The method

The Guide Asks. The Student Thinks.

Two instructors sit with every student: one human, one machine. Neither is permitted to close the question for you.

The pairing

A trained human guide holds the relationship, reads the student's frustration curve, and decides when a question should get harder. The AI companion holds the material: it can reformulate a problem twenty different ways without tiring, track which reformulation the student responded to, and surface the record to the guide.

The companion runs under a strict instruction set. It returns questions, counterexamples, and structure. It does not return conclusions, finished essays, or solved problems. When a student pushes for the answer, it asks what the student expects the answer to look like and why.

Recursive questioning, in practice

1. Claim

The student states a position in one sentence. Vague positions are returned, not corrected.

2. Ground

What would have to be true for that claim to hold? The student lists the load-bearing assumptions.

3. Falsifier

What observation would show the claim is wrong? A claim with no falsifier is sent back as unfinished, not marked wrong.

4. Steelman

The student builds the strongest version of the opposing case before defending their own.

5. Revision

The student rewrites the original sentence. The revision, not the first draft, is the unit of assessment.

6. Recall

The chain is re-entered days later, often through a melodic or physical cue, to test whether the reasoning survived.

Music and outdoor time are scheduled blocks

Instrument practice occupies a fixed block because it is the most reliable daily instance of deliberate practice a child will encounter: immediate error feedback, a clear target, and an unforgiving clock. Time outdoors occupies a fixed block because long attention needs somewhere to go that is not a screen or a desk.

Neither block is reward, therapy, or decoration. Both are on the schedule with defined cognitive purpose, and both are subject to the same measurement as everything else.

How we speak to young learners

Adult vocabulary, with definitions offered when asked rather than pre-emptively substituted.
No mascots, no character narrators, no cutesy framing of difficulty.
Difficulty is described accurately: hard things are called hard, and the student is told roughly how long the hard part usually lasts.
Developmental honesty: we account for a child's immaturity in pacing, rest, and emotional load without pretending their reasoning is less real.
A student is allowed to disagree with a guide in writing, and the disagreement is answered on the merits.

Status of the method

Hypothesis in development

Recursive questioning with an AI companion that withholds answers produces more transferable reasoning than direct instruction with the same content. Falsified if transfer tasks in an unfamiliar domain show no advantage over a matched control.

Open question we have not resolved

We do not yet know the right ratio of guide contact hours to companion hours, or where the companion becomes a crutch. The pilot is designed partly to find that boundary.