Scope and methodology
Scope and methodology
What was checked
For each claim, the analysis identifies the assumptions, constructs an example and works through the contradiction. It then asks which later results depend on that claim, whether a repair is possible and whether the implementation needs to change. A failed lemma alone does not overturn a paper’s experimental results.
Classification
Counterexample: an explicit construction contradicts the selected statement as written. Addressed: the issue has relevant treatment in a later version of the paper or a separate work, public discussion or code source. Material already present in the original publication, including its appendix or original supplement, does not qualify. A separate source may have the same authors. These entries identify the treatment and link to it; discussion alone is not a formal correction.
Review procedure
All analyses and written explanations were produced by AI, including the cross-checks by separate AI agents and the code used to reproduce calculations. They have not been independently verified by human reviewers or formally verified. Authors have not been contacted.
Selection and citation counts
I used AI to examine selected arguments in 334 papers published from 2015 onward, each with at least 300 recorded citations. This page shows 40 of them; the screening did not cover every proof. I’d welcome anyone who wants to help check these analyses, challenge a counterexample or contribute a correction.
Citation counts are dated snapshots, not live totals. Each count links to the corresponding Semantic Scholar record; the paper title links to the primary source.
Correction history
Addressed issues are kept separate from the detailed counterexamples. The linked material may resolve the issue, restrict the statement or discuss the relevant limitation; each entry specifies which. Failure to retrieve a correction for a detailed analysis does not establish its absence or the priority of an observation.
Suggest a correction with a source ↗