You open Anki before work and it says 312 due. You do them on the train, at lunch and after dinner. You take a weekend off, and on Monday it says 640. Japanese has a word for growth like that, 雪だるま式, the way a snowball gets bigger the further it rolls, and it describes a lot of Anki decks well.
The usual advice is to be more consistent: never miss a day, do your reviews first, add fewer cards. That helps, but it treats the pile as a willpower problem. Most of it is a design problem. Once you see where the reviews come from, you can cut a surprising share of them without learning any less.
Every card is a subscription
When you add a card to Anki, you aren't scheduling one review. You're signing up for a review now, another in a few days, another in a few weeks, and on and on for as long as the card stays in the deck. The gaps get longer, but they never end.
That's why reviews grow even when your habits don't change. Twenty new cards a day, Anki's default, doesn't mean twenty cards' worth of work a day. It means twenty more subscriptions every day, stacked on all the earlier ones. A commonly quoted rule of thumb is that daily reviews settle at around ten times your daily new cards, so twenty new a day ends up as roughly two hundred reviews a day once the deck matures, which happens to be Anki's default review limit.
Skipping a day doesn't pause the subscriptions either. The cards keep coming due, so a missed day lands on top of the next one.
None of this is a flaw in spaced repetition. It's the price of remembering things for years. The question is whether every card in your deck is worth that price.
Many of your cards are words you already knew
Premade vocabulary decks start at word one. The core 2k or 6k deck someone recommended doesn't know that you finished a textbook, played through two games in Japanese or spent a year in Osaka. It hands you 食べる (to eat) and 今日 (today) in your first week, right alongside words you've never seen.
A known word costs almost nothing to answer, but it costs something every time, for years. If 800 cards in a 2,000-card deck are words you already knew, those 800 still get the full schedule, and they still make the due count look heavier on the mornings you are deciding whether to open Anki at all.
What you can do in Anki:
- Go through the next few hundred new cards in the browser before they reach you. Suspend the ones you know cold, or tag them and suspend them all at once.
- For words you know but want to check eventually, use Set Due Date in the browser to push them a year out, instead of letting them start from day one.
- Prefer decks sorted by frequency, so you can skip a block at the start and know roughly what you skipped.
This works, but it's slow, and you are the one doing the sorting: deciding, card by card, what you know.
"New cards per day" counts cards, not words you're learning
Anki's new-card limit is a fixed number. It doesn't know whether today's twenty new cards are twenty words you've never seen or twelve you knew and eight you didn't. Those two days feel the same while you're doing them and make very different futures: one adds eight words that will come back again and again because you keep missing them, the other adds twenty.
The cards that drive your workload are the ones you miss. A missed card restarts at short intervals and comes back many times before it settles. Words you get wrong more than a handful of times turn into leeches: by default Anki tags a card as a leech after eight lapses and leaves it in the deck.
What you can do in Anki:
- Make new cards the thing that gives, not reviews. On a heavy day, add none.
- Watch how often you press Again on new cards. If you're missing most of them, the deck is above you; drop the new-card limit until that rate comes down.
- Deal with leeches instead of grinding them: rewrite the card, add an example sentence, or suspend it for now.
Your retention target is higher than you need
Anki now offers FSRS, a scheduler that models how likely you are to remember each card and times reviews to a desired retention you choose. The default is 90%: each card comes back just as your chance of remembering it falls to nine in ten.
That number matters more than it looks. The closer it gets to 100%, the more often each card has to come back, and the curve gets steep near the top: each extra point of retention costs more reviews than the last. For vocabulary you will meet again in reading and conversation anyway, 90% is often more safety than you need.
What you can do in Anki:
- Turn on FSRS if you haven't, and let it fit its parameters to your review history.
- Try a lower desired retention, such as 85%, and watch your daily load for a few weeks.
- Recent versions also spread reviews across nearby days and let you mark lighter days of the week, which smooths the peaks without changing the total.
