paper.json: A Standard for LLM-Agent-Readable Academic Papers
The recently introduced paper.json convention suggests a streamlined JSON file to accompany academic PDFs, tackling the challenges faced by LLM agents when interpreting traditional text. This initiative comprises five key elements: stable claim IDs (C1), a clear does-not-claim list (C2), precise shell commands for each figure (C3), stable definition IDs (C5), and a minimum viable compliance guideline (C4) indicating that creating hand-written JSON alongside the PDF can be completed in less than an hour. The goal of this proposal is to facilitate citation at a sub-paper level, avoid scope overreach, and ensure figure commands are readily available within the paper itself.
Key facts
- paper.json is a companion JSON file for academic PDFs.
- It addresses failures when LLM agents read standard prose papers.
- C1 provides stable claim IDs for sub-paper granularity citation.
- C2 is an explicit list of what the paper does not claim.
- C3 includes exact shell commands for each figure.
- C5 provides stable definition IDs.
- C4 states minimum viable compliance is achievable in under an hour.
- The convention is proposed for arXiv paper 2605.16194.
Entities
Institutions
- arXiv