Science Slop Index

Paper-level scientific slop

Science, or slop?

AI text detectors read sentences. The Science Slop Index reads how a paper holds together: whether its sections build on each other, its claims are argued, and its evidence can be inspected.

Structure

Do the sections build on one another?

  • Cross-section references
  • Macro redundancy

Argument

Are claims and citations argued, not just stated?

  • Argument graph
  • Citation isolation

Artifacts

Can a reader inspect the method and the evidence?

  • Figure exposition
  • Evidence gap

How it works

Six measures, three planes, one number

Each measure counts the share of a paper's units that show one pattern. The index averages the measures within each plane, then averages the planes, and reports the result out of 100. Higher means more of the patterns that set AI-generated papers apart from matched human papers.

It separates AI from human papers

As reported in the paper: pair accuracy on SciSlopBench, 390 AI-generated papers each matched with a human paper on the same problem. 0.5 is chance.

What happens to your paper

Your file is deleted when the analysis finishes. The report gets its own link to share.

  1. Read. LaTeX source is read directly; arXiv links fetch the source. PDFs are rebuilt into sections, captions, references, and citations.
  2. Measure. Four measures are exact counting rules. Argument graph and Figure exposition ask a language model to label sentences and read the method figure.
  3. Locate. Every flagged unit is listed where it occurs, so each number can be checked against the paper.
MeasureUnitCountsPair acc.*

* Pair accuracies are the paper's results for its own implementation. This site re-implements the measures; see the notes in each report.