Sign up free

A mock exam that predicts the real one: how to build and read it

How to set a mock exam that forecasts the real one: what studies say about mocks and predicted grades, the blueprint, the timing and an honest margin.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Teaching from the results · · 13 min read

How do I set a mock exam that predicts how students will do in the real exam?

A mock exam predicts the real one best when it follows the real paper's blueprint, length and timing, uses questions nobody has seen, is sat under exam conditions and is marked to the real mark scheme. Give the forecast as a range: teachers' predicted grades run high, an early mock mark runs low, and one student's result is far less certain than a class's.

A mock exam does three jobs at once: it forecasts the result, rehearses the conditions and shows what to revise. A mock built for one does the others badly, which is why mock marks so often surprise everyone on results day.

This guide is for a teacher, tutor or head of department who wants the mock to say what the real exam will bring. Each source was read on 10 October 2026; "abstract read" marks one where only the abstract was open.

EdProcto publishes this guide. It runs a mock under a time limit held on its server and sets out the figures in its results. The forecast is yours to make.

What is a mock exam for, and why do its three jobs pull apart?

A mock is a forecast, a rehearsal and a diagnosis, and each wants a different paper on a different day.

  • A forecast wants the whole paper, unseen, late in the course, with no marks back until everyone has sat.
  • A rehearsal wants the real room, clock and rules, and is worth repeating.
  • A diagnosis wants to be early, with enough questions on each topic to trust that topic's mark, and feedback at once.

So they pull apart. A later mock forecasts better and leaves less time to act; a real paper asks too few questions on some topics to diagnose them; a forecast holds back marks that revision needs. Decide which job comes first and tell the students. The table is our recommendation.

The mock is mainly When The paper What goes back In EdProcto
A forecast Course nearly taught Full, unseen, real blueprint A range, once all have sat The Exam preset in EdProcto
A rehearsal More than once Real format and minutes Notes on pace "Minutes overall" as the real paper's
A diagnosis Early enough to reteach More questions a topic Marks by topic, at once A topic on every question

How well do mock exams and predicted grades forecast the real result?

Well for a class and poorly for one student, and the two err in opposite directions: teachers' predicted grades run high, and an early mock mark runs low.

Predicted grades run high

Beynon and Thomson (2025) set about a million predicted GCSE grades, collected in May and June, beside that August's results, for 140 thousand pupils in 740 secondary schools. Averages were close: where they differed, predictions were nearly always the higher, by 0.4 of a grade at most, and 78% of predicted grades were a 4 or above against 74% of results. Pupil by pupil they were not: in all but four subjects fewer than half of the predictions matched the result, and in English literature 34% did. The top grade ran the other way, with 4.3% of predictions a 9 and 5.2% of results. The authors caution that schools predict in different ways.

At A level the gap looks wider: UCAS (n.d.) reports around half of UK 18-year-olds predicted AAA or better in 2025, and 26% of accepted applicants achieving it. Murphy and Wyness (2020), whose headline figures are in our guide to proving progress to parents, show who is missed: high-attaining students from disadvantaged backgrounds were significantly more likely to be predicted too low (abstract read).

An early mock mark runs low

Chisnall et al. (2015) followed 774 second-year students at a London medical school who sat a three-station practice, under exam conditions, two months before a clinical exam of sixteen stations. Of them 46.6% passed the practice and 85.0% the real exam. Of those who passed the practice 92.5% passed for real; of those who failed it, 78.5% did too. A pass in the mock held, and a fail forecast little.

That is one study of one clinical exam. We found no published study that follows a school's mock marks through to its exam results.

Does a mock sat under exam conditions tell you more?

Yes, about readiness, on the little evidence we read. In a fully online introductory physics course, Zhang et al. (2023) changed the practice exams to one attempt a question with the feedback delayed, and the changed format gave a more accurate reading of how ready students were (abstract read).

What do the regulator and UCAS say about mocks and predicted grades?

