Citation Grounding: Detecting and Reducing LLM Citation Hallucinations via Legal Citation Graphs

Volodymyr Ovcharov · 2026

Read Citation Grounding: Detecting and Reducing LLM Citation Hallucinations via Legal Citation Graphs by Volodymyr Ovcharov (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-00898.txt.

Description

Ovcharov shows that a popular check for fake legal citations mostly measures its own database in this 2026 paper. The check, citation grounding, looks up model-made citations in a graph drawn from real court rulings. No human graders are needed. He scores 400 replies: 100 Ukrainian legal queries run through four commercial models.

Two snapshots of the same national graph give wild gaps. Against a sparse graph of 470,000 records, scores run from 79.1 percent to 85.5 percent in the sparse graph test, which looks like 15 to 21 percent fake cites. Against a dense re-draw of 330 million records, the same replies score from 98.9 percent to 99.9 percent on the dense graph. A simple sampling model, fed only the record count, replays the sparse scores to within two hundredths. Worse, no pair of models can be told apart at 95 percent trust at any graph size. Sparse graphs rank harvest coverage, and dense ones rank nothing, he concludes.

Catalog record

CREATORVolodymyr Ovcharov
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.00898

Browse the full library · About the Tower of Babel