Test whether your published papers still run

Mike Frank had Claude inspect the code and data behind his recent journal papers and it uncovered serious reproducibility problems.

Done with Claude· Advanced· Nov 2026

What they did
Mike Frank used Claude to go through every journal paper he had co-authored over the last five years and check whether the computational work could be repeated. The agent relied on the papers’ publicly posted code and data and worked from those materials alone.
How it went
The check surfaced major reproducibility trouble even though the files were openly available. The post does not include a paper count, specific failure modes, or a breakdown of which papers passed.
Tips
This approach depends on having public code and data to inspect. If you want a similar review, have the agent do the work in a duplicate copy and pause before any action that would change, send, or delete something.

Try it yourself with Claude

I want to verify the reproducibility of my published papers using the code, data, and PDFs in [locations]. For each paper, set up a fresh copy, try to execute the analysis, and compare the outputs with the reported figures and tables. For every item, tell me whether it fully matches, partly matches, or does not match, and name the reason if something breaks, such as missing inputs, broken paths, or version conflicts. Do not touch my original files. If anything would involve publishing, deleting, sending, or another irreversible step, stop first and ask for my approval.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on bsky.app ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.