Why learning that works feels like it isn't working

You answer a question correctly, and the app asks about the same idea again. It feels like wasted time. Rereading a chapter feels smooth and productive. Watching a clear lecture feels like understanding. The research on all three feelings says the same thing: they are unreliable, and often exactly backwards. This post is the evidence.

You cannot feel yourself learning

Across randomized experiments, how much people learn and how much they feel they learn are different measurements, and they regularly move in opposite directions. Smooth methods, like rereading and watching, feel productive and fade fast. Effortful methods, like being made to answer, feel slow and hold. The feeling of learning is a poor gauge of learning.

The reason is mundane. What you can feel in the moment is fluency: how easily the material flows past you. A well-written page and a good lecturer produce a lot of fluency. But fluency measures how familiar something is while it is in front of you, and learning is what remains when it no longer is. Those are different quantities, and only one of them is available to your senses while you study.

The experiment that separated feeling from fact

In 2019, a Harvard physics group ran the clean version of the test (Deslauriers et al., PNAS). Students were randomly assigned to identical material taught two ways: a polished, passive lecture from an experienced instructor, or an active session where students had to work through the problems themselves. Then both groups took the same test and rated how much they felt they had learned.

The active group scored higher. The passive group felt they had learned more. Across students, actual learning and the feeling of learning were anticorrelated. Notably, the authors’ practical advice was not to make the effective format easier. It was to tell students, early and explicitly, why the harder format is harder, so that the effort reads as the method working rather than failing.

If you choose study methods by how they feel, this result says you will systematically select against the ones that work.

A correct answer is the halfway mark, not the end

The strongest version of the “why is it asking me again?” question has a precise answer, and it comes from a 2008 experiment in Science (Karpicke & Roediger). Students learned foreign vocabulary until they got each word right once. Then the conditions split. For some students, a word they had answered correctly kept coming back in later rounds of testing. For others, it was dropped, since they clearly knew it.

A week later, the students who kept being tested on words they already knew recalled about 80 percent of them. The students whose words were dropped after the first correct answer recalled around a third. Same material, same study time within a session, and the difference between remembering and forgetting was whether retrieval continued after success. Extra silent restudying of known words, for comparison, did nothing measurable.

The students could not feel any of this. Asked to predict their own recall, the groups gave similar estimates. Dropping an item the moment you get it right is what every instinct suggests, and it is the single most expensive mistake in the study.

Three successes, then stop

More is not better forever. A follow-up line of research (Vaughn & Rawson, 2011) had learners continue until one, three, five, seven, nine, or eleven correct recalls. Retention climbed up to about three successes and then flattened. Past three, extra repetitions in the same sitting bought almost nothing.

So the repetition that works has both a floor and a ceiling. One correct answer is not enough, and ten are a waste. A method built on this evidence re-asks after success a small number of times, makes each return harder in kind, and then leaves the idea alone until a later day.

How to grade a method, an app, or a class

Four questions do most of the work:

• What did you have to produce? Picking the right option from a list is the weakest evidence of knowledge, because the answer was on the screen. Producing the answer yourself, or applying it to a case, is stronger.
• Does it return after you were right? If being correct once retires an idea permanently, the method is optimized to feel good, not to hold.
• Do the returns go deeper? Repeating the identical question drills a reflex. A good second touch asks more than the first did.
• When do you check? Judge by what you can produce days later, not by how the session felt while it was happening.

These four apply equally to a flashcard deck, a language app, a university course, or a tutor.

Where Ulern stands

We build Ulern, a learning system, so read this section knowing that.

A session in Ulern runs the way the studies above suggest. It opens with a short calibration check: a few quick multiple-choice questions that sort what you already know from what you still need. You can skip the check; skipping just means Ulern assumes nothing and teaches everything fresh. Anything unfamiliar is taught once, properly, then confirmed with an easy first activity. The check that counts comes later in the session, on purpose. An immediate re-test would only measure short-term recall, so each idea returns after a gap, and this time you produce the answer yourself, at the level you set out to reach. When you miss, you get feedback, a fresh explanation from a different angle, and an easier retry. Across our internal test runs, this averages about three activities per idea.

So there is a moment in every session where you have just answered correctly and Ulern asks again anyway, harder. People notice it, and some tell us it feels redundant. That moment is this post. We have not yet run an external trial, so the honest claim stops here: the studies above are the reason a session looks the way it does.