For researchers
Learning studies rarely follow real coursework for long: a laboratory can keep participants for an hour or a week, seldom a term. Cogniloop is used across a whole course, by learners studying for their own exams, and it keeps a detailed record of how they work. We want that record to be useful for research well beyond our own questions.
What the record holds
- Every attempt and each line of working, with the line where the reasoning stopped holding and whether the mistake was a slip or a gap, and how each grade was reached: by computation, by a model, or by a model held to what computation proved.
- How handwritten work was written: strokes, pauses and crossings-out, kept as numbers once the pen strokes themselves are deleted.
- Optional background: level, field and year of study, and the last grade in the course's subject.
- Every hint, in order and by strength, so “solved with help” stays apart from “solved alone”.
- Every spaced review, when it was due and how it went, read from the work rather than a self-rating.
- The engine's prediction before each outcome, such as “will solve this alone”, so its predictions can be scored against what happened.
- Exam dates and study plans, so behaviour can be read against a real deadline.
Bring your own question
The record is not collected for one study. These are examples of questions it could reach that short studies cannot:
- How help-seeking changes over a course, and whether asking for less help predicts doing better later.
- Where reasoning breaks: whether many learners break at the same step, and how that differs between topics and subjects.
- How forgetting looks in real coursework, not word lists, and how well a scheduler like FSRS fits it.
- How learners use an AI tutor when the answer is one request away: when they ask for it, and what happens to them afterwards.
- Learner modelling: how well a model of what a learner can do alone predicts their next attempt, scored against a simple baseline.
If you have a question we have not thought of, we would rather hear it than guess.
Data from consenting learners
Learners choose at sign-up, and can change later in Your data, whether their data may go to outside researchers, with the switch “Share my pseudonymised data with outside researchers”. It is off by default, tutoring works the same either way, and they can withdraw at any time. We plan to share research data on these terms:
- Only learners who switched research use on, with the wording and date of what each one agreed to kept on record.
- Pseudonymised: names, email addresses and account details removed and replaced with random identifiers.
- Only what your question needs, after your ethics approval, under a written agreement that rules out re-identifying anyone and passing the data on.
- Learners who withdraw are left out of anything shared after they do.
- The record is the learners' work and what came of it. How the engine reaches its decisions is not part of it.
See the privacy policy for how learners' data is handled.
Ways to work with us
- Analyse the record for your own question, once enough learners have opted in.
- Run the comparison study. The working paper sets out the claim and how it could fail: learners who use Cogniloop should solve more new problems on their own 7 and 30 days later than learners who use a general chatbot or study their usual way, with the outcome and analysis fixed before any data is collected. We would rather an independent group ran it.
- Run a study inside Cogniloop. For example, comparing two kinds of help with learners who agree to take part. This is planned, not built, and would need consent for that study and an ethics review.
Where we are
No research use has happened yet, and none will without consent and an ethics review. There is no dataset to share yet: Cogniloop is in invite-only early access, and so far its real data comes from one learner. There are no learning-outcome results. The record grows as testers join, and working with us now means helping decide what it should capture.
Get in touch
If you study learning, tutoring, memory or AI in education, write to hello@cogniloop.app with your field, your question and your ethics process.