ProofContext: Dependency-Aware Retrieval of Verifiable Context for Mathematical Reasoning

Read full text
Authors: Maximiliano Lucius
Cited by: 6
References: 1

Abstract

Reference implementation, adversarial near-miss benchmark (MathScopeBench), deterministic experiments, an affine-plank case study, and the arXiv manuscript for ProofContext. ProofContext compiles a mathematical query into a typed Problem Signature and returns the smallest auditable Evidence Bundle in which a claim can be safely used, using hard mathematical compatibility filters, an authority membrane (retrieval ranks but never promotes), and version-aware invalidation, layered read-only on the MIRADOR Mathematical Cell substrate. On near-miss categories decidable by a typed filter, ProofContext serves 0% of incompatible near matches vs 44-75% for lexical/dense/hybrid retrieval, at preserved recall. Source: https://github.com/maximilianolucius/proofcontext

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
12 July 2026