POLICY
How we handle
what we find.
We check specific calculations in published astrophysics papers and publish our code so anyone can rerun them.
Claims, not people
Every published conclusion identifies its scope and supporting evidence. We may report a bounded reproduction attempt, including a documented barrier or unresolved result. An unsuccessful attempt does not by itself establish an error in the paper. We do not comment on authors’ conduct.
Authors hear first
Before we publish a discrepancy that changes a result, we send it to the authors. After 14 days, the project owner decides whether to seek an independent check. The deadline does not trigger contact or publication. Author responses are represented fairly; private correspondence is not published without permission.
A paper that reproduces, or has only minor notes that change no result, is published as a package without writing to the authors. If they contact us about it, we record their reply and correct the package where they show we were wrong.
A confirmed finding identifies who confirmed the specific discrepancy: the authors or an independent researcher. We may also publish a narrowly supported technical finding labelled “Technical finding — not externally confirmed”, with its assumptions, evidence and unknown consequences stated explicitly. This requires the owner’s approval of that exact scope and package; it is not external confirmation. Author silence is not agreement. Unresolved substantive disputes remain on hold, and a documented attempt cannot be used to disguise an unsupported finding.
AI assistance
AI models help us derive and check results. A named person approves every released report and is responsible for it. Each report discloses the tools, checks and human involvement relevant to its conclusions.
Corrections
If we get something wrong, we correct it publicly and keep the history. Write to us at [email protected].