Search papers, labs, and topics across Lattice.
This paper introduces HALLMARK, a comprehensive benchmark designed to diagnose the failure modes of citation verifiers used in large language models (LLMs). By evaluating various verification approaches鈥攊ncluding rule-based, LLM-based, and tool-augmented agents鈥攁cross 2,526 BibTeX entries, the study reveals that the false-positive rate (FPR) is the critical factor determining the deployability of these verifiers, rather than recall. The findings highlight that while agentic lookups can improve recall, they significantly increase false positives, underscoring the need for a balance between detection accuracy and noise in citation verification systems.
False-positive rates, not recall, dictate the deployability of citation verifiers in LLMs, revealing a critical trade-off for researchers.
Large language models (LLMs) now routinely draft literature reviews and assist with academic writing, which means a higher risk of fabricated references: GPTZero found 53 papers with hallucinated citations among NeurIPS 2025's accepted set. Rule- and LLM-based verifiers are emerging, but no shared benchmark compares them and gives detailed failure diagnostics. We close that gap with HALLMARK (Hallucination benchmark): 2,526 BibTeX entries spanning 14 hallucination types, three difficulty tiers, six diagnostic sub-tests per entry, and a contamination-resistant held-out split. On it we evaluate a DOI-lookup baseline, frontier LLMs zero-shot, tool-augmented agents, and our own rule-based, co-designed verifier bibtex-updater. Across the benchmark one result is consistent: the false-positive rate, not recall, decides whether a verifier is deployable. HALLMARK makes it concrete through three failure modes: agentic lookups buy recall but inflate false positives; at a venue-realistic base rate, the order-of-magnitude spread in false-positive rates (FPRs) -- not recall -- governs whether a verifier's flags are mostly true catches or mostly noise; and most LLMs over-flag papers published past their training cutoff, where only the two latest-cutoff models hold their false-positive rate near in-distribution levels (a signal we report as descriptive, since it is confounded with possible recall of those entries). Thus FPR is the deployment bottleneck, but an undetected fabrication remains the costlier error for the scientific record.
