Most progress dashboards report at subject level, because subjects are what the curriculum is divided into and the numbers look tidy. As a guide to what to do next, it is nearly useless. "Pathology: 61%" does not tell you whether to open a book, and if it did, it would not tell you which page.
A useful map has to be finer than that, and it has to be built from what you actually did rather than what you feel.
Three levels, three different jobs
- Subject. Only useful for time allocation — which of twenty gets next week. Twenty numbers, reviewed monthly.
- Chapter. The level at which you plan a session. Fine enough to open the right material, coarse enough to have sensible sample sizes.
- Topic. The level at which you actually fix something. This is where a misconception lives, and where a single hour changes an outcome.
The trap is sample size. At topic level you may have three attempts, and three attempts tell you almost nothing — one unlucky guess swings the percentage by a third. Treat a topic-level number as a pointer to look, never as a conclusion.
Plan at chapter level. Diagnose at topic level. Allocate time at subject level. Using one number for all three jobs is why dashboards feel useless.
What belongs on the map besides accuracy
- Coverage. How much of the topic you have actually seen. Low coverage and high accuracy is not strength, it is a small sample.
- Confidence calibration. Whether your certainty matches your correctness. A topic where you are consistently over-confident needs different treatment from one where you are simply weak.
- Recency. When you last touched it. A topic at 80% from four months ago is not at 80% now.
- Direction. Whether it is improving. A weak topic that is climbing needs patience; a weak topic that is flat needs a different method.
Reading a topic-level number honestly
At topic level you often have three or four attempts, and at that sample size the percentage is close to meaningless — a single lucky guess moves it by 25 points. Treat a low topic score as a prompt to go and look, never as a measurement. The number's job is to direct attention, and it does that well even when it cannot support a conclusion.
Coverage is the number people avoid
Accuracy is flattering; coverage is not. A candidate three months in is often at well under half the curriculum genuinely attempted, while feeling reasonably prepared because everything they have attempted went acceptably. Coverage is the number that predicts the unpleasant surprise, and it is the one worth checking monthly.
High accuracy over low coverage is not strength. It is a small sample wearing strength's clothes.
Rebuild the map monthly, act on it weekly
Looking at it more often than monthly invites over-reaction to noise; less often and you spend weeks working on a picture that has changed. The weekly decision is narrower: given the map, what gets this week's attention. That separation — slow diagnosis, fast action — is what stops the map becoming something you admire instead of use.
Keeping it honest
A map built only from questions you chose to answer inherits all your avoidance. If you never select forensic medicine, its numbers stay flattering and empty. Mixing in questions you did not choose — random sets, or an app that deals from your weakest areas — is what keeps the map connected to reality.
In Recallab, Blind Spots and Fresh Eyes exist partly for this reason. Pro Learner Insights keeps the longer-term view, with a review of each finished week built from your own results.