Developer PreviewPhilosophy

A model can know the answer without being a good tutor.

Tutor Benchmark asks what a model does with a learner’s actual state—not only whether it can produce a correct answer.

“A model can know the answer without being a good tutor.”

That is why this project focuses on observable tutoring behavior: diagnosis, guidance, adaptation, and the learner’s next action, alongside correctness.

Independent by design

Benchmark, not tutor product.

Tutor Benchmark evaluates a TutorUnderTest. It is not a chat application, a prompt playground, an admin dashboard, or a Review Workspace module. Provider-specific behavior enters through an adapter boundary.

Inspect the source ↗

Current release posture

Website
Developer Preview
Package
v0.1.0 published
Dataset
tutor-eval-v0.2a@0.2a.6
Leaderboard
No calibrated public runs
License
Apache-2.0 / CC BY 4.0 / Brand Policy

Public boundary

The site is read-only and secret-free. It packages public development metadata, not credentials, raw provider payloads, private calibration files, production conversations, or hidden reasoning.