Updated 10 October 2026.
AMBOSS Knowledge Profile addresses a reasonable frustration with question-bank revision: an overall percentage correct does not necessarily tell you what you understand, what you should practise next or how prepared you are for an examination. A learner working through unfamiliar, difficult questions may look weaker than someone repeating material they already recognise. A useful dashboard should distinguish those situations rather than reward whichever activity produces the most reassuring number.
The important qualification is that a more sophisticated estimate remains an estimate. Knowledge Profile should help organise revision, not become an authority that overrules the examination blueprint, independent assessments or evidence that you cannot explain a concept without help.
What changed in October 2026?
AMBOSS's 5 October announcement describes a knowledge-tracing model that uses question responses and item difficulty to estimate topic-level ability. Its adaptive sessions began in April 2026; October was not their initial launch. The announcement lists passing probabilities for USMLE Step 1, Step 2 CK and COMLEX Levels 1 and 2, with score estimates for Step 2 CK and COMLEX Level 2. Some support pages describe narrower availability, so the selected objective and current account interface matter.
That distinction prevents a common misunderstanding. The existence of a general learning profile does not establish that every examination on a platform has a validated score forecast. A feature can help choose questions for several objectives while providing an exam-specific prediction for only some of them.
EPC is not simply your percentage correct
AMBOSS's Analysis documentation distinguishes Equated Percent Correct, or EPC, from historical accuracy. Depending on the objective, EPC concerns expected performance on a standardised examination or the relevant AMBOSS question pool. Metrics appear after at least 40 relevant answers. Topic relevance reflects representation within the bank, while a focus area can be a relative weakness rather than an absolute deficiency.
Imagine two hypothetical candidates who both answered 70% of their questions correctly. One worked through unseen questions under timed conditions. The other repeated a familiar set, checking explanations between attempts. Their identical historical percentages should not be interpreted as identical evidence of readiness. Difficulty, familiarity, sampling and testing conditions all matter to the interpretation, even when a dashboard cannot fully account for every difference.
EPC attempts to offer a more informative signal than a simple tally. However, the user should still ask what the model has actually observed. An estimate based on a narrow selection of topics cannot remove the need to inspect curriculum coverage. Forty answers may unlock a display; it is not a recommended preparation volume or a certificate that uncertainty has disappeared.
The live estimate and Score Predictor answer different questions
The Knowledge Profile product explanation separates routine question-bank estimates from the standalone Score Predictor. The latter uses practice-assessment results and timing information rather than treating ordinary question-bank activity as the same input.
More specifically, the Score Predictor documentation describes a forecast for a target examination date, assuming preparation continues, with a range intended to cover 95% of outcomes. It requires at least two relevant practice-exam results, including one within four weeks of the target date. The live Analysis score range, by contrast, is described as covering 68% and concerns present-day performance.
It would therefore be a mistake to compare the widths of the two ranges and conclude that the narrower one must be more precise. They use different inputs, time horizons and coverage levels. A narrower interval can simply express less coverage. The first question is not whether a forecast feels encouraging, but what event and timeframe it is actually forecasting.
This also explains why a strong future forecast should not be read as permission to stop preparing. Its interpretation depends on the conditions under which it was generated. Changing the study plan substantially may change the relevance of the original projection.
A knowledge profile is still an incomplete view of knowledge
AMBOSS's August explanation says that the profile is built from question responses and does not yet incorporate Library use or AI Mode interactions as evidence. October's announcement describes closer personalisation of AI recommendations and future expansion of profile inputs. Those are different directions of information flow: a profile can inform a conversation without the conversation already updating the calibrated profile.
This is worth watching because learners do more than answer multiple-choice questions. They discuss uncertainty, identify missing facts, interpret feedback and practise explaining decisions. A question response may reveal that something went wrong without distinguishing whether the cause was factual recall, reading the stem, applying the wrong jurisdiction or changing a correct answer unnecessarily.
The profile should therefore be treated as a starting hypothesis about where to investigate. Calling a topic weak is useful only when the next activity addresses the particular misunderstanding. Otherwise, an adaptive system risks providing more exposure without resolving why the earlier exposure failed.
How to use the dashboard without letting it take over revision
A practical approach is to review the profile at a planned interval rather than after every question. Select a small number of actionable weaknesses, inspect actual errors and decide what kind of work each requires. Some errors need a clear explanation; others need fresh application, timing practice or a later check without prompts.
Keep a separate view of the examination blueprint. A model may be confident about the topics you have practised while relatively uninformed about those you have avoided. A high estimate is not a reason to let an unvisited curriculum area disappear from the plan.
Preserve some fresh, mixed practice under the conditions you intend to assess. Targeted work asks whether a particular gap is improving. Mixed work asks whether you can recognise the relevant problem without being told the topic first. Those are complementary questions, not competing philosophies of revision.
The same restraint applies to iatroX. A progress display or adaptive recommendation should not be presented as a validated probability of passing unless evidence supports that specific claim. Personalisation is useful when it improves the next decision; it does not need to pretend that it can predict every eventual outcome.
Where iatroX fits alongside Knowledge Profile
iatroX's Socratic Tutor offers a different entry point: discussion of the reasoning behind an attempted question, with the selected examination context retained. That makes it relevant when the problem is not discovering that a topic needs work, but understanding the distinction that keeps producing mistakes.
A learner could use a weakness identified elsewhere to choose corresponding practice within the iatroX examination catalogue, then explain their reasoning and attempt another relevant question. This is a proposed study workflow, not an automatic import of AMBOSS data. There is no need to copy proprietary question text between services to practise the same underlying concept.
For uncertainty involving clinical guidance rather than an exam distractor, askiatroX provides a separate reference route, described in iatroX's methodology. Checking a source and demonstrating unaided recall should remain visibly different activities.
iatroX's development direction is to connect these activities more closely across examination preparation and continuing professional learning. That is an ambition for a more continuous learning experience, not a claim that every activity already feeds a single validated knowledge score. The useful competition with AMBOSS is over the quality of the next learning step, not simply the appearance of the dashboard.
Frequently asked questions
Is AMBOSS EPC the same as my raw question-bank percentage?
No. EPC is a model-based estimate with a meaning that depends on the selected objective; raw accuracy describes answers already recorded. Their differences are explained in the Analysis guide. Neither should be interpreted without considering coverage and practice conditions.
Can I use the Knowledge Profile score range like the separate Score Predictor range?
Not interchangeably. They concern different inputs and time horizons, and their stated coverage levels differ. The Score Predictor guidance should govern interpretation of that particular tool rather than assumptions drawn from another dashboard.
Does iatroX offer an AMBOSS-equivalent validated score prediction?
This comparison does not establish one. iatroX's adaptive practice, tutoring and planning should be judged as learning tools, not described as an equivalent exam-score forecasting model without published validation for that purpose.
Turn an identified weakness into focused practice with iatroX's exam-specific learning tools.