ThinknCheck: Grounded Claim Verification with Compact, Reasoning-Driven, and Interpretable Models

2026-04-02

AI Summary

This work addresses the challenge of achieving efficient, interpretable, and highly accurate fact verification in resource-constrained settings. The authors propose a reasoning-driven lightweight verifier with 1B parameters that first generates structured reasoning rationales under explicit supervision before producing a binary verdict. Built upon a 4-bit quantized Gemma3 model, the approach integrates supervised fine-tuning with domain-specialization strategies—such as ThinknCheck-Science—and is trained on LLMAggreFact-Think, the first reasoning-augmented fact-checking dataset. Experimental results show that the model attains a balanced accuracy of 78.1% on LLMAggreFact, outperforming MiniCheck-7B—a model seven times larger in parameter count—and achieves a 14.7-percentage-point improvement on SciFact. Ablation studies further confirm the critical contribution of the explicit reasoning step to overall performance.

Abstract

We present ThinknCheck, a 1B-parameter verifier for grounded claim verification that first produces a short, structured rationale and then a binary verdict. We construct LLMAggreFact-Think, a 24.1k reasoning-augmented training set derived from LLMAggreFact, and fine-tune a 4-bit Gemma3 model to follow this format. On LLMAggreFact, ThinknCheck attains 78.1 balanced accuracy (BAcc), surpassing MiniCheck-7B (77.4) with 7x fewer parameters; removing the reasoning step reduces BAcc to 57.5. On SciFact, ThinknCheck reaches 64.7 BAcc, a +14.7 absolute gain over MiniCheck-7B. By contrast, zero-shot chain-of-thought on the base Gemma3-1B harms accuracy relative to direct answers, and preference optimization with a simple format+accuracy reward underperforms supervised reasoning. To probe the latter, we introduce GSMClaims and a domain-specialized variant, ThinknCheck-Science, which improves across benchmarks, including 61.0\% accuracy on GSMClaims. Overall, explicit, supervised reasoning enables compact verifiers that are competitive while remaining resource-efficient and interpretable.

Problem

• claim verification
• reasoning
• interpretable models
• compact models
• fact-checking

Innovation

• grounded claim verification
• reasoning-driven models
• compact verifiers
• supervised reasoning
• interpretable AI