What Do Claim Verification Datasets Actually Test? A Reasoning Trace Analysis
This work identifies limitations in existing claim verification benchmarks, which is incremental but important for researchers aiming to develop more robust evaluation methods.
The study analyzed claim verification datasets and found that they primarily test direct evidence extraction, with under-representation of multi-sentence synthesis and numerical reasoning, revealing biases such as lexical matching dominance in some datasets.
Despite rapid progress in claim verification, we lack a systematic understanding of what reasoning these benchmarks actually exercise. We generate structured reasoning traces for 24K claim-verification examples across 9 datasets using GPT-4o-mini and find that direct evidence extraction dominates, while multi-sentence synthesis and numerical reasoning are severely under-represented. A dataset-level breakdown reveals stark biases: some datasets almost exclusively test lexical matching, while others require information synthesis in roughly half of cases. Using a compact 1B-parameter reasoning verifier, we further characterize five error types and show that error profiles vary dramatically by domain -- general-domain verification is dominated by lexical overlap bias, scientific verification by overcautiousness, and mathematical verification by arithmetic reasoning failures. Our findings suggest that high benchmark scores primarily reflect retrieval-plus-entailment ability. We outline recommendations for building more challenging evaluation suites that better test the reasoning capabilities verification systems need.