Knowledge Universe › School › Learning Engine
Learning Engine · pilot conceptThe School Learning Engine.
School should not only map curriculum — it should demonstrate an evidence-informed learning strategy. This page models how a learning session, a course, mastery, feedback, motivation and oversight actually work together. II itself is not claimed to be proven effective. Every method below is checked against real evidence, or explicitly marked as II's own hypothesis, on the Evidence Panel →.
1 · The learning journey
Eleven steps, not one explanation and a quiz.
Each step is backed by a specific evidence class — see the badge, and the full detail on the Evidence Panel.
Diagnose
Find out what the learner already knows and where the gaps are, before teaching anything new.
Set goal
A specific, achievable next step — not "get better at maths", but "divide by a fraction correctly".
Explain
A clear explanation at the right depth — the "Explain to" idea already live in the Mathematics demo.
Worked example
See a full solution worked through before trying one alone. Evidence A — Barbieri et al. 2023
Try
Attempt it yourself — this is where retrieval starts, not just reading.
Feedback
What was correct, where the error is, why, and what to try next. Evidence A — EEF Feedback
Retrieve from memory
Recall the idea again without looking it up — one of the most reliable study techniques in the evidence. Evidence A — Agarwal et al. 2021
Check error
Find exactly where the reasoning went wrong, not just that it did.
Connect to prior knowledge
Link the new idea to something already understood.
Spaced review
Come back to it again later, not all in one sitting. Evidence A — Carpenter, Pan & Butler 2022
Mastery / support
Either move on with confidence, or get targeted help — never just "time's up, next topic". Evidence A — EEF Mastery learning
2 · Course modes
Four real modes, not one fixed speed.
Regular path
Normal curriculum progression, with spaced review built in — not a single pass through the material.
Catch-up intensive
Targeted repair of specific missing prerequisites, for a short, bounded period. Not simply more homework of the same kind.
Exam preparation
Retrieval + spacing + mixed (interleaved) practice + reviewing your own errors — the combination the evidence supports, not just more past papers.
Advanced / challenge
Transfer to new situations, deeper problems, projects, explaining to others — not more repetitions of what is already mastered.
3 · Mastery
Learned now is not the same as retained later.
Prototype rule: a learner does not progress just because time has passed. Understanding has to be demonstrated — and then checked again after a delay, because the two states are genuinely different.
Example: a learner solves five fraction-division problems correctly today (Learned now). Whether that holds up is only known after a later, unannounced check — a week on, a month on (Retained later). This prototype does not yet run that later check; it names the distinction so a real one can be added honestly.
A commonly used illustrative mastery threshold is around ~80% — shown here as one configurable pilot setting, not a universal scientific constant this prototype claims to have proven.
4 · Student self-regulation
Learning skills, taught through the journey — not a separate lesson.
Each of these already has a home in the journey above: Check for errors → step 8; Explain your reasoning → steps 4 and 6; Connect new to old → step 9; Monitor and Reflect → steps 6, 7 and 11. Evidence A — EEF Metacognition and self-regulation
5 · Feedback model
Four questions, not one word.
Not "WRONG. 0 POINTS. TRY AGAIN." as the whole experience — a worked example, using the fractions demo's own "⅔ ÷ ⅓" question:
- What was correct?
- You correctly flipped ⅓ to its reciprocal, 3.
- Where is the error?
- The final multiplication — ⅔ × 3.
- Why?
- 2 × 3 = 6, over the original denominator 3, gives 6/3 = 2 — not ½.
- What should I try next?
- Redo just the multiplication step, then check by multiplying your answer back by ⅓.
See it live (simplified to right/not-yet): Mathematics demo → Try yourself. Evidence A — EEF Feedback
6 · Motivation / reward
Reward the behaviour that actually helps, not activity for its own sake.
Reward this
Persist · use a better strategy · find your own error · explain · verify · master · help
Not raw activity
Screen time · number of clicks · number of exercises · perfect streaks
Never: casino mechanics, loot boxes, loss aversion, public child rankings, shame, "you are smarter than X%". See why — external rewards for an already-interesting activity can reduce a child's own interest in it (the overjustification effect).
7 · Parent view — concept only
A weekly summary, not a surveillance feed.
Parent role: support goals, routine and motivation — not constant monitoring or doing the homework. No real child data or accounts exist anywhere in this prototype; the box below is a fictional illustration.
Weekly Learning Summary · fictional example, no real child
- Mastered this week
- Adding and subtracting fractions with the same denominator.
- Needs support
- Dividing by a fraction — correct on worked examples, inconsistent on the "try yourself" step.
- Next goal
- Confidently explain why dividing by a fraction makes a number larger.
- Recommended parent action
- Ask them to explain "why" out loud once this week — no need to check the maths yourself.
- Recommended teacher / tutor escalation
- None needed yet — revisit if "needs support" repeats next week.
8 · Control model
Four roles, one boundary: AI does not issue official school grades.
Student
Self-monitoring — sees their own progress map, nothing else.
II coach
Low-stakes feedback and practice recommendations — never an official grade.
Parent
A supportive weekly overview — not a dashboard of every action.
Teacher / human tutor
Educational judgement and escalation — the only role that can assign a real grade or make a placement decision.