For learners
温故而知新
An instrument, not a scoreboard. It measures what you have actually acquired — how much you retain, what you can produce rather than merely recognise, and where your practice is thinner than it feels.
The Lab reads your own practice, so opening it asks you to sign in.
Why it exists
Mandarin Temple already knew how much you had done. It did not know what you had learned — and those come apart much further than anyone expects.
A progress bar measures how far you scrolled. It moves identically for the lesson you understood and the one you skimmed at midnight, which makes it useless for deciding what to do next.
The Lab counts retrieval separately from presence, and only retrieval feeds a mastery figure.
“HSK 2: 74%” averages together recognising a character, hearing it, saying it and writing it — four abilities that come apart badly in Chinese and need different practice.
Everything is measured per facet, so a strong reader with weak production can see exactly that.
Review schedules are usually built on a textbook curve. Yours is not the textbook’s: your own intervals, your own accuracy, your own decay.
Your forgetting curve is drawn from your real review history, not from an average of other people.
What actually happens
Four steps, in order. Nothing is inferred, nothing is modelled from people who are not you, and the last step is allowed to refuse.
A flashcard, a practice item, a spoken phrase, a written character. Nothing here is a separate test — the Lab reads the practice you were doing anyway.
What was asked, what you said, whether it was right, how long it took, how long since you last saw it, and which of the seven facets it exercised. One row, appended, never edited.
Each word you have met carries its own running state: how often it has been tested, how often correct, how stable it has become, and which mistakes recur.
Every figure arrives with the evidence count behind it and a confidence band. Below the threshold it does not render a smaller number; it says how much more is needed.
What gets measured
A Chinese word is not one thing you either know or do not. It is seven abilities that grow at different rates — which is why a single score for “知道” tells you nothing you can act on.
Seven ways of knowing a word — meaning, character, pinyin, tone, listening, speaking and writing — counted apart, because in practice they come apart.
A free account opens the first three — Meaning, Character and Pinyin — with real evidence behind them. The other four are measured on Pro.
Create a free account to see what it measuresWhat you do not get here
The Lab never grades you, never changes your schedule, and never touches XP or streaks. Your points and your streak carry on exactly as they did.
They are counting different things on purpose. XP and streaks measure activity, which is what keeps a habit going. The Lab measures evidence of learning, which is a harder and less flattering question. A long streak with a thin Lab is real information, not a bug.
How to trust it
Every figure carries the number of observations behind it, and a metric that has not earned the right to be shown says so instead of printing a number.
Reading a page proves you were there. Answering proves you retrieved it. The Lab counts those apart, everywhere, because collapsing them is how “74%” comes to mean nothing.
No webcam, no eye-tracking, no emotion or attention inference — not now and not later. What is recorded is what was answered and how long it took.
Questions
No. The Lab only reads the record your ordinary practice already produces — it never grades you, never changes your schedule, and never touches XP or streaks. The five cognitive tasks are entirely optional and change nothing about your learning either.
Because a percentage from three attempts is noise wearing a number’s clothes. Each metric has an evidence floor — eight retrievals for a facet, twelve reviews for a retention figure — and below it the Lab tells you what is missing instead of printing something you might act on.
Because each facet fills from a specific kind of practice, and no amount of a different kind will fill it. Meaning moves from reveal-mode flashcards and reading; Character from choosing the character in the Read step of a lesson practice; Pinyin from Write mode on a flashcard or the write step in a lesson. Those three are measured on a free account, and each names what would count, so their thresholds are something you can act on. Tone, Listening, Speaking and Writing are a different case: they are measured on Pro, so on a free account they are withheld by plan rather than by evidence — practice will not open them. The page tells you which of the two you are looking at.
Reading proves you were on the page; it does not prove you could recall anything. Those checkpoints are counted, and you can see them on the Activity page, but only answering — a review, a practice round, a spoken phrase — produces the evidence a mastery figure is built from.
You. A teacher sees aggregate patterns for their own circle and the individual learners they teach; a guardian sees a plain-language view of a learner they manage. Your cognitive-task results are never shown to a teacher, a guardian or an institution. Beyond your circle, only Mandarin Temple staff can reach your record, and only where running the service requires it — support, safety, or administering a study you joined; even then your cognitive-task results stay pseudonymous to us.
Only if you explicitly opt in, on a separate screen, and you can withdraw at any time. Research data leaves with a pseudonym that cannot be joined across studies, carries no free text and no direct identifiers, and a learner whose account is managed by a guardian cannot consent for themselves.
No webcam, no eye-tracking, no emotion or attention inference, now or later. What the Lab records is what you answered, when, and how long it took. That commitment is written into the ethics page and is not conditional on a future product decision.
Yours. It is drawn from your own reviews at the intervals they actually happened — how much you still recalled after a day, a week, a month. It needs real gaps of days and weeks to take shape, so it is the one part of the Lab that cannot be filled by practising harder today.
Yes. Every table the Lab writes to is covered by the account export and by deletion — the evidence log, the per-item state, task results and consent records. Deleting your account removes them rather than retaining them under a research exemption.
There is nothing to set up and nothing extra to sit. The Lab reads the practice you were doing anyway, and a first profile appears at around eight observations.
Signing in takes you straight to your own evidence.
