Stress-Testing the Reasoning Competence of LLMs With Proofs Under Minimal Formalism

TL;DR

ProofGrid is a comprehensive benchmark suite that evaluates large language models' reasoning abilities through machine-checkable proofs in minimal formal notation, providing fine-grained, reproducible assessments.

Contribution

The paper introduces ProofGrid, a novel benchmark with formal proofs in NDL, enabling precise evaluation of LLM reasoning beyond final answers, and develops a new proof-checking pipeline.

Findings

Models perform well on foundational tasks but struggle with complex reasoning.

ProofGrid reveals epistemic instability where models generate flawed proofs but reject incorrect inferences.

The evaluation framework includes advanced analysis methods like IRT and Fisher information.

Abstract

We introduce ProofGrid, a benchmark suite for evaluating LLM reasoning through machine-checkable proofs rather than final answers alone. ProofGrid contains 15 tasks spanning proof writing, proof checking, proof masking, and proof gap-filling. Tasks are expressed in minimal formal notation, especially NDL, a compact natural-deduction language that fits in short prompts and supports precise, auditable verification. This yields mechanical, reproducible, and fine-grained evaluation rather than judgments by humans or LLMs. ProofGrid covers a calibrated difficulty spectrum, from foundational reasoning tests to structurally rich challenge tasks that no current model solves, while minimizing reliance on domain knowledge, solver delegation, and long-context artifacts. We also develop a comparative framework for reasoning benchmarks and use it to situate ProofGrid relative to existing work in terms of representation, verification guarantees, and reasoning depth. Methodologically, we introduce an instrumented proof-checking pipeline that tolerates minor surface deviations while locating the first substantive reasoning failure, improving measurement resolution and separating proof planning from low-level execution noise. Using this pipeline, we evaluate a broad range of open and proprietary models. Results show rapid progress but substantial remaining limits: frontier models perform well on several foundational tasks, yet difficult tasks, especially those requiring global combinatorial reasoning or low-level proof synthesis, remain far from solved. We also identify epistemic instability, where models generate flawed proofs yet correctly reject those local inferences in isolation, and formalize this with an Epistemic Stability Index. Finally, we complement accuracy with 2PL IRT analyses, Wright maps, and a normalized task-discrimination measure based on Fisher information.