Most aviation training stops at recall.

You set the depth

Five rungs, and you choose where to stop.

Plenty of aviation material goes deep. What nobody does is hand the depth control to the learner — because depth is the industry's upsell, and giving it away contradicts the business model. Every concept is written as rungs you climb only as far as you care to.

Depth and purpose are separate dials. Point it at a commercial oral and it drills the band that ride actually examines — it will not quiz you on rulemaking history. The deeper rungs stay there to read whenever the question is bugging you at eleven at night.

A student wants d1. A CFI candidate needs d3. Somebody who just wants to understand their own aircraft goes to d5 on the four topics they care about and stays at d1 everywhere else. Same corpus, three different products.

The tutor

It withholds, then teaches.

The obvious objection to all of this is: why not just ask ChatGPT? There is a real answer. In a 2025 PNAS study, Bastani et al. found students given a plain chat interface scored 17% worse on the exam than students who had never used AI at all — once the AI was taken away. A scaffolded tutor erased the harm entirely. An assistant that answers well is not the same as one that teaches.

So it does not open with the answer. It escalates hints, names the misconception it thinks you are carrying, and then gives you a model answer. That last step is the one oral-sim tools skip: staying in character while a learner flounders is a quiz, not teaching. An examiner finds out what you know. This is the other job.

Hints are counted separately from questions, so needing three of them is recorded as needing three of them. Answering cold and dragging yourself there are different events, and what you see next week reflects which one it was.

References

Every rung names its source.

The objection pilots actually voice about AI study tools is not that they are shallow. It is that they are confidently wrong, and that a learner without the background cannot tell the difference. Fair. It is the thing this was designed around first.

Every rung carries the regulation, advisory circular or handbook section it rests on, so you can go and read the source rather than take our word for it. Behind that, each supporting record tracks its own provenance — where the claim came from and who stands behind it — and a claim carries that history with it for as long as it exists.

The contrast worth drawing is with the incumbent. Sheppard Air's method is built on the Primacy Law — get the right answer in first so it sticks. That is from the same FAA handbook that names Bloom's taxonomy. They quote the FAA on how to make an answer stick. We are quoting it on what level you should reach.

It knows what it doesn't know

Two of the five answer buttons are about us.

When you grade a question, three of the outcomes are about how you did. The other two are about how we did. Ask CFI flags a point that wants a human instructor and hands it to yours. Gap records that the material came up short — logged against us, not you, and routed to whoever writes that topic.

A study tool with no way to say I don't know will eventually teach you something wrong, convincingly, because it was engineered to make things stick. This one has somewhere to put that.