Grading yourself is harder than it sounds
Anki shows you the answer and asks how it went. That's flexible, but it means you're the judge, often at speed. Recognizing 懸念 (concern) once you flip the card feels like knowing it. Pressing Hard when you really mean Again feels kinder. Both make the schedule less accurate, and both tend to push the work later rather than remove it.
One-way cards add a second gap. Japanese on the front trains recognition only. Many learners add the reverse card to practice producing words, which roughly doubles the reviews for every word.
What you can do in Anki:
- Use a type-in field (
{{type:Back}}in the card template) for production cards, so the check is the word you typed, not your impression of it. - Press Again whenever you didn't have the answer before you flipped. Keep Hard for answers you got, slowly.
- Add reverse cards only for words you actually want to say. Recognition is enough for many words.
You can't see what the work adds up to
Anki's statistics tell you about cards: how many are mature, how many you retained last month, how long reviews took. They don't tell you how much Japanese you now know. When the only number in front of you every morning is the due count, the habit slowly turns from learning into paying off a debt, and a debt is easy to start avoiding.
What you can do in Anki:
- Use a frequency-sorted deck, so "cards learned" maps roughly to how much of everyday Japanese you cover.
- Keep the deck to what you care about. A word you'll never use is a subscription you'll pay for anyway.
What all these fixes have in common
Look back over the list and a pattern shows up. Anki schedules whatever you give it, very well. It doesn't know which words you already know, how hard a word is for you, or which words are worth learning. So every fix above is you doing that work by hand: sorting known cards out, sizing new cards to your level, lowering the retention target, policing your own grading, picking the right deck.
That's a reasonable trade for a general tool. You can put anything into Anki, from kanji to anatomy, and it will hold it. But for vocabulary, where the words have a known order of usefulness and your level can be measured, an app that starts from those facts can skip most of the pile before it forms.
How Tangoya is built around the pile
Tangoya is a Japanese vocabulary app that starts from the opposite end. Instead of scheduling every card you add, it schedules only the words you show it you don't know.
It learns your level, then skips what you already know. Tangoya sorts its words into six kinds (single kanji, kanji compounds, compound verbs, loan words, onomatopoeia and kana words), and the first 30 answers of each kind measure where you are. From then on, every word you haven't been asked has a chance of being known, worked out from your level and how common the word is. The words you almost certainly know are counted without asking them, so nobody drills 食べる for a year to prove a point. New words come from the edge of your level, a little below it, so you get about three in four right, and only about three new words in a set are ones you'll miss.
Only misses come back. Get a word right the first time and it doesn't come back: Tangoya takes your word for it. The trade-off is real. A word you knew today and forget next year isn't re-asked. In exchange, the hundreds of reviews that would have gone to words you already had never happen.
Misses come back on FSRS, at about three in four. The words you miss return in a minute, then ten minutes, then days, weeks and months. Each comes back when you'd still remember it about three times in four, a lower target than Anki's default, because these are the words you're working on, not words you need to hold at nine in ten forever.
Breaks don't snowball. After a break, the next set is a catch-up day: up to 40 words coming back and 4 new ones, the most recent misses first. And once a kind of word finds your level, words you missed far above it leave practice, since those were guesses, not gaps.
Nothing to grade. Each word is asked one way at a time: Read (see it, pick its meaning from four), Listen (hear it, pick its meaning) or Recall (type the word that fills the gap in a sentence; any form of the word counts). The app checks the answer. If you're not sure, I don't know is right there, and it counts as a miss, so the word comes back for you to learn.
The number that moves is the language. Instead of a due count, Home shows two shares of the 10,000 words most used in everyday conversation: how many you understand, and how many you can produce. Finish Quick, the first ten questions, and the day counts.
If you like Anki for what it does well, keep it. It's still a good home for sentences you mine from shows and books, or anything outside the most common words. But if the everyday vocabulary is where the pile comes from, you don't have to sort it by hand.