Comparing Practice Exam Attempts: When Improvement Is Real
You sat a practice exam, scored 64%, studied for a week, sat it again and scored 78%. Fourteen points. It feels like the clearest evidence you have. It is, unfortunately, the least reliable number in your whole preparation, and the reason is simple enough to be uncomfortable: you had seen the questions before. This page sets out when a comparison between two attempts means something, when it does not, and what to look at instead.
Start the trend with one honest data point
A full 103-question CCNA paper, free, with an explanation and an Exam Tip on every item. It becomes the first point on a trend that only ever compares like with like - and the ranked domains tell you where the next week should go. Create an account on the next screen and the code applies itself.
CCNA-EXAM-FULL
Start the free exam
Studying Security+ instead? Use SECPLUS-EXAM-FULL · Already have an account? Log in to start
On this page
The retake problem
When you sit the same paper twice, the second score is measuring two things mixed together: how much you learned about the subject, and how much you remember about this specific set of questions. Those are different skills, and only one of them is on the real exam.
The contamination is larger than most candidates expect. You do not need to consciously recall an item for it to help. Having seen a question once, the shape of the correct answer is familiar even when the reasoning behind it is not, and distractors you have already rejected once are rejected faster the second time. The effect is strongest exactly where you are weakest, because those are the questions you spent time on afterwards.
Which produces the trap: the second score always goes up. Up is what it does. A number that rises whether or not you learned anything cannot tell you whether you learned anything.
The four comparability rules
A movement between two attempts is only reported when all four of these hold. When any one fails, the comparison is refused rather than approximated, because a refused comparison costs you nothing and a wrong one sends you to study the wrong thing.
Same exam
Two different papers of the same certification are different instruments, even at the same length. One may lean harder on subnetting, the other on OSPF. Comparing their totals produces a difference that is mostly about which paper you drew.
Same domain map
If the question-to-domain mapping was rebuilt against a new blueprint version between the two sittings, per-domain figures are no longer measuring the same categories. The delta would be an artefact of the taxonomy changing, not of you changing.
Both covered the whole paper
A full paper is never compared against a capped preview. A 9-question preview and a 103-question sitting differ by sample size alone, and putting the two on one line manufactures a swing that nobody made. Partial sittings are counted and reported separately, never plotted.
Enough signal to say anything
If every domain came back at the same mastery, the ranking carries no information and is suppressed rather than shown. A list that looks like a finding but is really a tie is worse than no list, because it will send you somewhere at random with the confidence of a recommendation.
Only full sittings are plotted on the trend, and the window is the last 24 of them. Anything older is history rather than trajectory.
Compare domains, not totals
The total is the least informative number available, because it mixes six or seven separate readings into one and hides which of them moved. Here is the same pair of CCNA sittings, read two ways:
| Domain | Weight | Sitting 1 | Sitting 2 | Move |
|---|---|---|---|---|
| IP Connectivity | 25% | 50% | 54% | +4 |
| Network Fundamentals | 20% | 79% | 83% | +4 |
| Network Access | 20% | 75% | 80% | +5 |
| Security Fundamentals | 15% | 64% | 93% | +29 |
| IP Services | 10% | 93% | 93% | 0 |
| Automation | 10% | 75% | 88% | +13 |
The headline is that readiness went from 70% to 78%, eight points, and that reads like a good week. The domain view says something much more specific and much more useful: the week went almost entirely into Security Fundamentals and Automation, which together are 25% of the exam, while IP Connectivity - a quarter of the exam on its own, and still the worst domain at 54% - barely moved.
This candidate has been studying the material they find satisfying. Eight points of readiness is real, but they are no closer to solving their actual problem than they were a week ago, and the total was never going to tell them that. Next week goes into routing whether they enjoy it or not.
Across the pack: latest, never best
Once you have sat more than one paper of a certification, a second question appears that no individual attempt can answer: where do you stand on the certification, rather than on any one paper?
That is answered by pooling per-domain evidence across every paper of that certification you have attempted, then running the same blueprint weighting over the pooled totals. One rule keeps it honest, and it is worth stating plainly because it is the opposite of what most progress screens do:
Your latest sitting of each paper contributes, not your best
If you sat paper 1 three times, the most recent attempt is the one that counts. Cherry-picking your best historical sitting of every paper would produce a flattering number describing a version of you that no longer exists. The pooled view is meant to report where you stand today, which sometimes means it goes down.
