How Termy Improved FSRS for Real-World Vocabulary

FSRS stands for Free Spaced Repetition Scheduler. It decides when a saved word should return for review. Easy words wait longer, while difficult words come back sooner.

Reviewing the same word twice in one afternoon is usually too easy. Waiting until it has completely disappeared turns review into learning it again. FSRS for language learning schedules the next review between those two extremes.

It does this separately for every card. A word you remember easily can wait longer, while a difficult one returns sooner. If you take a break and still remember a word when you come back, FSRS uses that success instead of punishing you for missing the planned date.

This is why Termy's Training system now runs FSRS-6. It gives each word a better calendar than a fixed ladder of one day, three days, then seven days. The calendar is only the beginning, though. FSRS still needs a trustworthy rating after each answer, and it does not decide which exercise will help you most. That is where Termy's work begins.

What makes FSRS a better scheduler

The Open Spaced Repetition project describes memory with three values. Difficulty records how hard a card tends to be for you. Stability describes how long that memory can last. Retrievability is the estimated chance that you could recall it today. Every answer updates those values.

Termy currently asks FSRS to aim for 90% recall. A miss lowers stability and brings the word back sooner. A successful answer lets the interval grow. Remembering after a longer gap can produce a larger gain because the memory survived more time than expected. The schedule adjusts without asking you to rebuild an arbitrary streak.

This matters when two words saved on the same day behave differently. One may keep appearing in the content you use and become familiar quickly. Another may look simple but disappear from memory every time. FSRS lets their calendars separate instead of forcing both through the same sequence.

Termy uses the default FSRS-6 parameters and the same 90% target for everyone. With enough review history, FSRS-6 can learn how quickly each learner forgets. Termy records that history now, but personal fitting has not shipped. Reviews use whole-day intervals rather than short steps measured in minutes. A missed card can still return at the end of the current session.

Termy gives FSRS a cleaner rating

FSRS expects a rating such as Again, Hard, Good, or Easy after every review. Many apps ask the learner to choose that rating manually. The extra choice creates uneven data because people disagree about what “Hard” means, and the same person may use the buttons differently from one day to the next.

Termy's automatic grading derives the rating from the answer instead. A wrong choice becomes Again. Correct answers are graded by response time, with fast answers treated differently from slow ones. A long sentence receives extra reading time, so the scheduler does not mistake reading for weak memory. Then Training moves to the next card.

Typed answers provide more detail. A clean answer is graded by time. A small typo or a case-only mistake receives a cautious grade, while using the Hint button caps the result at Hard. Accents still matter. Turkish text follows Turkish case rules rather than English ones. If an exercise shows the first letter from the start, that planned help carries no penalty.

Screenshot recall needs one small decision from the learner because Termy cannot see an answer held silently in someone's head. Recall time measures how easy the answer felt. After the answer is revealed, one “Missed it” action records a real lapse. The learner reports the part that software cannot observe, and FSRS handles the interval.

Termy changes the exercise as the word becomes familiar

A good review date can still produce a weak review. Choosing a definition tests recognition, while typing a missing word tests recall. Termy changes the exercise as the memory becomes stronger. New or recently missed words start with gentler tasks.

The first card may show the word in its saved sentence and ask for the meaning. Later, the sentence returns with a gap. Training can play the word or the full sentence and ask you to type what you hear. Another exercise supplies the first letter. Cold typing arrives when the card is ready for it.

Cards do not climb this progression at the same speed. Easier words can reach typing early. A recent Hard or Again answer lowers the next exercise even when the word used to be strong. Training also varies tasks within a safe range, so a session does not become twenty copies of the same question. Premium training can bring back a saved screenshot for visual recall as well.

Different forms of a word can share one card. Forms such as goes, went, and going belong to the same base word when the language data identifies them correctly. Their saved sentences remain available as extra contexts rather than becoming nearly identical schedules. This keeps the unit of learning close to the word itself.

Termy keeps FSRS connected to real encounters

A scheduler normally sees only reviews. Termy also knows when the same word was checked again in a different sentence. That new encounter is useful evidence: the word appeared in content that mattered to the learner, and it still needed attention. Words with more distinct encounters move up the training queue.

New saves wait in a holding pool before they enter active training. For each selected training language, Termy normally activates up to 15 words a day, with repeated words going first. Intake pauses when the due queue grows too large. Saving many words during one long reading or gaming session should not bury the learner later.

Training also works when someone starts a session before any card is due. Termy fills the remaining space with active words that are closest to review. An early success updates the memory state, but it cannot push the word beyond its old due date. Recalling a word six hours after the previous review does not prove that it will survive an extra week.

Late reviews keep their real timing. Remembering a word after a long break can earn more stability. Missing it lowers stability, sends the card to the end of the current session, and makes the next exercise easier. The saved context gets another chance to help before cold recall returns.

Some cards need more than another short interval. After four misses, Termy opens a relearn panel with the definition, translation, and two or three saved encounters. The learner can continue, snooze the card for 30 days, remove it from training, or delete it. Repeating the same failed prompt forever would add work without explaining the word any better.

Strong cards eventually leave the active queue. Retirement requires two recent successful answers, no more than one lapse, and a long FSRS interval. A word below the learner's current level can retire at a 21-day interval, while words at or above that level wait for 45 days. If a retired word is checked again in Termy, it becomes active and due immediately. A fresh encounter carries more weight than its old status.

For a wider look at how selective review supports real content, read what the Antimoon Method gets right about input and spaced repetition. Termy will use the review history it records now to evaluate personal FSRS parameters later; that work has not shipped yet.

Termy builds this training from words checked in games, videos, ebooks, and websites, with their meaning and original context still attached.