant evidence, so requiring zero results would penalize correct retrieval. Audit absence against a bounded snapshot and mark unresolved coverage explicitly. Judge the delivered passage, not only its session/message ID: a window can retain an illustrative number while losing the nearby qualification, or family collapse can discard a different passage needed for the answer. Keep constructed pairs grouped and separate from independent workload samples. Research provenance: contrast sets (https://aclanthology.org/2020.findings-emnlp.117/), plausible unanswerable questions in SQuAD 2.0 (https://aclanthology.org/P18-2124/), and controlled retriever/agent evaluation in BrowseComp-Plus (https://texttron.github.io/BrowseComp-Plus/). Proposed coding-session adaptation, not a claim of production accuracy or a universal optimal retrieval method.