Your Grammar profile shows the same evidence the course generator reads. It is there so you can inspect the system's opinion of you, disagree with it, and get one thing to do next.
A learner-facing view of the model's own estimate is called an open learner model. The largest review of them and of learning dashboards is descriptive: it reports no pooled effect on learning outcomes, and it found learner control and opportunities to act on the model unevenly implemented.
So this page will not tell you that reading your Overview makes you learn faster. What the review supports is narrower: open learner models are the better-evaluated dashboard family, and outcome evidence for dashboards is thin across the board.
The uneven half of that finding is the one we could act on. A later case study of a dashboard that made explicit recommendations reported that learners valued being pointed at specific material rather than at charts. Its sample was small, so we treat it as a design argument and not as proof.
The next action is the point
The Overview's first job is to offer one runnable recommendation: what to do, why that was chosen, and roughly how long it takes. The charts underneath are the supporting evidence for that choice, not the product.Every recommendation names the evidence that selected it, and you can reject it. Picking a different rule from the list is a normal thing to do, not an override of the system.
Check-up score and practice accuracy are not the same number
A check-up score describes one bounded assessment: these items, on that day. A practice accuracy describes your recent independent work across distinct items as you did it. They can disagree without either being broken, because they are answers to different questions.Neither is a grade. Both are reported next to the amount of evidence behind them, which is usually the more useful half.
Recognising a form is not producing one
Recognition evidence means you picked or identified a form from supplied options. Productive evidence means you produced it without options in front of you. Recognition can show that you notice a form. It cannot show that you can use it, so the two are counted and displayed separately.Each rule says which kinds of evidence it actually has: recognition only, productive use only, or both.
What the rule states mean
New and Exploring mean the evidence is still sparse. Nothing stronger appears until a rule has at least four independent items across two separate study sessions, so a single lucky answer cannot produce a confident-looking label.Once a rule has that much evidence, the estimate decides. Needs practice means the estimate is low. Building means it is middling, or high but still without enough productive evidence.Secure additionally requires two source-distinct productive items answered correctly across at least two sessions. A high estimate built only on recognition stays Building, and says so.
Secure is a current estimate about recent evidence. It is not a certificate, and a rule can leave it again.
Review due is about the age of the evidence and nothing else. A stale Secure rule stays Secure with a Review due badge until something new comes in. It does not mean you got worse.
The four-item and two-session thresholds are our reporting rule, not a research finding. They were chosen to be conservative, and the learner evaluation is what will tell us whether the labels read as credible.
Why it is keyed to rules
A single level label does not tell you what to work on. Corpus-derived can-do statements, the approach the English Grammar Profile established, break a level into things you can actually practise. A row can say your dative prepositions are shaky while your past tense is solid, and the next course reads exactly those rows.
Where the estimate comes from
Underneath the labels sits a decayed probability that updates from independent answers, with older evidence counting for less. Models in this family are well studied at predicting answers and much less studied against actual learning. We treat that as a warning rather than a licence, which is why the internal number orders your practice and never appears as a score.Supported work, where the Tutor or the Validator helped, does not enter that function at all. It still counts as study time and exposure.
Getting the most out of it
Start from the recommended action. If you disagree with it, the reason it gives is the fastest way to find out which of you is wrong.
Read the Overview as a to-do list, not as a grade.
Ignore a rule marked with thin evidence until you have answered more items on it.
Check which evidence a rule has before believing its state. Recognition only is a different claim from productive use.
Treat Review due as a calendar entry, not as a demotion.
Under the hood
Each rule carries an estimate, a confidence value, an evidence-coverage kind and a timestamp. The estimate updates on independent check-up items, independent course exercises and independent follow-up attempts, with a 30-day decay half-life.The state labels are computed from that estimate plus the evidence counts. Sparse evidence caps the label regardless of how high the estimate is.The course generator reads the same rows you see. There is no second hidden model behind the visible one.Study time is an estimate of intentional engagement on learning surfaces, with server-side caps so that multiple tabs and retries cannot inflate it. It is not a stopwatch.