AI Use & Student Data

1 Where does the AI run?

At authoring time only, on our systems, before a teacher reviews and assigns the lesson. It matches the problem to a catalogued problem type, drafts a student-and-teacher dialogue, speaks it in licensed synthetic voices, and animates the working.

Nothing runs at playback. The delivered lesson is a standard video file plus a quiz — it plays from a course site, a projector, or a memory stick with no internet at all.

2 What student data do you receive?

None.

There are no student accounts and no logins. The package we deliver makes no outbound network calls whatsoever. It communicates only with your own learning platform, through that platform's standard interface, writing two values: a completion status and a score.

Those records stay in your gradebook. Nothing is transmitted to us, so there is nothing for us to store, sell, lose, or be compelled to produce. No cookies, no trackers.

3 Who checks it before a student sees it?

A machine, then a person, in that order. Before publication an automatic checker recomputes every step of the working and compares the final answer against the answer the problem was set with. On any disagreement the lesson is refused and the publishing script stops with an error.

This has caught real errors: one lesson divided 0.7818 by 204.23 and wrote 0.0038289 instead of 0.0038280, carrying a 0.02% slip into a wrong final answer. Every other stage had passed it. A human then reviews and approves each lesson.

4 Where do the quiz's wrong answers come from?

From 661 hand-written notes recording what learners actually get wrong on each concept — each one a real mistake paired with its correction. A question is that pair turned around: the correction becomes the right answer, the mistake becomes the option beside it.

No distractor is invented by a language model. A wrong answer therefore tells you which misunderstanding the learner is carrying, not merely that they missed one.

5 What does it deliberately not do?

It does not talk to students, adapt to them in real time, or hold a profile of anyone. It cannot answer a question the lesson did not anticipate, and it does not branch — a wrong quiz answer is not routed to a different explanation.

It is not a monitoring or content-scanning product. No learner's name, image, voice or written work is ever an input. It replaces the first explanation, not the tutor.

6 Where does the source material come from?

Every problem carries a provenance record that travels with the finished lesson. A lesson built from your own course material is stamped as yours, delivered to you, and is not added to any catalogue we license to anyone else.

The pre-publication check verifies that stamp survived into the finished file rather than assuming it did. Lessons are authored with AI assistance and disclosed as such.