- Who it helps
- Researchers reviewing AI-generated papers and evaluation claims
- Availability
- Public preprint submitted October 28 and revised October 31, 2026.
A plausible value can still be misattributed
The authors examined 540 judgments and manually labeled 256 cases from eighteen submissions. Four in five unsupported values in their analysis were linked to attribution errors.
Certification covers only part of the evidence
Their checker certified a subset with reported precision of 78 out of 85. A value left uncertified is not automatically false, and a correct number alone does not validate the surrounding conclusion.
What you can do
- Read the primary description and its stated limits before planning an evaluation. Open reference ↗
Where the evidence stops
Abstract checked; labels, checker behavior and reported precision were not independently reproduced.
Sources read. Public source text checked November 14, 2026 UTC. No installation, model inference or complete workflow executed.
Sources & further reading
Paper · paper
Misattributed numbers in AI-written research ↗
October 28 submission, revised October 31. Abstract distinguishes unsupported numbers from mismatched experiment, metric or run attribution.
Paper authors · page checked 2026-11-14 · Published 2026-10-28
Coverage note: Abstract checked; labels, checker behavior and reported precision were not independently reproduced.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
A study of AI-written research finds that unsupported numbers often refer to the wrong experiment, metric or run.