The Open Measurement Checklist

Five questions every district should ask every education-AI vendor

Quantum Learning Machines · v1.0 · July 2026 · CC-BY-4.0 — share freely

Any vendor claiming their AI “works” should be able to answer yes — with links — to all five. “Trust us” is a no.

Can we inspect the model?

Are the weights (or a meaningful evaluation surface) available for independent testing — not just a scripted demo?

Good looks like: Downloadable weights or a public evaluation endpoint.

Does the model card state what the system cannot do, with numbers? A card with only strengths is an advertisement.

Good looks like: Published error rates and failure modes, even unflattering ones.

Ask specifically how contamination was controlled: held-out data, redacted detection signals, versioning.

Good looks like: A public benchmark with a documented held-out split — and the vendor's actual scores on it.

What was it trained and evaluated on, under what licenses and consent? Student data handled under what agreements?

Good looks like: A data statement you could hand to your privacy officer.

If a researcher wanted to check the headline claim, is everything they need public?

Good looks like: Open evaluators, open data, versioned results — reproduction possible without the vendor's permission.

Why we publish this: we hold ourselves to it. Our models, cards, evaluation data, and honest numbers are public at play.quantumlearningmachines.com/developer and huggingface.co/QuantumLearningMachines. Ask us all five — and ask everyone else too.