Why Spaced Repetition Stops Working: The Interval Is Not the Problem

The interval in a spaced-repetition system is not a magic timer that fixes weak cards. SuperMemo says it schedules each review at the point where the probability of remembering an item drops to 90%, and a successful retrieval earns a longer delay while a failed one brings the card back within days. When a deck stops working, the culprit is rarely the algorithm. It is almost always the card.

English · 724 words

How the Interval Actually Works

SuperMemo's first optimum interval for review, calculated at 90% retrievability, is 3.96 days. SuperMemo says the first repetition is usually after two to three days, the second after about one week, and later repetitions can stretch to months or a year or longer. When an item is forgotten and rated "I don't know," the next repetition arrives after one to a few days. When rated "I know," the interval grows faster; difficult ratings pull it back sooner.

The system aims to schedule repetition at the latest possible point before forgetting risk becomes too high. These are individually calculated intervals, not fixed schedules. Same-day cramming is explicitly excluded: SuperMemo's FAQ states that the spacing effect makes immediate re-review ineffective or even counterproductive. The interval is adaptive, responsive, and entirely conditional on what happens at retrieval.

What Strengthening Memory Actually Means

A forgetting curve, as described by RemNote, plots recall likelihood against time since last recall. Memory decays exponentially after learning. Reviewing at the right moment decreases the future rate of that exponential decay. The curve becomes shallower. Information persists longer on average.

But this process strengthens what already exists. It does not create memories from nothing. The distinction is plain: retrieval practice—actively pulling information from memory—outperforms passive rereading. Yet the format of that retrieval matters enormously. Multiple-choice questions produce recognition-based retrieval. Free recall demands generative retrieval and builds stronger traces.

The schedule can extend a memory's lifespan. It cannot substitute for how that memory was formed.

The Recognition Trap

Some learners can run mature decks at high accuracy rates and still fail to speak, write, or apply. This is the recognition-production gap. Recognition cards let the answer surface from context, partial cues, or familiarity. Production cards require the learner to generate the target from minimal prompting.

A card front that lets the answer be guessed from nearby text or from visual shape teaches the shape, not the knowledge. A card that always appears with the same surrounding material, the same font, the same screen position—none of this is the knowledge itself. It is scaffolding that the real world will not provide.

SuperMemo's algorithm assumes it is strengthening the item. If the item being strengthened is "recognize the answer when prompted with these specific cues," that is exactly what improves. The learner masters the deck, not the domain.

Where Cards Actually Break

The hilog.dev guide specifies one question, one answer per card. Complex cards are harder to review consistently because they test multiple things simultaneously. A card that asks for a definition, an example, and a counterexample in one prompt is really three cards pretending to be one. When the learner fails, the system does not know which component failed.

Cards should force generation of the essential idea. A front that reads "What is the capital of France? (It starts with P)" has already given away the retrieval target. The learner practices completing "Paris" from "P...", not recalling the capital from the country alone. This distinction is invisible to the algorithm. The interval will lengthen. The memory will not transfer.

What the Schedule Really Controls

The interval's operational role is allocating limited daily study time. New material demands fresh encoding effort. Review demands retrieval effort. A fixed daily budget—twenty minutes, forty cards, whatever the learner sets—must be split between these competing demands.

SuperMemo's adaptive intervals determine how many reviews accumulate and when. They do not determine whether those reviews are worth doing. A deck full of recognition cards, complex multi-part prompts, or context-dependent cues will run smoothly through its intervals. The algorithm will faithfully schedule. Time will pass. Retention curves will flatten.

The learner will have spent minutes each day practicing the wrong thing.

The schedule buys time and distributes review load. It does not rescue card design. The real test of any spaced-repetition system is whether each card, at the moment of retrieval, demands exactly the memory the learner actually needs.