What opens a loop
One wrong answer does not open a loop. The app looks at how many sessions the wrong answers on a topic are spread across, because three wrong answers on one tired evening and three wrong answers on three separate days are not the same evidence. The screen states it plainly: under each topic it shows how many incorrect answers fell across how many sessions. Spread is a sign of a gap. Clustered in one night is usually a sign of the night. The session is the unit for a reason: practice distributed across sessions leaves a different trace than one long sitting.
Four states, all of them named on screen
A topic sits in new evidence while the record is still thin. Once enough has accumulated it becomes recall ready, which opens your earlier questions for a closed book attempt. Passing that moves it to verification pending. It becomes verified only when the same knowledge is used correctly on a different question. All four states are written out, so you are never guessing which stage a topic is in.
Why the same question does not count
Getting a question wrong and then getting the identical question right a minute later is short term memory, not learning. Transfer of learning is the ability to use knowledge in a new context, which is what an exam actually asks for. That is why the closing condition is the same knowledge on a different question. The app will not re-ask the identical item immediately and count that as success.
Why the system does not guess
No extra AI call is made for any of this. The screen prints the cost returned by the server, and that cost is zero provider calls and zero model tokens on every response. The tracking reads three observed things: which question you answered, which session it happened in, and what you verified afterwards. So the reason a topic is open is a record of your own answers rather than an estimate.
What a hand kept mistake list still does
A hand kept list has two real costs: writing it takes time, and the decision to close an entry never gets made. Writing one by hand still works, and the reason it works is the same one measured in the testing effect literature. The difference here is not that the writing is automatic. It is that the closing rule is written down. An entry is not removed because you feel you know it now. It is removed because it was verified.
What a session costs
A recall session generates nothing new. It uses questions you have already seen, so it does not come out of your daily quiz allowance and it spends no AI credit. When you miss one you see the worked path, then move to the next item. You can stop wherever you like, because what is being measured is the evidence that accumulates rather than the length of the sitting.
Where it lives
Loops sit inside the study section, with the count of open topics shown there. When you finish a quiz, the result screen tells you how many wrong answers were added to the model. For the same data gathered by topic rather than by loop, weak topic analysis collects it, and spaced repetition carries the schedule for topics that have already closed.
| State | Meaning | What it waits for |
|---|---|---|
| New evidence | Wrong answers accumulating | Spread across sessions |
| Recall ready | Earlier questions opened | A closed book attempt |
| Verification pending | Recall passed | Same knowledge, new question |
| Verified | Loop closed | Hands over to the schedule |
Sources used on this page
- WeSolve+ App Store
- Transfer of learning
- Testing effect
- Active recall
- Desirable difficulty
- Distributed practice
- Spacing effect
- Forgetting curve
- Recall (memory)
- Free recall
- Recognition memory
- Judgment of learning
- Metacognition
- Metamemory
- Self-regulated learning
- Formative assessment
- Educational assessment
- Mastery learning
- Overlearning
- Varied practice
- Practice (learning method)
- Cognitive load
- Working memory
- Long-term memory
- Memory consolidation
- Encoding (memory)
- Retrieval-induced forgetting
- Feedback
- Learning curve
- Karpicke and Roediger 2008, Science
- Roediger and Karpicke 2006
- Dunlosky and colleagues 2013
- The Learning Scientists, retrieval practice
What is a learning loop?
A topic that stays open after a wrong answer and closes only when the same knowledge is used correctly on a different question. Answering the identical question again does not close it.
Does one wrong answer open a loop?
No. The app looks at how many sessions the wrong answers are spread across. Answers clustered in one night usually say something about the night rather than about a gap.
Does a recall session use my quiz allowance?
No. The session generates no new questions, it reuses ones you have already seen, so it does not come out of the daily quiz allowance.
Do loops spend AI credit?
No. No extra model call is made for the tracking. The screen prints the cost the server returns, and it is zero provider calls and zero model tokens.
Why is a topic still open?
You may have passed the recall stage while the same knowledge has not yet appeared on a different question. The topic stays open until that verification arrives.
Does this replace a mistake notebook?
It replaces the writing, not the thinking. Putting a reason in your own words still helps. What changes is that closing an entry now follows a stated rule.
Last updated: 2026-09-23 · Written with AI assistance and reviewed before publishing.