Why one review calendar doesn't fit everyone

Suppose you give every new word the same plan: leave gaps of one day, three days, a week, and a month between successive reviews. These are example intervals, not a recommended formula.

The plan is easy to follow and spreads practice over time. You don't need a complicated app to use it. When you're starting with a few words, that simplicity can be helpful.

Then two words begin to behave differently. One comes to mind easily. The other keeps getting confused with a similar word. If their dates must stay identical, you may spend time checking one more often than needed while still struggling to recall the other.

Words arrive with different histories

Imagine an English speaker learning Spanish. The written form of hotel is familiar; desarrollar, meaning "to develop," may need more work. This is an example, not a ranking of difficulty for every learner. Familiar spelling doesn't mean that pronunciation and use are already learned.

Word length alone won't explain the difference. What the learner already knows, what they confuse, and the context in which they meet a word can all matter. Another learner may have a different starting point.

A calendar that considers only the review number can't respond to those histories. The third review is the third review, whether the first two answers were easy or needed help.

A fixed schedule is a starting assumption

There is nothing inherently wrong with a simple schedule. When you have little information, a sensible default helps you begin.

The trouble comes when the schedule cannot respond. If you miss an answer, should the next gap shrink? If you recall it easily after an unexpectedly long delay, should the gap grow? How much should either result matter?

As the collection grows, those choices affect how much time you spend checking familiar words and how often difficult ones become inaccessible. A useful schedule needs a way to change its mind.

An adaptive system treats its initial schedule as revisable. It can use the answer and the time since the previous encounter to reconsider the next date. That is more specific than moving every word to the same next box on a calendar.

The same learner can change

Even a schedule built around one person needs updating. A word that was rare last month might now appear in every chapter of a book they're reading. A distinction that once caused confusion might become clear after a useful conversation.

The app won't observe all of that. It has a smaller record: the answers its review system saves and when they happened. This is evidence about performance, not direct access to the learner's brain or every conversation outside the app.

An answer after one day and an answer after a month can have the same label while giving different evidence about retention. The elapsed time helps explain what that answer tested.

An adaptive estimate can still be wrong. More information gives a system something to work with; it doesn't make the next answer certain.

Turning answers into a new interval

One way to make this practical is to keep information for each card. The SM-2 algorithm, for example, uses response grades and an easiness factor, a value associated with the item, to adjust intervals.

That is a move toward a more useful question: given what happened with this item, how should its schedule change?

FSRS explicitly models memory state and calculates recall probability. Updating an estimate for one word is not the same thing as fitting all the model's parameters to one person. We'll return to that distinction later in the series.

For your own practice, notice what keeps going wrong with a word. Do you confuse its meaning, spelling, or the question you're being asked? Changing the date alone can't explain the difference between two meanings or clarify an unclear prompt. Sometimes the learning material needs attention too.

The calendar remains the place where the date appears. The decision behind that date becomes more specific to the item and its learner.

Next, we'll look at how review systems moved from boxes of cards to models like FSRS.