They say a mock should be sat like the exam, and that a predicted grade is a hopeful one. Ofqual, the exams regulator in England, asks schools to have students sit mocks in exam conditions and to keep the papers in case exams cannot go ahead; one set is enough (Office of Qualifications and Examinations Regulation, 2025). Its guidance, first published in 2023, spells the conditions out:

  • questions not known beforehand, and no books or notes;
  • independent work, supervised, though outside invigilators are not required;
  • the real paper's time, with the adjustments a disabled student will have in the summer;
  • papers like the real ones, marked to the board's mark schemes, to one standard across teachers.

Students should get feedback, and a teacher may tell a student the grade a mock indicates. For the standard at each grade Ofqual points to papers from 2019 and earlier and from 2023 onwards, the boundaries between having been more lenient.

UCAS (n.d.) defines a predicted grade as the grade a school believes an applicant is "likely to achieve in positive circumstances", and asks that it rest on evidence such as internal assessments. So a predicted grade is hopeful by definition, and a mock is part of its evidence, not the prediction.

What do students expect, and what does a mock do to anxiety and revision?

Students expect more than they score, a mock seems to calm them, and by itself it changes their plans more than their revision.

Students' own forecasts

In two semester-long studies, Miller and Geraci (2011) found college students' predicted exam scores almost always above the score earned, most of all among low performers. With concrete feedback and extra credit for accuracy, those students' predictions improved and their scores did not (abstract read).

Test anxiety

Lima et al. (2026) randomised 67 medical students at one Brazilian university; 28 took part in a mock clinical exam, setting it or sitting it, a week before the real one. Straight after the real exam their state anxiety averaged 54.8 on a scale of 20 to 80, against 63.6 for the rest, and exam scores did not differ significantly. Seven students allotted to the mock could not attend and joined the comparison group, which weakens the design. In Chisnall et al. (2015), 56.8% of the 308 students who answered a survey said the practice had left them less anxious. We found no controlled study of a written school mock and anxiety.

Revision afterwards

Zhang et al. (2023) added a mock a week before the exam: it changed what students intended to study, and not how they studied or their sense of what they knew. A mark by itself is not enough: the work has to be set.

How do you build a mock that forecasts?

Copy the real paper in everything but its questions.

  1. The same blueprint. Take the topics and their marks from the specification and a specimen paper, and keep the share of each question type (writing an exam blueprint).
  2. The same length and timing. A full paper in the real minutes, with agreed extra time. Time limits covers what a tight clock does to a mark.
  3. Questions nobody has seen. A past paper the class has practised forecasts their memory of it. Use one they have not met, sat as its board's terms allow, or write new questions to the blueprint. Paste a board's paper into a tool only where its terms say you may.
  4. Marked to the real scheme, with a second marker on a sample (moderation).
  5. The real conditions, as Ofqual lists them, for every class at the same time or on different papers.

When should the mock be sat?

When the course is nearly taught and there are still weeks to act: six to ten teaching weeks before the exam is our working rule, since no study we read tests the timing. Two months out, on a short paper, most who failed went on to pass (Chisnall et al., 2015); a week out, revision did not change (Zhang et al., 2023). A tutor's weekly practice papers do the diagnosing beforehand, which leaves the mock free to forecast.

How do you turn a mock mark into a forecast with an honest margin?

Add the gain your past classes made between mock and exam, and give the spread of that gain as the margin. This is our working rule, not a published method.

  1. Use the right boundaries: those of the paper's own series, none from the lenient years.
  2. Measure your history. Set last year's mock marks beside last year's results and take, for each student, the real percentage less the mock percentage. The average is the usual gain; the range holding the middle two thirds of students is the margin.
  3. Forecast a range: the mock mark plus the usual gain, give or take the margin.
  4. Sort the class into three: those who pass with no gain, those who need a gain inside the usual range, and those who need more.
  5. Check on results day, so that next year's gain is measured, not assumed.

With no history yet, the margin is at least the mock's own standard error (test reliability).

One class, with invented figures

Everything here is invented: a physics class of 28, a paper of 70 marks in 105 minutes, a pass at 60%, and a class last year that gained 7 points on average between mock and exam, two students in three gaining between 0 and 14.

Run through EdProcto's analysis code, the 28 sets of marks give "Average score" 53.8% with "spread (sd) 13.9", "Median" 54.3% and "Passed" 35.7%.

