Notes

Each note takes one question, runs it against something real — a tool, a datasheet, our own corpus — and publishes the raw result before the conclusion. Every one ends with the section that makes it worth reading: what this test does not establish. Where a note counts something, the count comes from a script in the repository, so the note and the data cannot drift apart without the script saying so.

What this page does not cover

  • Comparative marketing. A note reports what a tool returned on one question on one day. That is not a verdict on the tool.
  • Our own conformance. Written up separately and unflatteringly on the self-assessment.