Research & Evidence
Product decisions derived from research — not a bibliography.
For each framework: the research idea, the design principle it implies, and how ThinkFlow puts it to work. ThinkFlow is independent and not affiliated with these organisations.
Core design principles
Five principles, one learner at the centre.
Pedagogy before technology
Learning objectives drive AI use — not convenience.
Human-in-the-loop thinking
AI supports planning, monitoring and analysis; it never bypasses them.
Scaffolded AI literacy
Capabilities, limits, bias and ethics are learned before high-stakes tasks.
Process visibility
Drafts, logs and reflections are required artefacts.
AI-resistant by design
Tasks grounded in personal experience, local context and live process.
Analogies from AI engineering
Designing for students is like engineering for models.
Context engineering
In AI engineering
Deciding exactly what information, examples and instructions a model sees, so it reasons well.
In the classroom
The teacher curates what students bring into the AI moment — their own first idea, the success criteria, the local context — so the AI responds to student thinking, not a blank prompt.
Where ThinkFlow does it
ThinkFlow's THINK stage: students must write their idea before AI sees anything.
Harness engineering
In AI engineering
The structure around a model — rules, tools, checks and loops — that keeps it safe and useful.
In the classroom
The routines around AI use — boundaries, verification steps, reflection checkpoints — that keep students doing the thinking.
Where ThinkFlow does it
ThinkFlow's AI Boundary and Bot Lab rules.
Evidence of impact
Measure progress from a baseline to the end of the unit.
Take a baseline score at the beginning of the AI unit, then measure with the same short thinking rubric each week for five weeks. The line shows the measured gain from baseline to final score — progress made visible without reducing learning to one final mark.
Progress over time
Thinking made visible
Illustrative class thinking score: one baseline measure at the beginning of the AI unit, then the same short rubric each week. Progress is the measured gain from baseline to final score.
Human Judgement & AI Literacy
Digital Promise
Understand → Evaluate → Use
Research idea
AI literacy progresses from understanding AI, to evaluating it, to using it — with human judgement kept central.
Design principle
Human judgement must remain central in every AI-supported activity.
How ThinkFlow uses it
Every design explicitly separates STUDENT DOES | AI DOES and states an AI Boundary — what AI can and must not do.
Research idea
Learners develop AI literacy by actively exploring, testing and creating with AI — not by passively consuming outputs.
Design principle
Students should interrogate, test, compare and manipulate AI outputs.
How ThinkFlow uses it
All six workflows position AI output as material to investigate — claims to verify, drafts to critique, explanations to compare.
Learning Through Experimentation
Stanford AI Tinkery
Play & Tinker → Create → Shape What's Next
Research idea
Educators learn about AI by playing, tinkering, making and iterating together.
Design principle
Learning design should be experimental and iterative, not form-filling.
How ThinkFlow uses it
The Design Studio lets educators swap challenges and workflows freely, and the model loops BUILD back to THINK: Reflect → Refine → Rebuild.
Learning Sciences
Learning Sciences
Metacognition • Productive Struggle
Research idea
Understanding grows through metacognition, productive struggle, cognitive engagement, learner agency, retrieval, explanation and evidence-based reasoning.
Design principle
Preserve the cognitive work through which understanding develops, and make reasoning visible.
How ThinkFlow uses it
The CHALLENGE stage builds in productive friction; the AI Boundary keeps AI in a deliberate role; reflection prompts capture change in understanding.