Cognivia · Cognitive diagnostics for learning
Every mindforgets differently.
A learning-diagnostics system, built on cognitive science. It measures what a learner's memory is doing behind an answer, and hands back a reading any teaching tool can act on.
What Cognivia records
Cognivia records what a learner answered, how long it took, how sure they were, and how that changes across repeated checks, then organises it into a testable working hypothesis about one concept. It does not read the mind, and thin evidence stays uncertain.
It is an independent research project, built for students and teachers in rural government schools. It leaves scheduling, grading, and lesson content to the tools a class already uses, and keeps observation separate from interpretation.
The rule that keeps it honest: observations and interpretations stay distinct, thin evidence abstains, and any response is checked later rather than assumed.
Why memory fades
Why review timing matters.
Memory fades along a steep, well-studied curve, and each review resets it and slows the next drop, a finding scientists have mapped for well over a century, yet almost no classroom acts on. Cognivia uses this to decide when a concept is worth checking again. Drag the review points below to see the effect.
Where memory becomes measurable
See what the answers show.
Every answer, how long it took, and how sure the student felt is recorded. Across repeated checks these build a picture of which concepts are slipping, a signal an exam score hides. A single session is only a first look, not a profile.
Read what kind of wrong it is.
From the timing and the confidence behind an answer, the confirm-or-clear diagnostic works out which kind of wrong it is, then routes it to the fix that kind needs.
Let the diagnosis pick the next move.
Show a misconception a counter-example that breaks it. Send a fading memory back for retrieval before it goes. A missing prerequisite means stepping back a level first, then returning. Each move is checked at the next review, so a fix that did not land gets caught instead of assumed.
A system, not a feature
A running record, not a one-off score.
The record updates across sessions, keeps a separate view per concept, and carries forward instead of resetting at term's end.
So a teacher meeting a student in a new year starts from what is already known about each concept, what is holding, what is slipping, rather than a blank slate.
Four kinds of wrong
Four ways to be wrong. Each one needs a different fix.
How long a wrong answer took and how sure the student was help sort it into one of four commonly-distinguished kinds. Whether two trained raters agree on the kind, and how often, is one of the things the study is set up to test; it is not assumed.
Plain forgetting
Looks like: held it before, now slower and less sure.
Fix: retrieval, timed to how fast this idea is slipping.
A missing prerequisite
Looks like: fails on something it depends on but never held.
Fix: step back to the prerequisite, then return.
A half-formed grasp
Looks like: right sometimes, hesitant, breaks when the wording changes.
Fix: worked examples to consolidate before testing.
A confident misconception
Looks like: fast, certain, and wrong, the same wrong answer each time.
Fix: a contrasting counter-example, never more drilling.
Cognivia · The diagnostic engine, live
Watch it read five
answers, and diagnose.
Answer five questions and watch the engine work: it sorts each error by type and builds a learning record right in the browser, updating as you go.
Research
Most learning tools open with their best number. Here are ours, before the trial has run.
Cognivia is tested in a pre-registered, three-arm randomised trial. The design, the analysis plan, and the data are public before the results are known. No retention number or speed multiple appears here, because the trial has not produced one yet. When it does, it goes up whether or not it helps the case.
Every formula and threshold Cognivia runs on is written down and public, so you can read the method instead of trusting a black box. The science is named: Ebbinghaus (1885) on the shape of forgetting, and Wixted & Carpenter (2007) and Bjork & Bjork (1992) on how memory strengthens and fades.
Registered as a Stage 1 pre-registered protocol; the analysis plan was locked before any data was collected. Every link above opens a real, public document you can read right now.
A student is not a 78%.
A student is a particular forgetting rate on photosynthesis, a stable retrieval strength on chloroplasts, a fatigue threshold at seven minutes, and an active misconception about chlorophyll that needs to be flagged before tomorrow.
Know what to
check next.
Walk one teacher decision end to end, the evidence, the choice, and a later check, with synthetic learners. Free for students, built in Kanpur, tested in the open.