Mock mark Students To pass, needs Forecast
60% and over 10 No gain A pass
46 to 59% 10 A gain of up to 14 points In the balance
Under 46% 8 More than the usual range Below without a change

On last year's average gain the 15 students at 53% or over pass. So the head of department hears "about 15 of 28, and between 10 and 20 if every gain stays inside last year's range". A student on 54% hears "61%, give or take 7", which is not a predicted pass.

In the full analysis the code writes "Reteach before moving on: Electricity (40.1%), Forces (53.7%)." Electricity carries 20 of the 70 marks, so the first lesson back is on circuits.

How do you hand a mock back so that it changes revision?

Hand it back by topic, within days, with the next piece of work already set.

  1. Grade everything first. A written answer still to mark counts as zero in every figure.
  2. Ask for a forecast before the marks. Each student writes down the mark they expect, and the gap is the lesson: concrete feedback is what sharpened forecasts in Miller and Geraci (2011).
  3. Give marks by topic beside the total, and the student's group, not a grade.
  4. Set the work. Reteach the weakest topic in the first lesson back and quiz it within the week, in new questions, then again later (spaced revision).

How does EdProcto run a mock exam?

EdProcto is the best tool for a mock that has to forecast, because what a forecast needs is kept on its server (why it is built this way): the questions reach a student only when their attempt begins, the time limit is held there, and no score leaves until you release it.

  1. Build to the blueprint. Write each question in any of the templates, or have Procto AI draft them from your own notes for you to check. Give each its marks in "Points" and its blueprint area in "Topic (for the results by topic)", worded the same every time.
  2. Press Exam, then match the real paper. The preset sets "Time limit" to "Set" and "Show the score" to "When released", and switches on "Submit by itself when time runs out", both shuffles and six browser measures, from "Full screen" to "Leaving the tab": Medium integrity. It switches off "Allow going back to change an answer" and "Allow skipping a question", and sets "Show the correct answer" and "Show the explanation" to "Never". A paper exam lets a candidate skip and return, in a printed order, so to rehearse one switch those two back on and "Shuffle the questions for each student" off. Put "Minutes overall" at the real paper's and "Pass mark" at the boundary you forecast against. With the answer and the explanation at "After submitting", they arrive with the release.
  3. Sit it. Students with extra time sit a copy with a longer limit, made with "Duplicate this quiz" at the foot of the builder.
  4. Read the figures. Every plan shows "Average score", "Median", "Passed", "Score distribution" with a dashed line at the pass mark, and "Got it" for each question. "Passed" is the share at or above the "Pass mark": on the invented class it read 53.6% with the mark at 53 and 71.4% at 46. On the Educator plan the full analysis adds "By topic", weakest first, and "Reliability".
  5. Release, and keep the marks. Press "Release the scores", then "Yes, release them". "Download the marks" saves a row for each attempt: "Name", "Email", "Sitting", "Started (your time)", "Handed in (your time)", "Score %", "Points", "Out of", "Reduced by %", "To grade", "Minutes", "Security events", "Notes on the record" and a column for each question. On results day, put the real result beside "Score %".

Free takes thirty students a quiz, Tutor 100, and Educator 500 with the full analysis; the Exam preset's Medium level is saved from the Tutor plan up. See pricing and the page for teachers. EdProcto gives the figures and never a prediction: no setting makes cheating impossible, a logged event is a record and not proof, and a mock mark is not a grade.

Quick questions

How accurate are mock exam results as a forecast?

Close for a class and loose for one student. Across 740 schools, fewer than half of teachers' predicted grades matched the result in most subjects (Beynon and Thomson, 2025).

When should students sit a mock exam?

When the course is nearly taught, with six to ten teaching weeks left. That is our working rule, not a finding.

Can I use a past paper as the mock?

Yes, on paper and as its board's terms allow, if the class has not seen it. Otherwise write new questions to the same blueprint.

Which EdProcto plan does a mock exam need?

Tutor for a class of up to 100 under the Exam preset, and Educator for the marks by topic. The pricing page has both.

References

The results

Difficulty, discrimination, the wrong option that beat the right one, reliability, and what to reteach, in plain sentences.

Basic on Free, full on Educator

What the results say

Read next