If you want to know how you will score on the real exam, your raw question-bank percentage is one of the worst numbers to trust. For the USMLE, cumulative UWorld percentage correlates with the real Step 2 CK score at only around r = 0.6 to 0.7, while a timed UWorld self-assessment correlates at around r = 0.85 to 0.88. The lesson generalizes: dedicated, timed, exam-like assessments predict performance, and loose percentages do not. Here is what the data actually shows and how to read your own numbers honestly.
Key takeaways
- Raw question-bank percentage is a noisy, selection-biased predictor, roughly r = 0.6 to 0.7 for Step 2 CK.
- Timed self-assessments predict far better: UWSA 2 correlates around r = 0.85 to 0.88.
- Take predictive assessments late, under real conditions, and trust the less flattering number.
- For UK and other exams with fewer formal predictors, timed mock exams are the equivalent signal.
- What moves your score is targeting real weaknesses, not admiring a high percentage.
Why is raw question-bank percentage a poor predictor?
Because a percentage is contaminated by how you generated it. Your cumulative question-bank score mixes early blocks when you knew less with later ones, timed with untimed sessions, first-pass with review questions, and an uneven spread of subjects. That is why two students sitting at 62 percent can end up 15 points apart on the real exam. In the USMLE data, cumulative percentage lands around r = 0.6 to 0.7 with the real Step 2 CK score, and students routinely overvalue it because a high percentage feels reassuring. It is a useful learning signal, but a weak forecasting one.
What predicts your real score well?
Dedicated, timed, exam-like self-assessments. For the USMLE, the standout is UWorld Self-Assessment 2, which correlates with the real Step 2 CK at around r = 0.85 to 0.88 and, taken 7 to 14 days out under full timed conditions, usually lands within about 5 to 8 points of the real score, with a slight optimistic bias of a few points. NBME forms, especially the newer ones, are also strong anchors and tend to run a little below the real score. The pattern is consistent: an assessment built to mirror the real exam, taken under real conditions, predicts far better than any running percentage.
How should you read your practice numbers honestly?
Follow a few rules the data supports. Take your predictive assessments late in preparation, when they reflect where you actually are, not where you started. Sit them under exam-like conditions, timed and uninterrupted, because a relaxed, paused practice test overstates you. When two assessments disagree, weight the timed self-assessment and the recent NBME over your question-bank percentage, and be willing to believe the less flattering number, since the evidence says it is usually the more accurate one. And recognize diminishing returns: once two or three recent scores cluster, a fourth number rarely tells you anything new and mostly feeds anxiety.
What does this mean for UK and other exams?
The predictor ecosystem is richest for the USMLE, with NBME forms and UWorld self-assessments. UK exams like the UKMLA and PLAB, and pathways like the AMC and MCCQE, have less formal score-prediction machinery, but the underlying principle holds exactly. For these exams, timed mock exams under real conditions are the equivalent signal, and your loose question-bank percentage is just as unreliable a forecaster as it is for the USMLE. Whatever the exam, trust the assessment that most closely reproduces the real thing, sat the way you will sit the real thing.
The deeper point: prediction is not preparation
Chasing a predicted number is not the same as improving. The reason timed assessments predict well is that they surface what you cannot yet do under pressure, and the value is in acting on that, not in the number itself. This is where a tool that maps your actual weaknesses earns its place. iatroX uses semantic adaptive learning to identify the underlying concepts you keep missing, rather than only reporting a percentage, so your study time targets the gap instead of padding a score. It covers UK and international exams and is £29 per month or £99 per year with free sample questions. Try the free questions, and for exam-specific difficulty context see our USMLE pass rates and difficulty guide.
Frequently asked questions
Does my UWorld percentage predict my Step 2 CK score? Only weakly. Cumulative percentage correlates around r = 0.6 to 0.7 and is biased by how you used the bank. Use a timed self-assessment for a real forecast.
What is the best Step 2 CK score predictor? UWorld Self-Assessment 2 has the highest single correlation, around r = 0.85 to 0.88, with a slight optimistic bias. Recent NBME forms are also strong and tend to run a little low. Use both.
When should I take predictive assessments? In the last one to three weeks of preparation, under full timed conditions. Taken early or untimed, they misrepresent where you are.
Which number should I believe when they disagree? The timed self-assessment and the recent NBME over your question-bank percentage, and usually the less flattering number, which tends to be the more accurate predictor.
Do UK exams have score predictors like the USMLE? Not to the same extent. For the UKMLA, PLAB, AMC, and MCCQE, timed mock exams under real conditions are the equivalent signal, and the same caution about raw percentages applies.