Why You Should Learn Before You Review: Imprimo's 4-Phase Learn Mode
The moment I released the first version of Imprimo, people started complaining about the same thing: they had a queue full of cards they didn't actually know. They'd added them from a shared deck, or from a PDF import, and now spaced repetition was dutifully scheduling reviews for material they'd never properly learned in the first place.
That's not a scheduling problem. That's a learning problem. And no algorithm, however sophisticated, can fix it.
Learn Mode is my answer. Before any card enters your FSRS review queue, it goes through four distinct phases designed to take you from zero to genuine understanding. Only after passing all four does the card get scheduled for long-term retention.
The core problem with standard spaced repetition
The standard flashcard workflow looks like this: create a card (or import one), and it immediately enters your review queue as a "new" card. The algorithm schedules it for review, you see the front, you try to recall the back. If you can't, it comes back soon. If you can, it comes back later.
The assumption buried in this model is that seeing the card and attempting recall is itself enough to build initial memory. Sometimes it is, especially for simple vocabulary or facts you've encountered before. But for new and complex concepts, cold recall is premature. You haven't formed a mental representation of the concept yet. You're just pattern-matching against text you vaguely recognize.
The result is that a lot of people's review queues are full of cards they consistently fail. They see the front, draw a blank, flip to the back, think "oh right," and click Again. This repeats. The card keeps coming back. The backlog grows. Eventually the person concludes that spaced repetition doesn't work for them and stops.
Spaced repetition does work for them. The encoding step was missing.
What Learn Mode actually does
Learn Mode organizes the initial learning of each concept into four sequential phases. Each phase has a specific cognitive purpose, a specific activity, and a specific threshold you cross before moving to the next.
The total time for a typical concept is around eleven minutes: two minutes for Explore, three for Connect, four for Practice, and two for Confirm. These aren't arbitrary. They reflect the cognitive load each phase places on working memory and the time needed to do the activity meaningfully rather than rushing through it.
Sessions are resumable. If you start a concept and stop partway through, you can pick up where you left off within seven days without losing progress. The app stores your current phase, completed phases, and any explanations you wrote, so returning to a half-finished session picks up cleanly.
Phase 1: Explore
The first phase introduces the concept fully before asking you to do anything with it. The Explore screen shows a plain-language definition pitched to your subject level, an analogy connecting the concept to something you already know, two to four key points, and a concrete example. That's it. Nothing to fill in, no quiz.
This structure mirrors elements of SQ3R (Survey, Question, Read, Recite, Review), a reading comprehension method with decades of research behind it. The idea is to give your brain a full structural scaffold for the concept before asking it to retrieve anything. You're building the storage location in memory before trying to put something in it.
The AI generates all of this content from the topic and subject profile you specify. If you're studying Python fundamentals, the definition uses programming language. If you're studying pharmacology, the example uses clinical context. The lesson adapts to your domain, not the other way around.
When you feel you have the concept, you tap "Got it, Continue." There's no quiz here. The goal is orientation, not testing. Testing comes in Phase 3.
Phase 2: Connect
Phase 2 is where the real encoding happens. It's built around the Feynman Technique, named for physicist Richard Feynman's observation that explaining something in simple language is the most reliable way to find out whether you actually understand it or just recognize it.
The Connect phase has one task: write your own explanation of the concept.
There's a text field, a prompt, and no minimum or maximum length. You can write three sentences or three paragraphs. What matters is that you're generating language from your own mental model rather than copying the definition back. The process of finding your own words forces you to engage with the concept structurally, not just aesthetically.
The app saves your explanation. When the concept shows up later in Confirm, and eventually in review, your own words are part of the record. Your explanation from Phase 2 becomes a reference point for measuring whether your understanding has held up over time.
There's a "Skip this step" button for people who want to move faster. Don't use it. The Connect phase is the highest-leverage activity in the entire workflow. Skipping it is like skipping the weight-bearing part of a workout and wondering why you're not getting stronger.
Phase 3: Practice
Phase 3 is standard active recall with the AI-generated flashcards for the concept. You see the front of each card, attempt to recall the answer without looking, then flip to reveal the back and grade yourself.
The grading scale is the same one used in the main review queue:
| Grade | Meaning |
|---|---|
| Again | Didn't recall it, needs to come back soon |
| Hard | Got it but it took real effort |
| Good | Recalled it correctly with normal effort |
| Easy | Instantly and confidently correct |
Every card gets its own grade. This matters more than it might seem. After you complete Phase 4, each card enters the FSRS review queue with a stability value derived from how you graded it here. A card you marked Easy gets a longer initial interval than one you marked Hard. The algorithm has calibrated data from your very first review, rather than defaulting to a fixed one-day starting interval regardless of how well you know the material.
This per-card grading during Learn Mode is one of the mechanics I'm happiest with. It means your first real review of each card is scheduled based on evidence, not a guess.
Phase 4: Confirm
The final phase has two parts: a celebration moment and a self-assessment.
The celebration is brief: a checkmark, a "Great job," a prompt to rate your confidence. Completing the practice cards is real progress, and the emotional tone of learning matters for whether you open the app again tomorrow.
