It is a reasonable thing to want. Months of effort, an expensive decision riding on the outcome, and a pile of practice data that looks like it should predict something. Unfortunately the honest answer is that it predicts direction well and absolute position badly.
Why a percentile cannot be computed from your practice
- It is relative. Your position depends on a cohort you cannot observe, sitting a paper that does not exist yet.
- Your practice is not a random sample. You chose most of what you attempted, and choice is biased towards what you can do.
- Difficulty is not comparable. One bank's 70% is another's 55%. There is no shared scale.
- Conditions differ. Untimed practice at home is a different task from a locked 42-minute section.
Anything that converts your practice percentage into a predicted percentile is performing confidence, not statistics.
What your data does tell you
Quite a lot, provided you ask questions it can actually answer:
- Direction. Is accuracy on material you have not seen before rising? That is the closest thing to a real progress signal.
- Coverage. What proportion of the curriculum have you genuinely met? Often the most sobering number available.
- Calibration. Does your confidence match your correctness? Improving calibration is real progress even when accuracy is flat.
- Evenness. How wide is the gap between your best and worst subjects? Narrowing it moves a percentile more reliably than deepening a strength.
What a plateau usually means
Accuracy that stops moving for a month is common and is rarely a sign that you have stopped learning. More often it means coverage has widened — you are now attempting harder or less familiar material, so a flat percentage is concealing real progress underneath a changing question mix.
The way to tell them apart is to look at coverage alongside accuracy. Flat accuracy with rising coverage is progress. Flat accuracy with flat coverage is a genuine plateau, and usually means the method needs changing rather than the hours increasing.
Trends need weeks, not days
Daily variation in practice accuracy is dominated by which questions came up, how tired you were, and how much guessing you did. Reading a trend from three days is reading noise. Four to six weeks is the shortest window in which a real change is distinguishable from ordinary variation.
Three signals worth more than a predicted score
- Unseen accuracy rising. Slowly, over weeks. The single best indicator you have.
- Coverage rising. More of the curriculum genuinely attempted. Sobering and actionable.
- The spread narrowing. Your worst subjects catching up. Under a percentile ranking this moves you more than deepening a strength.
A note on anxiety
The desire for a predicted percentile is usually a desire for reassurance rather than for information. A number invented by a tool cannot provide it, and a confident wrong prediction — in either direction — does real harm: complacency or despair, both on no evidence. The honest answer is that you can know your direction and not your destination, and direction is what you can act on anyway.
Accuracy on unseen questions is the honest metric
Overall practice accuracy is inflated by repeats — you have seen those items before, and recognition does a lot of the work. Accuracy restricted to questions you are meeting for the first time is far closer to what an exam measures, and it is usually a good deal lower, which is precisely why it is the more useful number to watch.
Pro Learner Insights holds this longer view week by week and shows the evidence behind anything it claims.