Skip to content
Anchorcite
Accuracy benchmark

How often an anchor actually supports its sentence

We test Anchorcite on a fixed set of questions across disciplines and publish the results, including where it falls short.

A scientist looking into a microscope
The first full run is under way

We will publish anchor precision, reference validity and the share of weak anchors by discipline here, with the full question set, every answer and every rating as a dataset. Until then, this page shows no numbers.

Anchor precisionShare of anchored sentences whose passage fully supports them.
References that existShare of references that resolve in Crossref or OpenAlex.
Weak anchors shownShare of sentences we flag as only partly supported.
Benchmark questionsQuestions in the fixed set, across ten disciplines.

Method

  1. 01Fixed question setQuestions written by researchers across 10 disciplines.
  2. 02Blind answersAnchorcite answers each question with no tuning to the set.
  3. 03Expert ratingTwo domain experts judge each sentence against its passage.
  4. 04Published in fullQuestions, answers and ratings are released as a dataset.