Home / Science

A century of evidence

Forgetting is measurable,
and so is why.

Memory decays on a curve, timing and confidence carry signal, and one kind of error can be told from another. A century of cognitive science established all three. Cognivia's work is not to restate it but to turn it into a live measurement of what a learner is doing, and why they fail. This page is the evidence that measurement rests on.

What nobody can see

Nobody can see what a mind is actually doing.

Over a century ago, Hermann Ebbinghaus showed that forgetting follows a curve: without reinforcement, most newly learned information disappears within days. Forgetting, in other words, is lawful. If it is lawful, it can be measured, and so can the things around it, the confusion, the confident wrong answer, the moment a shaky grasp turns solid.

Yet almost nothing in a normal learning environment measures any of it. Information is delivered, practiced, tested, and scored. The score records the output and throws away the process: it never says why an answer was wrong, or what was about to be lost.

Cognivia began with a smaller, more stubborn question: what if you could read that process directly, while it is still happening, instead of only marking the answer right or wrong once it is over? Everything on this page is the evidence that you can.

The rough sizes are not in dispute. Left unreviewed, a large share of new material, on the order of three-quarters of it, is gone within a week, close to the figure Ebbinghaus reported and to a direct replication of his curve in 2015 (Murre & Dros). Testing yourself, rather than re-reading, holds on to far more of it (Roediger & Karpicke, 2006). And a hundred and forty years after Ebbinghaus first plotted the curve, most classrooms still schedule as though it were never measured. Cognivia's wager is narrow: that those numbers become useful only once you can read them for one learner, in real time, rather than as population averages.

What makes it measurable

Three findings that turn learning from a black box into something you can read.

Each one is a reason a learner's state can be measured at all. This is the evidence the measurement rests on.

01

Forgetting follows a curve

Memory fades on a predictable curve, and each retrieval resets it. That regularity is what makes forgetting measurable: from a learner's answers over time, Cognivia estimates a personal decay rate per concept, with uncertainty attached, rather than assuming a fixed pace. The estimate is a measurement, and it carries its own error bars.

Each concept carries its own decay rate and sample size, so the model knows how much it actually knows, and says so plainly when the evidence is thin.

02

A retrieval attempt is a measurement

Reconstructing an answer from memory is the cleanest reading of memory state there is. When a learner retrieves, three signals appear at once, whether the trace held, how long it took, and how confident they were, that passive re-reading never produces. Cognivia records all three on every answer and reads the mechanism behind a miss from them.

The answer is revealed only after an attempt, so every response measures what a learner can actually reconstruct, rather than what looks familiar.

03

Feeling isn't knowing, and the gap is readable

Familiarity feels like mastery. Re-reading looks productive because the material seems familiar, but recognising something is a long way from being able to reconstruct it. Because Cognivia measures at the moment of retrieval, under real difficulty, it sees the gap between feeling and fact directly, the confident wrong answer a re-read would have hidden. That gap is the most useful thing on the page.

See it work on your own memory.

A three-minute demo session, no sign-in needed.

Try a session See the system →

References

The work this rests on.

Every model and number on Cognivia is tagged to a source below, verified against the primary paper. The phenomena are established science; the exact formulas and thresholds Cognivia runs are its own design choices, and it says so.

  1. Ebbinghaus, H. (1885). Über das Gedächtnis. Duncker & Humblot.
  2. Murre, J. M. J., & Dros, J. (2015). Replication and analysis of Ebbinghaus' forgetting curve. PLOS ONE, 10(7), e0120644.
  3. Wixted, J. T., & Carpenter, S. K. (2007). The Wickelgren power law and the Ebbinghaus savings function. Psychological Science, 18(2), 133–134.
  4. Bjork, R. A., & Bjork, E. L. (1992). A new theory of disuse and an old theory of stimulus fluctuation. In From Learning Processes to Cognitive Processes (Vol. 2, pp. 35–67). Erlbaum.
  5. Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95(4), 492–527.
  6. Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning. Psychological Science, 17(3), 249–255.
  7. Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks. Psychological Bulletin, 132(3), 354–380.
  8. Heitz, R. P. (2014). The speed-accuracy tradeoff. Frontiers in Neuroscience, 8, 150.
  9. Metcalfe, J. (2017). Learning from errors. Annual Review of Psychology, 68, 465–489.
  10. Brown, J. S., & Burton, R. R. (1978). Diagnostic models for procedural bugs in basic mathematical skills. Cognitive Science, 2(2), 155–192.
  11. Ackerman, P. L. (Ed.). (2011). Cognitive Fatigue. American Psychological Association.
  12. Gelman, A., et al. (2013). Bayesian Data Analysis (3rd ed.). CRC Press.
  13. Free Spaced Repetition Scheduler (FSRS-6), open-spaced-repetition. Power-law forgetting curve; retrievability = 0.9 at elapsed time = stability.