Evidence that your platform drives learning. Not just engagement.
AffectSense offers a behavioural-evidence layer for your platform. We design and run a pre-registered study with a partner school or college and an independent evaluator. You choose how deep the study goes. The evidence is yours to take to trusts, the DfE and buyers. The method also applies to universities, employers and other learning contexts.
Every study starts with interaction logs. Where a study needs it, consented video-based behavioural measurement can be added under a dedicated data-protection framework. Never emotion recognition.
Engagement is not evidence of learning.
Time on platform, clicks and completion rates show that learners were present. They do not show that learners understood anything. Schools, universities and employers are increasingly asking a more demanding question: does the way people behave on a platform tell us anything about what they learned, beyond the scores the platform already reports?
Observe
Behaviour drawn from interaction: how learners respond to difficulty, seek help, pace demanding tasks and explore. Every rule is written down before data are collected.
Test
Against an outcome the vendor does not control: a pre- and post-assessment designed, administered and marked by an independent evaluator.
Report
Whatever the result. Hypotheses are pre-registered, and a null result from a valid study is reported as a finding.
Choose how deep your evidence goes.
Every design shares the same core: logs only, an independent outcome, pre-registered hypotheses.
Behaviour · start here
Interaction logs only. Tests whether behavioural indices predict independently marked learning beyond in-tool accuracy. Lowest burden on schools.
Research underpinning the indicators: Pardos et al. 2014 (JLA); Sao Pedro et al. 2013 (UMUAI); D'Mello & Graesser 2012 (L&I); Goldhammer et al. 2014 (JEP); Aleven et al. 2016 (IJAIED).
+ Learner voice · add-on
Occasional one-tap check-ins: how is it going, and what would help? The pupil names the feeling; the system never does. Adds a randomised test of whether checking in helps pupils re-engage, and measures meta-affective capability.
Research underpinning the indicators: Rebolledo-Mendez et al. 2022 (IJAIED); du Boulay et al. 2010 (IJAIED); Pekrun et al. 2010 (JEP); Baker et al. 2010 (IJHCS).
+ Self-regulation · add-on
No prompts. Always-available Hint / Easier / Harder / Break tools; we measure how pupils recover after a dip in engagement. Validated by blind replay coding and, optionally, trained observers.
Research underpinning the indicators: Baker et al. 2010 (IJHCS); Goetz et al. 2023 (JEP); Gobert, Baker & Wixon 2015 (Ed. Psychologist); Ocumpaugh et al. 2014 (BJET).
Governance — nothing is analysed that stakeholders have not agreed.
Platform owner
Chooses the design, and the course or unit it runs on.
Independent evaluator
Usually a university. Owns ethics approval and the outcome, decides on optional components, marks blind.
Host institution
The school, university or employer is the data controller and holds the only key linking codes to names.
Learners
Adults consent for themselves; under-18s with their families. Withdrawing never affects access to the course.
First reference study: Break the Web (in partner review)
Our first worked example is a four-lesson primary science unit, from food chains to food webs. Learners predict what happens when an organism is removed from a woodland food web, then see the result; how they respond when a prediction proves wrong is what we measure. The method is not tied to science or to schools: any subject or training domain with an authentic task, designed points of difficulty and an independently marked outcome can carry a study.
Design
Single-arm predictive-validity study.
Outcome
Paper pre/post test, owned and marked by the university.
Critical test
Do behavioural indices add prediction beyond in-platform scores?
Safeguard
Indices sealed and deposited before outcomes are linked.
Validation
Indicators checked against blind human coding.
Target
About 120 consented learners across 2+ institutions.
Validation you can put before any buyer.
Independent & pre-registered
An evaluation you can cite whichever way it falls.
Light integration
iframe/HTML embed, pseudonymous codes, your identity system untouched.
Institution-ready paperwork
DPIA outline, data processing agreement and transfer risk assessment drafted for host institutions.
Speaks your buyer's language
DfE evidence principles for schools, independent assessment for universities, evidence of learning transfer for employers.
A roadmap
Start with Behaviour, add depth when the first result justifies it.
A path to efficacy
The foundation for a controlled, EEF-style trial if validity holds.
A sample evidence report
A redacted example report — question, method and result stated neutrally — will be made available in due course.
For schools & trusts
A free, curriculum-anchored science unit, and a part in better EdTech evidence. Four lessons plus two short paper tests; no extra marking.
You stay the data controller.
No judgements about individual pupils: indices are never shown to teachers, pupils or parents.
We draft the paperwork at no cost.
No admin burden on your team.
DfE & evaluators
Measurement validity first. Efficacy claims only after. Pre-registered on OSF. Sealed-deposit protocol. Instrument checks reported: a failure to create variation is reported as instrument failure, not a null.
Universities and colleges
Universities take part in one of two roles, and sometimes both.
As independent evaluator, the university:
- reviews and grants ethics approval for the study;
- designs, administers and owns the outcome assessment, and marks it blind to all of our outputs, with a proportion double-marked;
- codes anonymised replays of learner activity to validate our behavioural indicators independently;
- receives our sealed, time-stamped indices before linking them to outcomes;
- runs or verifies the pre-registered analysis, and is free to publish the results whatever their direction.
As host institution
The university runs a study on its own online or blended courses, remains the data controller, and students consent directly.
Three roles. No one marks their own homework.
Your platform
Delivers. Hosts the learning and works directly with the host institution or employer.
AffectSense
Observes. Adds the behavioural-evidence layer, computes the agreed indices and seals them.
Independent evaluator (usually a university)
Evaluates. Owns ethics approval, the independent outcome and the final analysis.
Strict defaults for every learner, and stricter still for children.
We collect the least we need, under pseudonymous codes, for school pupils, university students and adult learners alike. We hold no names or free text.
Logs first.
Every study starts with interaction logs. Video-based behavioural measurement is added only where a study needs it, with separate explicit consent and its own data-protection assessment. Raw video is not retained.
Behaviour, never emotion labels.
Indicators are named by what learners do. Where learners report their experience of a task, they choose to do so.
Never about the individual.
No grading, appraisal or performance management, and no special-category or diagnostic data.
UK GDPR first.
Processing agreements and approved transfer safeguards are in place before any learner data moves.
Who is behind the method
AffectSense was founded by Dr Genaro Rebolledo-Méndez, who holds a doctorate in artificial intelligence applied to education from the University of Sussex and has around 30 years of research and applied experience across Sussex, UCL and Tecnológico de Monterrey. His published work on motivation, help-seeking and meta-affective behaviour in intelligent tutoring systems underpins every study design.
Start with one bounded study.
One unit on your platform, measured honestly, evaluated independently. Add depth only when the evidence earns it.
For EdTech platforms in schools and colleges