The arithmetic nobody checks
The daily review load does not come from what you added today; it comes from everything in the deck coming due. Adding at a constant rate means the load rises steadily for months before levelling off.
That is why kanji study so often collapses at around the three to six month mark. Nothing went wrong; the arithmetic simply arrived.
Choosing a rate you can sustain when the load has doubled is the decision that determines whether the system survives.
A sustainable pace
Most learners start too fast because the early sessions are short. The workable rate is the one you can maintain on a bad week, in a busy month, when nothing else is going well.
Consistency beats rate decisively here, because a smaller deck reviewed daily for two years outperforms a larger one abandoned after four months.
| New per day | Deck after a year | Sustainable? |
|---|---|---|
| 20 | Very large | Rarely |
| 10 | Large | For some |
| 5 | Moderate | Usually |
| 3 | Modest | Almost always |
| 0 for a week | Unchanged | Always fine |
The last row matters. Pausing new cards while keeping reviews going is a legitimate response to a busy period and is far better than stopping entirely.
What does not need to be in the deck
The deck should hold what you cannot yet read in context, not everything you have ever met. Characters that appear constantly in your reading are maintained by that reading, and reviewing them is duplicated effort.
This is the main lever for keeping the load down, and it is the one learners are most reluctant to use.
- Characters you read fluently in words you know.
- Characters that appear in your reading several times a week.
- Characters relevant only to material you no longer study.
- Rare characters you added for completeness.
- Anything that has failed repeatedly; relearn it differently instead.
When a card keeps failing
A card that fails repeatedly is not telling you to review it more; it is telling you it was learned badly. More repetitions of a bad encoding produce more failures.
The productive response is to rebuild it: find the components, attach it to a different word, or contrast it with the character it is being confused with.
- Identify what it is being confused with; usually something.
- Break it into components and name them.
- Attach it to a different word, preferably one you use.
- Write it by hand for a week.
- If it still fails, delete it and meet it again in reading.
Handing over to reading
There is a point where reading maintains kanji better than review does, and it arrives earlier than most learners think. Once you can read material at your level regularly, characters are met in context repeatedly and for free.
At that point the deck should shrink rather than grow: new characters come from what you meet, and old ones are retired as reading maintains them.
- Read daily at a level where you meet known characters constantly.
- Add only characters that block comprehension repeatedly.
- Retire characters you have read comfortably several times.
- Keep the deck small enough that reviews stay short.
- Judge progress by reading ease rather than by deck size.
The system that survives
The kanji system that works is the boring one: a modest daily rate, short reviews, aggressive retirement, and reading doing most of the maintenance.
The one that fails is the ambitious one, and it fails on a timetable that is entirely predictable from the arithmetic at the start.
For the recognition speed underneath all of this, the kanji games on Unihongo's kanji page are the complement to review: reviews test whether you know a character, and drilling makes recognising it fast enough for reading.
If your reviews take more than fifteen minutes, the deck is too big rather than you being behind. Retire something.