Spaced Repetition for Chinese Characters: Why Cramming Fails

Here is the uncomfortable math of character learning: you can learn 谢perfectly today — stroke order, pronunciation, three example words — and by next week it's gone. Not faded: gone. This isn't a personal failing; it's the forgetting curve, documented by Hermann Ebbinghaus in 1885 and replicated ever since. Memory for new, unconnected material decays exponentially, losing most of its strength within days. Spaced repetition is the cheapest known way to beat that curve, and it's built into this site's review queue.

The trick: review at the point of almost-forgetting

The counterintuitive core finding is that a review is most valuable when the memory is weakest— right before it vanishes. Recalling something easy does little; the effortful, uncertain recall of a nearly-lost character is exactly what resets its forgetting curve and flattens it. Each successful hard recall roughly doubles how long the memory lasts. So the optimal schedule isn't "review daily" — it's review at expanding intervals: one day, then three, then a week, then two, then a month. Ten reviews spread this way can maintain a character for years. The same ten reviews crammed into one afternoon maintain it for about a week.

That's why cramming feels great and fails reliably: massed repetition produces strong familiarity ("I know this, I just saw it") that masks weak recall. The exam-week all-nighter is real learning's most convincing impostor.

How the queue on this site works

When you mark a character as learning — from any character page — it enters your review schedule. When it comes due, the review page shows you the bare character and asks for the pinyin and meaning from memory. After revealing the answer, you grade your own recall:

• Again— blank, or wrong. The character resets to short intervals and reappears tomorrow. This is not punishment; it's the system finding your actual edge.
• Hard — got there, slowly or shakily. The interval grows, but cautiously.
• Good — clean recall with normal effort. The default; the interval roughly doubles.
• Easy — instant, automatic. The interval leaps ahead so the queue stops wasting your time on it.

Honest grading is the entire game. Grading a shaky recall as Good feels efficient and quietly schedules a future failure; grading strictly costs one extra review now and saves three later. Everything stays in your browser's local storage — no account, and the progress page can export your data anytime.

Making it sustainable

Spaced repetition has one operational rule: the queue must reach zero (almost) every day. The daily load for an HSK 1-sized set is small — five to ten minutes once past the initial wave — but it compounds viciously when ignored, and a week's absence produces a wall of reviews that has ended many a learning streak. Two defenses: first, cap new characters (six a day, as in the 30-day study plan) since every new character is a loan against future review time. Second, after any gap, clear reviews before learning anything new. An old character maintained beats a new one met.

What the queue can't do

Spaced repetition is a maintenance system, not a learning system. A character that enters the queue as an arbitrary squiggle stays hard forever — reviews preserve whatever understanding exists, including none. So do the real learning first: watch the stroke order, write it from memory, meet it inside two or three words, notice its radical. A character with that much structure behind it glides through reviews at ever-longer intervals and effectively maintains itself. The queue is the flywheel; understanding is the push that starts it spinning.