The confidence rating is a five-star scale with labels from "Still learning" to "Mastered." This is metacognitive self-assessment: evaluating your own level of understanding, not just whether you could recall specific cards. Research on metacognition consistently shows that learners who accurately assess their own knowledge gaps retain material better and study more efficiently over time.
Your confidence rating is stored alongside the concept. It feeds into the initial FSRS calibration alongside your per-card practice grades, giving the algorithm a fuller picture of where you actually stand. A concept where you rated yourself 4/5 enters the review queue differently than one where you rated yourself 2/5, even if your Practice phase grades were similar.
Once you tap "Complete Learning," all cards from the concept are written to your review queue simultaneously with their individually calculated initial stability values. The concept is marked as learned, the deck progress updates, and the FSRS scheduler takes over.
How it connects to spaced repetition
A common question is whether Learn Mode adds too much friction before you can start reviewing. The answer depends on what you're trying to achieve.
If you're maintaining knowledge you already have, Learn Mode isn't the right tool. Import a deck, send it to review, and the standard FSRS queue will do its job efficiently.
If you're learning something new, and most of Imprimo's users are, the upfront investment of eleven minutes per concept pays for itself quickly. Cards you learn properly through all four phases fail at a much lower rate in the review queue. A lower failure rate means shorter daily review sessions, not longer ones. The algorithm schedules fewer revisits for cards with healthy stability, so you're spending review time on what actually needs reinforcement rather than re-encountering cards you never learned.
The relationship between Learn Mode and the FSRS review queue is sequential, not competing. Learn Mode handles initial encoding. FSRS handles long-term maintenance.
Multi-concept sessions
Most real topics contain more than one concept. A deck on "Python Decorators" might have concepts covering closures, function wrapping, the @ syntax, use cases, and common patterns. Each concept goes through all four phases independently.
After completing a concept, Imprimo shows how many concepts remain in the deck and offers two options: continue to the next concept immediately, or stop and come back later. The break suggestion actually comes from the app itself. After completing two or more phases within an hour, the session notes that a rest might help consolidation before continuing.
This is grounded in something real. Memory consolidation is aided by rest and sleep. Pushing through ten concepts in a single two-hour session is less effective than spreading them across multiple shorter sessions, even when the total time is the same. The app doesn't block marathon sessions, but it does say something when you've been going for a while.
What the AI generates, and how it adapts
Every lesson in Learn Mode is generated by AI, tailored to the subject profile you set when creating the deck. The profiles include general, programming, medical, law, language learning, and engineering, among others. Each profile calibrates the vocabulary, examples, and analogies to fit the domain.
For a concept in a software engineering deck, the analogy might compare a design pattern to a real-world manufacturing process. For a medical deck, the same concept gets explained in clinical terms with patient-facing examples. The AI isn't filling a fixed template. It's generating content that fits the mental models a learner in that domain already has.
The AI also generates the practice flashcards for each concept, targeting a distinct testable idea per card. Cards that test recognition ("Which of these is...") are avoided. The practice cards are retrieval-based: front asks a specific question, back provides a specific answer, and the card tests one thing.
If a concept comes back with zero valid cards after generation, it's discarded automatically. The same applies to concepts that fail to load during a session: the system skips them and moves to the next valid concept rather than showing an error.
Who Learn Mode is built for
Learn Mode is most useful for people studying subjects where most of the material is new. Medical students building a knowledge base from scratch. Engineers learning an unfamiliar codebase or API. Language learners expanding vocabulary in a language they're not fluent in yet.
It's less useful for people who already have solid conceptual foundations and are just trying to maintain memory of specific facts. If you know what a Python decorator does and you just want to remember the syntax, you don't need to go through an eleven-minute lesson. You need a review card.
The typical flow for a new user is to use Learn Mode for the initial acquisition phase of a subject, then shift to pure FSRS review once the conceptual foundations are in place. The two modes complement each other: Learn Mode builds the structure, spaced repetition maintains it.
The honest trade-off
Learn Mode adds time upfront. Eleven minutes per concept is a real commitment, and a deck with ten concepts is roughly two hours of initial work spread across sessions. For people who are used to importing a 500-card shared Anki deck in thirty seconds, this can feel slow.
The trade-off is that those 500 cards go into your review queue with no initial learning, and you'll spend the next several months struggling through cards that never quite stick because you never properly learned them. The same deck run through Learn Mode takes longer to start, but the review sessions that follow are shorter and more successful.
Whether that trade-off works for you depends on your timeline and your goals. If you have an exam in three days, Learn Mode is probably not the right tool. If you're building knowledge you want to keep for years, the upfront cost is worth it.
Download Imprimo and try a concept. The onboarding runs you through a complete Learn Mode session on a topic of your choice so you can experience the workflow before committing to a full deck. If it fits how you learn, you'll know within the first concept.
If you're weighing whether to learn a topic through Learn Mode or jump straight to review, the comparison of active recall and spaced repetition explains how the two methods fit together and when each is the right choice.