How Kematra decides

Kematra makes decisions about your studying, so you are entitled to know where those decisions come from. Some of them rest on well-replicated research. Some are our own inference from it. Some we invented.

The short version: the directions are research-grounded and the magnitudes are ours. That spacing beats cramming is not in serious doubt. The specific interval Kematra picks for you is a guess we made, bounded so it cannot do anything silly.

Supported by research

Directions with real published support behind them. The citations are for the direction, never for our numbers.

Kematra asks you to retrieve things from memory — answering, working a problem, explaining — rather than rereading them.
Retrieving something from memory produces better long-term retention than studying it again for the same amount of time. This is one of the better-replicated findings in the area, and practice testing was rated high-utility in a large review of study techniques.
Roediger, H. L., & Karpicke, J. D. · Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T.
It spreads your work on a topic out instead of massing it together.
Spacing repetitions apart produces better retention than the same repetitions close together, across a large body of studies. Distributed practice was also rated high-utility in the same review.
Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. · Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T.
It schedules a review further out when a topic is holding, and sooner when it is not.
Retention declines with time since study, and the useful gap between repetitions grows with how long you need to remember something. That the gap should scale is supported; the curve we use to decide *when* is ours.
Ebbinghaus, H. · Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H.
It mixes topics within a session rather than doing one topic to exhaustion.
Interleaving related problem types improves later performance compared with blocking them, at least for material where telling problem types apart is part of the skill. The evidence is strongest in mathematics.
Rohrer, D., & Taylor, K. · Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T.

Reasonable inference

Follows from supported findings, but has not itself been tested. Honest extrapolation.

It treats a check you marked yourself as weaker evidence than one checked against a mark scheme.
People are generally poor judges of their own learning, and fluency during study is a misleading signal. It follows that a self-judgement should count for less — but *how much* less is not something the research settles, and our weighting is a construction.
Bjork, E. L., & Bjork, R. A. · Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T.
It makes a topic's difficulty rise when you struggle with it and fall when you do not.
That difficulty varies by person and topic, and that recent performance is informative about it, is uncontroversial. Treating it as a single number that moves by a fixed proportion is our simplification.
Bjork, E. L., & Bjork, R. A.

Our own construction

We made these up. They are bounded and tunable, and they are the parts most likely to be wrong.

The specific numbers: how fast estimates adapt, how much a review is worth, where the thresholds for each stage sit, how a topic's demand is weighted against another subject's.
Ours. Chosen to behave sensibly, bounded so they cannot run away, and tunable. No published result fixes these values, and we do not claim one does. They are the part of Kematra most likely to be wrong, and the part we would change first given real data.
The split between 'essential' and 'supporting' items in a topic checklist.
A product construction. It makes the pass condition explicit and adjustable, which we think is useful, but it is not a finding about learning.
The estimate of how long a task will take, and the shortfall it reports when your time does not cover your goal.
Built from a starting table we invented, then calibrated to you from the time bands you tap. The calibration is real; the table it starts from is a guess, and early estimates are the least reliable thing the product shows you.

How methods are chosen

Kematra chooses study methods by rule, not by preference — and where it adapts which method you get, it adapts by your results, not your preferences. If a method is not producing evidence that you have learned something, it moves on to another one. It will not give you more of a method because you liked it. That is a deliberate choice and it can be uncomfortable: what feels productive and what produces retention are not the same thing, and the research is fairly consistent that fluency during study is a poor guide to later recall.

What Kematra cannot do

Kematra cannot watch you study. It knows what you tell it — that a session happened, roughly how long it took, and what the result was — and everything else is inference from that. If you record a pass you did not earn, the plan believes you.
It cannot assess speaking. For a language subject it can plan vocabulary, grammar and written work, but a spoken component is something it can schedule time for and not something it can check.
It cannot tell whether you did a task well or merely did it. Execution is unobservable to us; the evidence is the outcome you record, not the process.
It does not know your course better than your teacher. Where it derives topics from an official specification it says so, and where it is guessing it says that instead.
Its estimates start out wrong. The first few weeks are a starting table plus your own answers; it gets better as you record more, and it will misjudge how long things take before then.
It has not been trialled. Kematra is built on published findings about memory and practice, but the product itself has not been tested against a control group, and we are not going to imply it has.

References

Cited for the directions above. None of them prescribe the specific values Kematra uses.

Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380.
Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008). Spacing effects in learning: A temporal ridgeline of optimal retention. Psychological Science, 19(11), 1095–1102.
Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students' learning with effective learning techniques. Psychological Science in the Public Interest, 14(1), 4–58.
Rohrer, D., & Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35(6), 481–498.
Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In Psychology and the Real World.
Ebbinghaus, H. (1885). Über das Gedächtnis. (Memory: A Contribution to Experimental Psychology, trans. 1913.)

9 decisions listed · 7 references.