Knowing Is Not Choosing: What Explicit Verification Adds Beyond Generative Preference

Generating a correct answer does not ensure that a language model will select it. We track factual competence through three stages: whether a correct candidate is generated, how candidates are ranked, and which answer is ultimately returned. Pre-generation readouts predict factual recall beyond exposure and identity features across three model families and, without refitting, which questions sampling will cover; however, they add little about whether an available correct answer is ultimately returned. Explicit verification with $P(\mathrm{True})$ improves within-question ranking over mean log-likelihood in Gemma, Qwen3, and Llama, raising pair-weighted AUROC by $0.078$–$0.119$. In a prospectively specified Gemma cohort, verification improves plurality accuracy by $4.99$ points under a frozen stricter criterion and by $3.88$ points under blinded semantic labels. Against chat-template likelihood, a stronger generative baseline, the gain remains $2.03$ points ($95\%$ CI $[0.16,3.92]$), concentrated in relations with strong common-answer priors and dependent on the entity. In two 27B Qwen models, verification exceeds chat-template likelihood by $0.13$–$0.17$ AUROC; masking the entity removes this ranking advantage, while plurality gains remain at most $2.9$ points. The measured gain also depends on the correctness criterion: recall-oriented reference matching credits option lists favored by likelihood and reduces the plurality gain over mean log-likelihood to $1.29$ points, whereas human labels give $3.70$.