What it actually is
It records how a learner remembers.
How it works, part by part. Cognivia records the timing and confidence behind every answer, weighs whether a miss looks like plain forgetting, a misconception, or a missing prerequisite, and keeps a running record a teacher can act on, all on one shared classroom computer while a learner just answers questions.
In sixty seconds
In short: Cognivia watches how you answer, works out how quickly each idea is slipping for you in particular, does something about it before it goes, and then checks whether that something actually worked. The sections below cover that loop one step at a time.
Inside every session
Each feature below comes from a finding in the research.
- Four-option recall with diagnostic distractors. Answering from memory is itself practice (Roediger & Karpicke, 2006), and each wrong option is written to match a known kind of error, so the choice carries information (Brown & Burton, 1978).
- Review timed to your own decay estimate. Forgetting follows a regular curve (Ebbinghaus, 1885; Murre & Dros, 2015), and spacing reviews out improves later recall (Cepeda et al., 2006).
- Confidence rated on every answer. A confident wrong answer is a different problem from an unsure one and needs a different correction (Metcalfe, 2017).
- Every attempt logged automatically. How long an answer took is part of the reading (Heitz, 2014), and the log is what lets a later check say whether a response actually helped.
One full loop, every session
Several things happen inside every Cognivia session, all running automatically in the background. Each one is grounded in research on how memory works. Together they form a complete loop: the system reads how each learner is doing, tells a wrong understanding apart from an incomplete one, decides what to do about each kind of failure, and records every step so its own recommendations can be checked against what actually happened. It runs in the browser on a single low-cost computer.
A named, defined error taxonomy
As answers come in, it weighs the evidence for plain forgetting, a half-formed grasp, a missing prerequisite, or a confident misconception, and says so when the evidence is too thin to tell. Each type has a written definition, so two independent raters can check the same error against it. Whether they actually agree is one of the things the study measures. This is the instrument's core.
Scheduling comes after the diagnosis
One thing the diagnosis lets you do: time each review against the student's own error-type-specific decay curve, with a separate forgetting estimate for a fading memory and for a shaky grasp. Most spaced-repetition tools schedule against a single curve per item; splitting it by error type is the part that is ours.
Every answer is logged
Every response, its timing, and its session are logged automatically. For learners enrolled in the study, with consent, that log is the pre-registered research dataset; ordinary use is kept separate.
The learning record
Ten variables that describe how you remember.
They describe how you remember, which marks alone do not show. You don't need a single symbol to use Cognivia, they're the system's shorthand for noticing how you learn, how fast you forget, when you get tired, and which kinds of mistakes you repeat, so it can bring each thing back right before you'd forget it. The letters are for the researchers; the studying is for you.
Seven of these read from a single session. The last three, spacing benefit, automatic recall, and bounce-back after a miss, need several sessions before they mean anything, so they fill in over time rather than on day one.
One session is a first look. Over many sessions, each topic in your subject builds up its own λ, its own RS curve, its own consolidation efficiency, each with its uncertainty shown. That is how the record could come to show that chemistry fades faster for you than biology, or that accuracy drops late in a session. These are estimates the study is still testing.
How to read the math
These formulas and their cutoffs are Cognivia's own design choices, not results borrowed from the papers cited across this site. The decay form draws on Ebbinghaus and on Wixted & Carpenter (2007); the storage-versus-retrieval distinction on Bjork & Bjork (1992); the confidence and error-type signals on the metacognition literature. But the exact expressions, the smoothing coefficients (α = 0.75, α = 0.30), and every threshold (λ ≤ 0.7, CE > 0.85, FSI > 0.20, PDS > 0.6) are engineering defaults we chose, seeded from the n = 5 pilot and the literature. They are how the system computes today. Calibrating these thresholds against a real student population is one of the things the trial is for.
The Cognitive Passport
What the Passport is meant to show.
A percentage says what a student got right on a given day, and a percentile says where they ranked that day. Neither reaches how memory forms, how knowledge consolidates, or how someone thinks when no one is watching. The Cognitive Passport is a first attempt at answering those questions with evidence gathered continuously as a student works, where a transcript is a single impression at term's end. It won't replace other assessment, and it is not ready to sit beside a grade.
In practice it is a page per learner: a forgetting estimate for each topic, the mistakes that keep coming back, and where in a session accuracy tends to drop, each with its uncertainty. It is exploratory, so it is not meant for admissions, grading, or any other high-stakes decision.