Our evidence, and its limits.

🧭Novices learn more from explicit teaching than from discovery.

For learners without substantial prior knowledge in a domain, fully guided instruction — clear explanation, worked examples, step-by-step models — consistently outperforms minimally guided or pure-discovery approaches. The effect reverses only as expertise grows (the expertise-reversal effect), which is an argument for fading support, not for withholding it.

Where you can see it: Every lesson opens with a modelled worked example before any independent task, and guided practice offers hints on demand with worked solutions. Support is withdrawn ACROSS THE PHASES of a lesson — modelled, then guided with hints available, then independent with none — and within guided practice the learner decides how much help to take.

Where it is contested: Kirschner, Sweller & Clark is a genuinely contested paper — constructivist researchers argue it characterises inquiry teaching unfairly. We think the novice/expert distinction survives the argument; we do not think it settles every question about inquiry.

🧠Working memory is the bottleneck, so presentation matters.

Working memory holds only a few elements at once. Instruction that splits attention across sources, or forces learners to hold too much at once, spends capacity on the format rather than the content. Managing extraneous load is one of the better-replicated findings in instructional design.

Where you can see it: Phases are short and single-purpose; diagrams sit with the text they explain rather than in a separate panel; a lesson never asks a learner to hold an instruction in their head while doing something else.

🤝Teach in the gap between what a child can do alone and with help.

Vygotsky's Zone of Proximal Development frames learning as social and assisted: the productive target is what a learner can do with support, not what they can already do unaided. Assistance is then withdrawn as competence grows.

Where you can see it: Progressive Diagnostics locate a working level per strand, and the plan starts there rather than at the nominal year level. The included Learning Start consultation exists because the 'more knowledgeable other' cannot be a piece of software.

Where it is contested: ZPD is a powerful framing device rather than a precisely measurable quantity. We use it to justify assisted practice and diagnostic placement — not to claim we can measure the width of any child's zone.

🔁Retrieving beats re-reading, and spacing beats cramming.

Practice testing and distributed practice are the two techniques rated of highest utility in a large review of learning strategies — well-evidenced, generalisable across ages, materials and outcome measures. Re-reading and highlighting, the two things students actually do most, rate low.

Where you can see it: Checks for understanding are retrieval events rather than comprehension re-reads, and a missed item is re-asked once, reshuffled, after the practice phase — a second retrieval attempt, not a re-read. Scored warm-up retrieval from earlier lessons opens the capstone lesson of every unit, Prep to Year 12 — one hundred and one units, three items each. Across the whole program every wrong answer a learner can choose, 3,258 of them, carries the label of the misconception it was written to catch, so a miss is recorded as what was misunderstood rather than only that something was; where the same belief is tested in a later year it carries the same label, so a record shows a misconception recurring rather than several unrelated slips. That was retrieval at one interval. A check question from a finished lesson is now brought back on its own expanding schedule — a day later, then three, seven, fourteen and thirty days if it holds, and back to a day if it does not — six items at most in a day, the ones whose distractor named a misconception first. It sits at the top of the learner's own page, is skippable, gates nothing, and reports to nobody. Whether that changes what any individual child retains is not something we have measured, and we claim nothing about it.

✅Feedback during learning changes outcomes more than marking after it.

Formative assessment — eliciting evidence of learning while it is still forming, and acting on it — is among the better-supported school interventions. Feedback quality matters more than quantity, and feedback about the task outperforms feedback about the person.

Where you can see it: Every lesson states its learning intention and success criteria up front, gives immediate feedback on checks, and closes with learner self-assessment against those same criteria. Feedback is always about the work.

🔤Capable readers do four different jobs, and all four must be taught.

The Four Resources model describes reading as four simultaneous roles — code breaker, text participant, text user, text analyst. A programme that teaches only decoding, or only comprehension, or only critique, produces a reader who is capable in one role and stranded in the others.

Where you can see it: English units and reading diagnostics deliberately cover all four roles, and each is named so a parent can see which job is being taught.

🎛️Design for variability from the start, rather than retrofitting access.

Universal Design for Learning proposes multiple means of engagement, representation, and action & expression as a design discipline. It is a well-developed framework with real practical value; the evidence base for UDL as a whole intervention is still thinner than its adoption.

Where you can see it: Read-aloud on every phase, adjustable text size, reading-support font, high contrast and calm motion are in the platform itself, not an accessibility settings afterthought. Choice tasks are multiple means of expression against identical success criteria.

Where it is contested: We use UDL as a design discipline, not as an evidence claim. Reviews of UDL implementation studies repeatedly note weak study designs. It shapes how we build; it is not something we cite as proof that we work.

🌟Genuine choice raises engagement — that is why choice tasks exist.

Self-determination theory identifies autonomy, competence and relatedness as the conditions for intrinsic motivation. A meta-analysis of 41 studies found that providing choice did enhance intrinsic motivation, effort, perceived competence and task performance — so the basic effect is real and it is measured, not just theorised.

Where you can see it: Choice tasks are assessed against identical success criteria, so no option is the soft one. A learner's chosen interests reorder which option is offered first; they never change the level, the pathway or the expectation — that comes from the diagnostics alone.

Where it is contested: This is the claim on this page with the most awkward evidence, so read it carefully. Patall and colleagues found the effect of choice was STRONGER for instructionally irrelevant choices than for instructionally relevant ones — and our choice tasks are squarely the instructionally relevant kind. The honest reading is that offering a choice of how to demonstrate learning is likely to produce a smaller motivational benefit than, say, letting a child choose the colour of their pen. We keep choice tasks because a modest engagement gain is still a gain, because they serve Universal Design for Learning's multiple means of expression regardless, and because equal-criteria options cost the learner nothing. We do not claim they are a large effect.