Teaching across rural classrooms, I kept hitting the same wall: I could see exactly what a student did, which question, which wrong answer, how long they took, and still had no way to see why. Was it forgetting, a misconception, or a missing prerequisite? Every system I had access to recorded the output and threw the reason away.
That gap is the whole problem. A score marks an answer right or wrong; it never reads the process that produced it. The cognitive science needed to read that process, how memory decays, what a retrieval attempt reveals, how one kind of error differs from another, has existed for a century. It had simply never been turned into a live measurement anyone could act on.
So I built Cognivia, an instrument that records how a learner answers, makes a careful guess at why an answer failed (or says it cannot tell yet), and hands back what to try next. It runs in a browser on one shared device, with nothing to install, because the classrooms I had in mind cannot count on more than that.
I locked the trial's design before collecting any trial data, and I report the analysis as it comes, including the parts that go against what I predicted. If the main result comes back null, it stays on the site, and I fall back to the diagnostic on its own, which is still useful whether or not the scheduling idea holds up.
I started it because a question would not leave me alone, and I built it where that question is hardest to answer. If it works, there is plenty to build on top of it, but the science has to come first, or the rest is not worth trusting.
One plain note, so there is no confusion. One person built this, alone, over a gap year in Kanpur, and it still has rough corners. It has nothing to do with any other product or platform that happens to share the name. The protocol and the analysis plan are public, so you do not have to take my word for the design.
Manik Maurya
In brief
I locked the analysis plan before the trial's first student logged in, and I report whatever comes back.
Very few students anywhere get instruction that adapts to them, and classrooms are not going to be rebuilt around each learner. What can change is the few minutes of revision a student does on their own.
So this study asks a narrow question: does adapting the schedule to the type of error a learner makes improve retention? If the answer is yes, it is portable: it runs on a phone or a shared classroom computer with nothing to install. It is also something I could build on my own, from Kanpur.
This trial tests one claim that can be proven wrong: that error-type-aware remediation beats adapting on correctness alone. I report the result either way, with the data and protocol public. A null would mean that specific claim was wrong, not that the work ends, because the learning record and the Passport, the measurement the scheduling is built on, are useful as a diagnostic on their own, whatever the scheduling result.
Check the work yourself
- Pre-registration on OSF (8PJQH): the trial design and analysis plan, filed before any trial data.
- Protocol preprint on EdArXiv: the study written up in full, with no results claimed.
- Study design and method: what it tests, the roadmap, and what it cannot show.
- Decision log: the judgment calls, what each one cost, and what changed.
- One-page summary: the project on a single dated page.
- ORCID 0009-0005-3554-693X