Docs / Why it works this way

Why it is harder than it needs to be.

Several things Ignia does look like friction added on purpose. It gives you less credit for a right answer in an easy format. It limits the format you are quickest at. It hands you a partly worked exercise and then takes the support away one question later. This page is about which of those are deliberate, why, and where the deliberate part stops.

What you actually want.

A session that goes well. Answers that come out right, a sense that it is going in, and some evidence at the end that the half hour was worth it. That is a reasonable thing to want, and nothing on this page is going to tell you it is a character flaw.

What you run into.

The obvious way to give you that is to smooth practice out: keep the success rate high, favour the formats you do well in, leave the support in place until you feel ready to drop it. Every one of those is available to us, and each one would make the app more pleasant tomorrow.

It breaks twice, and the second break is the one specific to an app that grades you.

How well a session goes is a poor guide to what you keep.

Performance during practice and retention weeks later routinely move in opposite directions. Conditions that make practice go worse — spacing it out, mixing types together, producing an answer instead of picking one — tend to produce more durable knowledge, and conditions that make practice go smoothly tend to produce less. This is a well-replicated pattern, not a single study, and it has a name in the literature: desirable difficulties. Its practical consequence is blunt: the one signal available to both of us while you work is the wrong one to steer by.

A smoothed-out session does not only teach less — it also misinforms the schedule.

Every answer here becomes a number, and that number is what decides when the point comes back. So removing friction has a second effect that has nothing to do with how the session feels: the score goes up, the estimated durability goes up with it, and the next review gets pushed out past the point where you would still have had it. You do not find out from a bad session. You find out weeks later, from material you were told you had.

That is why difficulty here is not a matter of taste. An easier session buys a worse schedule, and the schedule is the part of the app you cannot check by feel.

What we do instead.

Not « make it hard ». Difficulty in general is not a design principle, it is a mood — and one that is very easy to mistake for rigour. What the app does is refuse the smooth option in four specific places, each of which is written down in the code and each of which you can see happening.

A right answer you could have guessed counts for less towards your schedule. Before a score reaches the scheduler it is capped by the guess rate of its format: at most 0.75 for a true-or-false, 0.875 for a four-option multiple choice, uncapped for an answer you typed out (chance_rate and clamp_fsrs_for_kind, mastery/mod.rs). The stored attempt is untouched — you are told you got it right, because you did. What is discounted is the confidence the schedule takes from it.
Recognition is capped at 30% of a session. Picking the right option out of a list can succeed on a memory far too weak to have produced the answer. It is not worthless, so it is not banned — it is bounded, in code, with a test that fails the build when a generated session exceeds it (MCQ_CAP, sessions/grid.rs). The reasoning is on why you are asked, not told.
Support is removed one exercise later, not when you feel ready. On a point you are shaky on, the first exercise arrives with part of the work already done — the card is labelled Faded — and the next exercise on that same point arrives without it. The ladder runs inside a single session and it is driven by your recorded state on that point, not by a sense of readiness that would arrive too late (scaffold, sessions/grid.rs).
A point you have met before is spread through the session rather than finished off in one block. Answering the same notion twice in a row means the second answer is given with the first still in mind, which is easier and which reports a spaced retrieval that never happened. Why that is gated on what you already know is its own page.

What this does not buy.

Five things, and the second one is a real gap rather than a modest disclaimer.

