How it works

A tutor that knows exactly where your child is, and refuses to pretend

The loop below runs every session. It is deliberately unglamorous: the value is in the diagnosis and the verification, not in the explaining.

The loop

Five steps, every session

  1. 01

    Diagnose

    Before teaching anything, probe the prerequisites. If a prerequisite is shaky, the session drops down the graph instead of pushing forward. This is the step ordinary tutoring skips and school cannot afford.

  2. 02

    Teach

    An authored explanation is selected for the learner’s reading level and delivered conversationally. The tutor is grounded in that authored material and is instructed to say when it does not know, rather than improvising facts.

  3. 03

    Practice

    A best-in-class leased engine supplies adaptive daily items in Phase 1. Freedom ingests that accuracy signal, requires shown work, and owns every layer that determines whether the concept truly counts.

  4. 04

    Check

    A rubric grades the reasoning, not only the result. A correct answer arrived at by a broken method is recorded as not mastered, and the specific misconception is logged by name.

  5. 05

    Advance or remediate

    Mastery is updated, spaced retrieval is scheduled, and the next node is chosen. Three stuck sessions on the same concept escalates to the parent instead of grinding.

Mastery

Four gates before a concept counts as learned

If mastery is fake, everything built on top of it is fake — the daily plan, the transcript, the reassurance you feel. So mastery is expensive to claim on purpose.

Item accuracy

At least 90% in controlled adaptive practice. Necessary evidence, but never sufficient by itself.

Delayed retrieval

Correct again days later, cold, without a warm-up. Recall on the same day proves nothing.

Produced work

A derivation, an explanation, a working artifact — something a skeptical adult can read.

Oral defense

Explain the reasoning out loud and answer live follow-up questions.

Streak counters and completion percentages are deliberately absent. Both are trivially gamed and neither predicts whether a child will still know something in six months.

Independent proof

Our own mastery score is not allowed to grade our school

The graph proves daily understanding. Nationally normed external tests provide a separate view of achievement and growth, so we can detect blind spots and eventually test whether learners are moving faster than comparable peers.

Baseline in the first 14 days; sparse growth checkpoints; annual state-appropriate assessment; PSAT/SAT or ACT when useful. We report achievement and growth separately and never turn one child into a claim about every public school.

See the parent dashboard

The obvious objection

Your kid has a phone with a better model on it than ours

Any system that assumes children will not use AI to shortcut the work is designed for a world that no longer exists. This one assumes they will try.

  • Credit is for reasoning, not answers

    There is nothing to gain by pasting in an answer, because answers are not what is scored.

  • Live follow-ups

    Oral defense generates questions from what the child just said. Borrowed reasoning falls apart in one follow-up.

  • Some assessment is closed-tool

    Observed, offline, periodically. Enough to keep the mastery numbers honest.

  • Flag, do not accuse

    Suspicious patterns route to a parent for a conversation, never to an automatic punishment.

The message to the child is straightforward rather than punitive: using a model to skip understanding is stealing from yourself, and this system is built so that shortcut does not produce credit.

Guardrails

Where the AI is not in charge

It teaches from authored material

Explanations, examples, and problems are written and versioned by a human, and the tutor retrieves from them. Free-associating facts at a child is the failure mode we are most careful about.

Every transcript is visible to a parent

There is no private channel between a model and a child. Younger learners get a bounded subject domain rather than open-ended chat.

Humans own the hard subjects

Ethics, meaning, conflict, and anything contested belongs to parents and mentors. The software's job is skills and knowledge, and it should be honest about that boundary.

Learner data stays put

A child's learning history lives in our own database, is never committed to source control, and is not fed to third-party analytics.