Pooling also fixes the small-sample problem properly. A single CCNA paper might only ask you eight Automation questions - too thin for a verdict, however confident the percentage looks. Across four papers that becomes thirty-odd and the reading firms up. The ranking gets more trustworthy the more distinct papers you sit, which is the exact opposite of how a repeated single paper behaves.
Retired papers are a deliberate special case: they still produce a study path, because somebody who sat one deserves a diagnosis from it, but they are excluded from "how much material exists on this domain", since you cannot go and drill them.
One full sitting, and the trend has something to sit on
90 questions against CompTIA's current SY0-701 objectives, free. Every full sitting after it is compared like with like, per domain, with the movements that are real separated from the ones that are sample size.
SECPLUS-EXAM-FULL
Redeem the code
Already have an account? Log in to redeem
The ideal number of papers to sit
The usual answer is a number, and the number is wrong because it is measuring the wrong thing. What matters is not how many sittings you have logged, it is whether each domain has a sample large enough to trust. So the ideal target is a sample size, not a count.
A practical target: enough distinct papers that your smallest domain has roughly twenty questions behind it. On a CCNA blueprint where Automation is 10% of a 103-question paper, that is about eight questions per sitting, so three or four different papers gets you there. On Security+, where the lightest domain is 12% of 90 questions, two or three will do it.
Ordered by what actually helps:
- A new paper you have not seen. Uncontaminated, and it adds sample to every domain at once.
- A targeted repeat of one paper, once, to check a specific domain you worked on. Read the domain, not the total.
- A second repeat of the same paper. By this point you are mostly measuring memory. It is not useless - it can build timing and stamina - but it is not a readiness reading.
And the honest limit of all of it: a practice exam measures the shape of your knowledge against a published blueprint. That is a genuinely useful thing to know and it is not the same as a prediction of exam-day performance, which no practice exam can honestly give you. For where the thresholds sit and what else is worth checking before booking, see am I ready for the CCNA or am I ready for the Security+.
Frequently asked questions
Should I retake the same practice exam?
Once, deliberately, to check the specific gaps you set out to close - then move to a different paper. A second sitting of the same paper measures how much of that paper you remember, and recall of 103 specific items is a different skill from answering unseen questions on the same topics. The score rises either way, which is precisely why the rise tells you nothing.
When can two attempts honestly be compared?
Same exam, same domain map, both covering the whole paper. Change any one and you are comparing different instruments. A full paper against a short preview would differ by sample size alone, and that difference would be reported to you as progress you never made.
My score went up but my readiness barely moved. Why?
Your gains landed in lightly weighted domains while a heavy domain stayed flat. The score counts questions equally; readiness weights domains by their share of the real exam. When they disagree, readiness is the more honest reading, and it is telling you the study time went somewhere cheap. Open the per-domain view and you will usually find one heavy domain that did not move.
What is the ideal number of practice exams before the real thing?
The ideal target is a sample size, not a count of sittings: enough distinct papers that your smallest domain has around twenty questions behind it. On CCNA, where the lightest domain is 10% of a 103-question paper, that is three or four different papers. On Security+, where the lightest is 12% of 90, two or three will do. Repeated sittings of one paper do not count toward that target in any useful way.
What is the best way to track practice exam progress?
Track blueprint-weighted readiness per domain across distinct papers, not the total score across repeats of one paper. The total mixes six or seven readings into one number and hides which moved - it can rise eight points while your heaviest domain has not moved at all, as in the table above. The best progress view is per-domain, uses your latest sitting of each paper rather than your best, and refuses to compare a full paper against a short preview.
What is the most efficient way to use practice exams?
Sit one early, before you feel ready, and treat it as a measuring instrument rather than a verdict. The most efficient pattern is: measure, fix the top two ranked domains, measure again on a different paper. Candidates who postpone the first paper until they feel prepared spend most of their study time spreading effort evenly across material they already know, which is where the bulk of wasted preparation goes.
Is a five point improvement meaningful?
On a fresh paper of the same certification, worth noticing. On a repeat of the same paper, close to meaningless. And on any single domain measured by fewer than about ten questions, five points is well inside the noise, which is why domain rankings break ties toward the larger sample rather than treating all readings as equal.
Why does my pooled readiness sometimes go down?
Because it uses your latest sitting of each paper rather than your best. If a recent attempt went badly, that is what gets reported. A progress view that only ever went up would be a motivational graphic rather than a measurement, and you cannot plan against a motivational graphic.