Understanding mathematical attention and judgement.
The public idea is simple: notice, measure, question and decide. The research challenge is harder: can we observe meaningful development in those capacities without teaching learners to perform for the measures?
Learning beyond the final mark.
A correct answer matters. It does not tell us whether the learner anticipated what was plausible, recognised an error, used feedback well, understood the limits of evidence or could transfer the idea somewhere new.
Anticipates sign, range, graph shape or scale before calculation.
Inspects units, assumptions, plausibility and interpretation.
Confidence becomes increasingly aligned with demonstrated capability.
Locates the first invalid step and chooses a useful next move.
Recognises familiar mathematical structure in a new context.
States what the evidence supports, what it does not, and why.
Data can direct attention. It cannot know the whole story.
REAL Analytics has developed from learner-facing personalised feedback into a connected system combining assessment, engagement, self-regulation and student voice.
Dialogue With Self and System
Learner reflection and system evidence are read together. Data is mirrored back to support agency, equity and informed action — not used as a substitute for the learner’s own account.
89% recall
The 2026 HEA exemplar reports identification of 89% of off-track trajectories by Weeks 5–7.
19% self-identify
Students can ask for support before a predictive model flags them.
Learner-facing
The richest analytical view belongs to the learner: trajectory, reflection and confidence are visible to them.
Assessment can become part of the learning.
The design work treats a question as more than a mark-producing object: it can carry mathematical intent, known misconceptions, feedback, retry behaviour and evidence about what happens after difficulty.
Design once. Learn from every response.
Question families can be mathematically verified, parameterised and linked across practice, feedback and later transfer.
Human judgement stays visible.
Automated practice can secure foundations. Interpretation, modelling choices, uncertainty and defence should not be automated away.
If AI helps, what happens when it leaves?
AI can explain, hint and critique. The stronger educational question is whether the learner later reasons independently when that scaffold is withdrawn.
The same discipline of attention travels.
The diabetes work examines continuous glucose monitoring as a temporal system: not simply a single metric, but patterns, risk and change across rolling windows. This strand provides a second domain in which models must remain accountable to the phenomenon they describe.