A plateau is easy to recognise and, in its way, easy to accept: the numbers sit still and you know something has to change. Inconsistency is harder to live with. You sit three papers in a fortnight and they come back noticeably different from one another, with no obvious reason for the spread. The temptation is to decide which one is the real you — usually the best result on a good day, the worst one at three in the morning — and to treat the others as noise. That instinct is understandable and it throws away the most useful thing you have. The spread between your attempts is itself a measurement, and it usually has a specific, fixable cause.
Part of the swing is arithmetic, not ability
Before diagnosing anything, it helps to know how much movement to expect from a test this short.
A Core subtest runs to roughly twenty items in about twenty-five minutes — confirm the current structure against the official sources, since the format can be revised between cycles — and on a test of that length a small number of questions carries a lot of weight. Two questions you guessed and happened to get right, plus one you knew perfectly well and misread, is a three-question difference. Nothing about your reasoning changed between those papers; the dice landed differently. Any single mock therefore gives you a point estimate of something that is really a range, and a candidate who is genuinely stable will still produce results that vary from sitting to sitting.
This matters most when you are tempted to act on one result. A single strong paper is not evidence that a weakness has been fixed, and a single poor one is not evidence that your preparation has collapsed. It also affects how you read the result itself: because the dMAT reports a scaled score and a percentile rank rather than a raw mark, a modest movement in questions can look larger than it is once it has been expressed relative to a cohort. Judge yourself on a band drawn from several papers rather than on your most recent one.
Most of the rest is the conditions
Once you allow for that, the majority of unexplained variance in practice scores comes from the fact that the papers were not really comparable.
Very few candidates hold their conditions constant. One mock is sat on a Sunday morning at a desk, fully awake, in one sitting. The next is squeezed into an evening after a full day of classes, on a laptop in a noisy room, with two short pauses that never quite got counted. A third gets attempted with rough paper on the table, which the real exam does not allow. Those are three different tests, and comparing their scores tells you about your circumstances rather than about your ability. This is the whole argument for simulating real exam conditions: consistency of setup is what makes the numbers mean anything.
Fatigue deserves a separate mention, because it produces variance that looks like inconsistency but is really stamina. If some of your papers are single subtests and others are full-length sittings across both modules, the full ones will tend to come in lower, and the gap will sit almost entirely in whatever you attempted last. That is not random at all. It is a predictable decline that only appears when a paper is long enough to expose it.
Reading the spread instead of the average
When the conditions have been held reasonably steady and the results still move around, the pattern of the movement is the diagnosis.
Look underneath the total first. A stable overall score can hide considerable churn — a strong figures section carrying a weak equations section on one paper, and the reverse on the next — and a total that swings can turn out to be one volatile subtest with everything else steady. Scoring each subtest separately across several attempts usually localises the problem quickly, and if it points at one area, the question becomes one of balancing your weak and strong subtests.
Then look at how the marks were lost, not just how many. The distinction worth making is between questions you never reached, questions you got wrong but understood immediately on review, and questions you could not have got right even with more time. The first is pacing, the second is carelessness or working memory, the third is method. Scores that swing widely are usually driven by the first two, because method changes slowly and predictably while pacing and attention can differ substantially from one morning to the next. Keeping an error log across attempts is what lets you see which of the three is actually moving, and a proper mock review is where that gets recorded.
When inconsistency is the thing to fix
Sometimes you do all of this and the spread stays wide. At that point, the inconsistency is not measurement noise obscuring your real level — it is your real level, and it is worth treating as the target.
Wide variation under controlled conditions almost always comes from decisions rather than knowledge, and specifically from what you do when a paper starts going badly. A candidate who falls behind and responds by speeding up produces a very different result from one who falls behind and holds their pace, even though the two have identical ability. The fix is to make those moments routine rather than improvised: a fixed rule for how long a question gets before you abandon it, a consistent approach to pacing, and enough repetitions that the decision does not have to be invented under pressure. Consistency is trainable, and it is trained by removing choices from the moments when your judgement is worst.
What to do with your next paper
Change one thing before your next mock: fix the conditions. Same time of day where you can manage it, full length, no pauses, no rough paper, screen rather than print. Then sit two or three under that setup — the free practice test works for this — and record the subtest scores and the three error types rather than only the total. What you are looking for is not your best result. It is the width of the band and where in the paper it comes from, which is a far better guide to what to work on next than any single number the papers have given you so far.