Difficulty is not the active ingredient. The benefit comes from effortful retrieval that succeeds, not from struggle as such. An exercise beyond your reach, or ambiguously worded, or asking for something the lesson never taught, is not a desirable difficulty. It is a defect, and there is nothing to be gained by pushing through it.
And we cannot tell those two apart from your score. This page opened by saying that in-session performance is a bad index of learning. That cuts against us too: when you do badly on an item, the app has no way to know whether that was a productive difficulty or a broken question. It has your answer and nothing else. This is a genuine limitation of the design, and it is why the Discuss action on an exercise is worth using.
The numbers are calibration, not findings. That recognition is capped, and that a two-way guess is discounted harder than a four-way one, are defensible directions. The specific 30%, and the specific « half the guess rate » discount, are notches we chose. The code says so where they are defined, and we would rather tell you than imply a precision we do not have.
There is no target success rate, and the famous one does not apply. You may have met the rule that training should be tuned to around 85% success. It is a real result, derived for two-choice tasks with continuously adjustable difficulty — it does not transport to a mix of formats whose guess rates differ by a factor of two or more, where the same 85% means three different things. Ignia does not aim at a percentage. It shifts which formats are offered depending on how a point is going for you, which is a coarser instrument and an honest one.
Not every friction in the app is one of these. The placement test withholds feedback on each question while you take it — but not to make it harder for your benefit. It is so the answers reach the planner uncontaminated by your having just been shown the answer to a related question. That is a measurement reason, not a learning one, and calling it a desirable difficulty would be exactly the sort of retrofit this section exists to avoid.

Questions.

I answered correctly. Why did the app act like I only half did?

It did not. You are told you got it right, and the attempt is stored as a success. What is discounted is only how much confidence the scheduler is allowed to draw from it. A correct true-or-false is a correct answer that a coin flip also produces half the time, so it moves your review date less than a typed answer would. If the discount did not exist, the schedule would push that point further out than your actual memory of it justifies, and you would meet it again after you had lost it.

Can I make sessions easier?

You choose how deep the course goes when you create it, and how long a session lasts. Both change what you are asked. Neither changes how forgivingly an answer is scored, and that is deliberate: a leniency dial would inflate exactly the numbers the review schedule is computed from, so it would buy a pleasant session by damaging the thing you are relying on the app for.

That exercise was not hard, it was badly written. What do I do?

Say so. Every exercise carries a Discuss action, and an item that is ambiguous, or that asks for something the lesson never covered, is a defect on our side — not a difficulty you are supposed to push through. We cannot tell the two apart from your score, which is why we would rather hear it from you.

Is this going to make me feel like I am doing badly?

Possibly, and that is the honest answer. Practice that is spread out, unaided and mixed up feels worse than practice that is massed, supported and blocked — that is the whole mechanism, not a side effect. What we try to avoid is compounding it: there is no streak, no leaderboard and no running tally of how you are doing, because those add a second thing to feel bad about without adding anything you learn from.

Sources.

What each one is doing here, so you can tell which claim rests on which work. Every one of these is a published paper you can look up.

Bjork, R. A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe & A. Shimamura (Eds.), Metacognition: Knowing about Knowing. MIT Press. · Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way. In Psychology and the Real World.

Where "desirable difficulties" comes from. Note that it is an umbrella over several effects rather than a single one — which is why this page names specific mechanisms instead of claiming difficulty is good in general.
Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176–199.

The load-bearing one: performance during practice is a poor and often inverted index of what is learned. This is the paper behind the claim that the signal available to you during a session is the wrong one to steer by.
Wilson, R. C., Shenhav, A., Straccia, M., & Cohen, J. D. (2019). The eighty five percent rule for optimal learning. Nature Communications, 10, 4646.

Cited here as the rule the app does NOT apply. Read the derivation: it is for two-choice tasks with continuously adjustable difficulty, which is why the number does not transport to a mix of formats with different guess rates.
Sweller, J., van Merriënboer, J. J. G., & Paas, F. G. W. C. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10(3), 251–296. · Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31.

Why support is faded rather than left in place: the scaffolding that helps someone new to a point becomes redundant, and then harmful, for someone who has it.

You will find effect sizes in these papers. We deliberately do not print them on this page: an average across studies does not predict what any one person will get, and a number quoted without its conditions reads as more precise than it is. In the papers they come with those conditions.

Related.

Why you are asked, not told — where the recognition cap comes from. · Why it mixes topics up — the difficulty this page mentions last, at full length. · Session engine — the caps, the scaffold ladder and the constants at source level, with the known gaps.