Step-labelled check reports for research papers: a proposal for Claude

An author may check a paper with AI before submission, and a referee may check the same paper again; neither sees the other's results. I propose a Claude tool with which the author checks a paper once, step by step, and a Claude page that lists checked papers together with their reports. Every step of a report states whether and how it was checked, gives its justification, and has its own problem thread, so that referees can read the report instead of repeating the check.