Learning Lab · For institutions
实事求是 · shí shì qiú shì — “Seek truth from facts”
Whether the programme is working, in evidence rather than in engagement. Seats actually activated, learners active in the window, median mastery and retention, and each circle read against its own earlier window — with the governance to stand behind the numbers.
The Lab itself sits behind a sign-in, because it shows real learners. Everything it does is described on this page, and the full record of what is collected is on the ethics page.
Why it exists
Usage reporting answers whether a licence is being consumed. It has never answered whether anyone learned anything, which is the question a renewal actually turns on.
Minutes, logins and completions all rise when a cohort is busy. None of them move differently when a cohort is busy and learning nothing, which makes them useless as an outcome measure.
Attainment is measured from retrieval evidence and reported separately from usage — never blended into one “engagement” score.
A cohort that arrives stronger finishes stronger. Comparing groups on raw attainment mostly ranks admissions, and it penalises exactly the teachers who take the hardest classes.
Each circle is compared with its own earlier window rather than with other circles, so the movement reported is the group’s own and not a reflection of who enrolled.
A healthy mean can contain a quarter of a cohort quietly stalling, and by the time it shows in results the term is over.
Patterns surface by cause and by cohort, early enough to change what happens next term rather than to explain it afterwards.
What actually happens
Four steps, in order. Every figure is traceable to the evidence beneath it, and the last step is allowed to refuse.
Ordinary practice inside the product. No separate assessment to administer, no proctoring, and no additional workload for teaching staff.
Appended, never edited, carrying the facet it exercised. An append-only log is what makes a later figure auditable rather than merely plausible.
Circles aggregate into programmes. A research cohort is frozen at enrolment, so a learner joining mid-term cannot retroactively rewrite last month’s numbers.
Adoption and outcomes: seats activated, learners active in the window, median mastery and retention. Below the anonymity floor a figure is withheld with the reason, rather than published in a form that identifies individuals. Intervention load and programme return are not computed yet, and are not shown.
What you do not get
The five optional cognitive tasks are research, they are never inserted into lessons, and their results are shown to the learner alone — not to a teacher, not to a guardian, and not to anyone on your staff, not even as a cohort average. Nobody at your institution sees what a learner wrote in their own words either.
What an administrator does see is programme-level roll-ups across your circles and, for any circle in your institution, the same cohort and individual-learner view its teacher has.
What gets measured
The reason a cohort can pass reading assessments and stall in speaking. Reporting one blended attainment number hides the specific capability a programme is failing to build.
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, on each learner’s own plan rather than on their teacher’s or their parent’s.
Create a free account to see what it measuresHow to trust it
Every figure carries its evidence count and a confidence band, and a cohort below the anonymity floor is withheld with the reason given. A number you cannot defend is worse than a gap.
There is no staff league table, and there will not be one. Circles are listed alphabetically with no rank in the data, and each one’s movement is read against its own earlier window, because a table that rewards intake quality drives exactly the wrong behaviour.
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
Programme-level roll-ups across your circles — seats activated, active learners, median mastery and retention — and, for any circle in your institution, the same cohort and individual-learner view its teacher has. What no one on your staff can see, at any level, is cognitive-task results or anything a learner wrote in their own words.
No, and it is built so it cannot be repurposed to. There is no staff comparison view, circles are returned alphabetically, and there is no rank field to sort on. Where movement is reported it is a circle against its own earlier window, because ranking on raw attainment mostly ranks admissions.
Because the cohort behind it was below the anonymity floor, or the evidence behind it was below the metric’s threshold. In both cases the Lab states the reason rather than publishing a number that identifies individuals or overstates its own certainty.
Three separate tiers. The learning record is the service the learner asked for. Product analytics is cookie-consented. Research is explicit, separate and revocable — and a managed minor cannot consent for themselves; their guardian does.
Learner exports are covered by the standard account export. A research dataset is event-level and pseudonymous — participant ids are salted per study so the same person cannot be joined across two of them, timestamps are hours since that participant’s own enrolment rather than wall-clock dates, and there is no name, email or free text in it. It is produced by us on request rather than downloaded from your own console, and it carries the study’s registered hypothesis, primary outcome and arms alongside the rows.
No. Everything is read from practice learners already do inside the product. There is nothing to administer, schedule or invigilate, and no additional workload on teaching staff.
No webcam, no eye-tracking, no emotion or attention inference, now or later — including for proctoring. That is a standing commitment, not a current configuration.
Yes, with pre-registration: the hypothesis and the primary outcome are frozen before the study starts and cannot be edited afterwards. Assignment is deterministic and reproducible, and every participant consents individually even when a circle opts in.
Sign in to open the Lab, or start with a pilot: one real cohort, your own roster, and an adoption report at the end. Pricing, seat counts and the pilot are on the institution plans page.
