Full Report
Everything behind the charts, in seven sections
The charts on the Benchmark page are the short version. This report has the full argument, method, tables and caveats. Read it section by section here, or download it as one PDF.
Download the full report (PDF)
01
Background
Why the women of the early church are a hard test for AI models.
02
Known LLM Biases
Hallucination, androcentric framing and flattened scholarly debate.
03
Methodology
The four-dimension rubric, the question bank and how annotators scored.
04
Data Collection
How the 210 responses were collected, and how to repeat it.
05
Results
Rankings, dimensions, figures, adversarial items and error tags in full.
06
Inter-Annotator Agreement
Krippendorff’s alpha under a ceiling effect, and how to read it.
07
Caveats
Scope, ceiling effect, single runs and unblinded annotation.
Back to top