Score prediction is the strongest claim a question bank can make, and the category's language hides how strong. A readiness indicator, an internal composite of your coverage, accuracy and mock trend, is a legitimate, useful instrument. A predicted examination score is a forecast of an external, equated, standard-set result, and it earns belief only through validation evidence: real users' predictions matched against their real examination outcomes, with the error honestly reported. Almost no provider publishes that evidence, and this page sets out what the claim requires, how to read the versions of it currently on the market, and, for symmetry, where our own readiness score does and does not stand.
What prediction would actually require
Four components, none optional. Training and outcome data: predictions must be built and tested against actual examination results from real candidates, not against internal mock scores, because the examination's equated scale is precisely what internal percentages cannot reach, /blog/why-qbank-percentages-are-not-comparable. Bias handling: the users who report outcomes are a selected population, and repeat question exposure inflates the inputs; both must be modelled or the forecast learns the wrong lesson. Calibration and discrimination reporting: predicted versus observed outcomes across the range, with intervals, so a "you will score 480" arrives as "480, plus or minus this much, and here is how often we are that wrong". And maintenance: examinations change format and standard, the MRCP's move to the 450 equated mark being this year's example, banks change content, and a prediction validated in one regime drifts silently in the next; a dateless validation is a lapsed one.
Reading the market's claims
Applying the standard to the public landscape, with evidence labels doing the work. Some providers make expressly modest claims: BoardVitals, for instance, describes its adaptive-testing score as intended to identify strengths and weaknesses rather than predict a board-examination pass, which is exactly the honest framing. Others advertise predicted score bands, MSRA Prep among them, without, as far as public materials show, a published calibration report; such features should be recorded as vendor-reported until the validation appears, which is a description of the evidence state, not an accusation. And timing-precision spaced-repetition language, "just before you would forget", belongs in the same file: a claim about a model's accuracy, published without the model's accuracy. The buyer's question is identical in every case: predicted against what outcome data, with what reported error?
Where iatroX's readiness score stands, by its own standard
Stated plainly because this page's standard would be worthless applied only outward: the iatroX readiness score is a composite of curriculum coverage, weighted topic accuracy and mock-performance trend. That makes it an internal readiness indicator, a structured answer to "how prepared am I, on this platform's evidence", and it is not a validated external score predictor, because we have not yet matched it against users' real examination outcomes at scale, and until we have, we will not market it as prediction. The indicator's honest uses are real: watching your trend, finding uncovered blueprint, catching weak-domain floors; the dashboard framework at /blog/why-qbank-percentages-are-not-comparable is how we recommend reading it, and validation against real outcomes is on our observatory programme's list, results to be published whichever way they run.
What learners should do with readiness metrics meanwhile
Three habits extract the value without the over-belief. Use trends, not points: a readiness figure's movement over weeks is informative where its absolute value is not. Triangulate: readiness composite, unseen-item performance, and full timed mocks are three imperfect instruments whose agreement means something and whose disagreement means more. And let the metric direct work rather than deliver verdicts: its best function is pointing at the uncovered domain or the decaying topic, and its worst is being read as a forecast it never earned. A number that changes what you practise tomorrow is doing its job; a number that tells you your future is doing marketing's.
Frequently asked questions
Has any provider published real calibration evidence?
Publication is rare and the situation moves; the durable skill is asking for the calibration report by name, and reading its absence as the answer.
Are mock scores better predictors than readiness composites?
Full timed mocks at true format are the best internal signal most candidates can get, and they still sit on the bank's difficulty, not the examination's equated scale; stability across mocks, not any single figure, is the readable message.
Would you publish an unfavourable validation of your own score?
Yes, and the commitment is only worth what the publication record makes it; that is what the observatory series exists to establish.
Why do candidates want predictions so badly?
Because uncertainty before a high-stakes examination is genuinely painful, which is exactly why the market supplies confident numbers; the mature substitute is triangulated readiness plus honest intervals, less comforting and considerably more true.