Outcomes

What we measure — and what we don't claim.

We don't publish efficacy claims yet — here's what we measure instead. Amistio Learn is in early access, and we have no independently verifiable outcome data. Until we do, you won't find a made-up statistic on this page, or anywhere else we write.

No invented numbers

Education products love a big percentage. We think a number you can't verify is worse than no number at all — so until we have real, honestly collected outcome data, we publish none: no learner counts dressed up as results, no cherry-picked success stories presented as evidence, no "proven" language.

For now, the absence of numbers on this page is the claim.

What we measure instead

Instead of advertising outcomes we can't prove, we instrument the thing we can control: whether the teaching itself holds up.

Real work, graded against rubricsA Journey isn't a feed of chat replies — it assigns real practice and homework. Every submission is graded against an explicit rubric, so feedback points at specific criteria instead of a vague “good job”.

Deterministic checks where the subject allows themWhere an answer can be checked mechanically — code that must run, an exercise with a verifiable answer — we prefer the deterministic check, so the grade isn't just a model's opinion.

Internal evals on how Ami teachesWe maintain an eval suite that runs Ami's teaching behavior against golden cases before changes ship: teach-back moments, lesson recaps, teaching moves, misconception probes, lesson notes, explanation quality, curriculum planning, and grading consistency, plus contracts that keep structured outputs well-formed.

What we commit to

Real outcome data, when it existsWhen we have outcome data worth publishing, it will appear here together with how it was collected, so you can judge it — and we will never invent numbers in the meantime.

The research we build on

The teaching loop is built on published learning science — retrieval practice, spacing, feedback, and the one-on-one tutoring research — and our blog covers it with citations to the actual studies.

Bloom's 2-sigma problem The science of learning blog

Judge it yourself

The honest pitch: try a session, submit real work, and see whether the feedback earns your trust. Free in early access — see pricing.

Start free