Past papers and mark schemes

The highest-yield material you have

If you do one thing differently after this course, make it this: past papers, done under exam conditions, marked against the mark scheme, starting far earlier than feels sensible.

Most students save past papers for the final fortnight, as a test of revision that is already complete. That is backwards. Past papers are the best study material available, not the assessment of it, because they tell you what is actually asked, in what form, with what weighting, and where the marks are.

Start in week three. Do one question badly. That failure tells you what to revise, and the information arrives while you can still act on it.

Reading a mark scheme properly

Mark schemes are published for most public examinations and available from many university departments on request. They are more informative than any revision guide, and almost nobody reads them.

What to look for:

• What earns the marks. Frequently a specific term, a specific comparison, or the statement of an assumption, not general quality.
• How many marks for what. A four-mark question wants four things. Students routinely write a paragraph for three marks and one line for six.
• The band descriptors. The difference between the top two bands is usually one identifiable thing: evaluation, engagement with counterargument, precision of terminology. Find out which, for your subject.
• The command words. Describe, explain, evaluate, justify, compare are technical terms with fixed meanings in most examination systems, and answering the wrong one costs the entire question.

Where the model belongs, exactly

Here is the pattern that works, and the order is the whole of it:

1. Do the question yourself, timed, closed book, on paper.
2. Mark it yourself against the mark scheme first. Be harsh.
3. Then paste the question, the mark scheme, and your answer, and ask: which marks did I miss, and quote the sentence where each was lost. Do not write a model answer.
4. Rewrite the parts you lost marks on, yourself.

Step three is legitimate everywhere, produces no submitted text, and is extremely useful. Steps one and two are non-negotiable, because an answer you did not write cannot tell you anything about what you can do.

The request to avoid is write me a model answer to this question. You will read it, it will look right, and you will learn nothing: the fluency trap in its purest form.

Where mark schemes do not exist

Many university modules do not publish them. You are not stuck:

• The learning outcomes in the module handbook are the criteria in another form. Paste those.
• Ask the lecturer what distinguishes a first from a two-one on this paper. Most will tell you, in detail, and are pleased to be asked.
• Old feedback. Comments on your returned work are a personalised mark scheme, and the repeated ones are the ones to act on.
• Ask the model to construct plausible criteria from the learning outcomes, then check them against a lecturer. Better than nothing; not a substitute for the real thing.

Generating questions when you run out

When you have exhausted the real papers, a model will generate more in the same style, and this is a good use: provided you paste real past questions as examples so it matches the form, and provided you check the generated questions against the syllabus. Generated questions drift towards the plausible and occasionally test something the course does not cover.

A useful prompt:

Here are six real exam questions from this paper. Generate eight more in exactly this style, at the same difficulty, covering these syllabus topics. Do not provide answers.

Do not provide answers keeps them usable as practice rather than reading.

The pattern-finding pass

One good analytical use. Paste the last ten years of questions and ask which topics recur, in what combinations, and how the phrasing has changed over time.

This is analysis of supplied text, which is the safe category, and it often reveals structure nobody told you about: a topic that appears every second year, two topics that always appear together, a command word that changed in 2022. Check the output against the papers, because it will occasionally over-claim a pattern from thin evidence, which is worth noticing, since inventing patterns from small samples is a failure you should be alert to in your own thinking too.