Evidence
The learning science RELAY is built on.
RELAY is built from published research rather than assumptions about what AI should do in a classroom. This page sets out which findings shaped which decisions, in plain language, with the studies they came from.
The approach
Four ideas we build on.
The academic names are in the citations below. These are the ideas themselves, said plainly.
Feedback works better without a mark attached
Known as formative assessment. When a grade is attached, students attend to the grade instead of the feedback. Take the grade away and the feedback starts doing the work.
Students can learn to judge their own understanding
Known as self-regulated learning. Support is heavier early and lighter over time, so a student ends up able to tell when they have understood something without an adult asking first.
Students speak up when it feels safe to
Known as psychological safety. Admitting you have not understood something has a social cost in a classroom. Lower that cost and more students say what they actually think, which matters most for the students who speak up least.
The teacher-student relationship is what carries learning
The strongest predictor of how a student goes is their relationship with the person teaching them. So the tool's job is to give that relationship better information, never to sit in the middle of it.
The studies
Every decision, and where it came from.
No grade, no judgement Black & Wiliam, 1998
Formative assessment produces some of the largest measured gains in education, but only where feedback is separated from grading. Attach a mark and students attend to the mark rather than the feedback. Removing the grade is what makes the feedback do any work.
Black, P. & Wiliam, D. (1998). Assessment and Classroom Learning. doi.org/10.1080/0969595980050102
Support that steps back over time Zimmerman, 2002 · Torrington et al., 2024
Self-regulated learning is learnable rather than innate. Students move towards it through support they gradually need less of. RELAY is built to work that way: more guidance early, less as a student shows they can carry the reasoning themselves.
Zimmerman, B. (2002). Becoming a Self-Regulated Learner. doi.org/10.1207/s15430421tip4102_2
The teacher stays in the loop McCaslin & Hickey, 2001
Co-regulation describes the adult support that precedes self-regulation. It is a relationship, not a feature. This is why RELAY routes a student’s confusion to their teacher rather than resolving it alone: the point is to prompt the conversation, not to replace it.
McCaslin, M. & Hickey, D. (2001). Self-regulated learning and academic achievement: A Vygotskian view. In B. Zimmerman & D. Schunk (Eds.), Self-Regulated Learning and Academic Achievement (2nd ed.). doi.org/10.4324/9781410601032
Why the relationship is the target Van Bergen et al., 2023 · Burns et al., 2022
Student-teacher relationship quality predicts engagement, behaviour and achievement. It is also uneven: a substantial minority of students describe their relationship with their teacher as difficult or distant. Making a student’s thinking visible gives a teacher something concrete to respond to.
Burns, E. et al. (2022), Journal of Adolescence. doi.org/10.1002/jad.12093
Guardrails change the outcome Bastani et al., PNAS 2025
In a randomised trial with nearly a thousand secondary students, a carefully designed, guardrailed tool produced a substantial lift in performance. Unrestricted access did not. How a tool is designed to behave with a student decides whether it helps.
Bastani, H. et al. (2025), PNAS. doi.org/10.1073/pnas.2422633122
Students already want real judgement, not a shortcut Microsoft, 2025
In Microsoft’s 2025 survey of US students, 82% agreed that knowing how to use AI effectively and responsibly is important for their future. That is the same distinction RELAY is built around: a guardrailed tool that builds a student’s own judgement, rather than an open one that lets them skip past it.
Microsoft (2025). AI in Education Report. 2025 Microsoft AI in Education Report
Why psychological safety comes first Edmondson, 1999
People only raise problems, admit error and ask for help where they believe doing so will not be held against them. In a classroom that belief is what decides whether a student says “I haven’t got this” or stays quiet, and it is unevenly distributed: the students least likely to speak up are often the ones with most to gain from doing so.
Everything RELAY does is arranged to lower that cost. Nothing is graded, nothing is shared with the class, and nothing a student says counts against them.
Edmondson, A. (1999). Psychological Safety and Learning Behavior in Work Teams. doi.org/10.2307/2666999
See it in action
Exceptional teaching deserves to be seen in action.
The evidence explains why RELAY is built the way it is. The quickest way to judge it is to see what a student puts in and what a teacher gets back.