Running online exams in a school: the decisions no feature list mentions

Mid-term exams are a few weeks out and the school has decided to run them online this time. The first question in the meeting is almost always the same: which software.

That is the easiest question to answer and the least decisive. What decides how the week goes is a set of small choices that appear on no feature list, and nearly all of them only show their consequences once the exam is underway.

This piece walks through those choices in the order an online exam actually happens — before, on the day, and after. Most of them apply whichever software you end up choosing.

Before: who is sitting it

An exam is set up weeks ahead. The class does not stop changing in the meantime — students move between classes, a new one arrives, another withdraws.

So there are two ways to name who sits it, and they look identical until somebody moves: write down the list of names, or name the class. A list of names is frozen on the day it was made. Naming the class means its contents get asked for again every time they are needed — when the notice goes out, when the exam opens, when you check who has not sat it yet.

For a single subject's test, that can be the whole class or a named subset. For a school-wide paper spanning several classes, it can be a combination: named students, one class, a whole year group and everything beneath it, or a filter such as grade level and enrolment year.

Worth checking: if a student joins that class next week, are they automatically included? If not, somebody has to remember to add them, and on exam morning nobody remembers.

Before: which questions a machine can mark

This is the most misunderstood part and the most expensive to get wrong, because expectations about "automatic marking" get set before anyone learns where it stops.

There are three kinds of question and only two can be marked without a person.

Multiple choice is marked instantly. Nothing to discuss.

Short answer is marked by matching against a key, and this is where the invisible problem lives: a student writes "Ir. Soekarno", the key says "Soekarno", and the machine calls it wrong. So one short-answer question can hold several keys that are all accepted, and the comparison ignores what is not knowledge — capitalisation, extra spaces, a full stop or question mark at the end.

That is still not enough, so short answers can be handled three ways, and the middle one suits most schools: marked automatically, but every answer it calls wrong is flagged for a human. The right ones pass straight through; the wrong ones wait for you. An answer a machine marks wrong may be a spelling variant rather than a gap in knowledge, and the only thing that can tell those apart is the teacher who taught the chapter.

Written answers are never machine-marked. They always wait for a person.

Which produces one thing worth knowing before you write the paper: while any written answer is still unmarked, that student has no score at all. Not zero, not provisional — none. A partial score is not a final score, and showing it as one makes half the class look like failures on the afternoon of the exam.

Worth deciding: how many written questions you can actually mark inside the deadline you promised parents. That is a real constraint, and it should shape the paper rather than the other way round.

On the day: two kinds of time that get confused

They sound like the same thing and are not.

Working time is how long a student gets once they start. Ninety minutes, counted from their first click.

The exam window is when the paper may be started at all. Monday 07:30 to 09:30, say.

Both run at once and whichever arrives first wins. A student who opens a ninety-minute paper at 09:29 does not get until 10:59 — the exam still closes at 09:30. Otherwise opening the paper at the last minute becomes a way to buy an extra hour, and somebody will find it.

Worth checking: whether your window is wide enough for a student who arrives late or whose device fails, without being so wide that invigilation stops meaning anything.

On the day: cheating, and what can genuinely be prevented

This section has to open with a sentence software vendors rarely write: no application can stop a student opening another tab, using a second device, or asking the person in the next room. Anyone who says otherwise is selling something.

What is achievable is making cheating more trouble than doing the work, and that is mostly a matter of settings rather than surveillance.

Shuffling the question order makes "number 7 is B" worthless.

Shuffling the answer options makes "the answer is B" worthless too. This is a separate setting from the one above despite the similar name: shuffling the order of questions does not shuffle the options inside them.

But the decisive one is neither. It is a setting that rarely gets attention: when the answer key becomes visible to students.

If the key appears the moment a paper is handed in, and your class sits the exam in two sessions — or even if just one student finishes early and walks out — then the exam stops measuring anything from the moment the first person submits. Not because the system leaked, but because you handed out the key yourself.

Worth deciding: if the exam is sat simultaneously under invigilation, showing the key on submission is good for learning. If it is not simultaneous, do not.

After: how many attempts, and which one counts

If a paper can be retaken, there are two separate questions: how many times, and which attempt becomes the mark.

The mark can be the highest, the last, the first, or the mean of all of them. None is universally right — for practice, highest makes sense; for a real exam, first or last makes more.

This is a school's assessment policy, not a technical setting. Unfortunately it lives on a settings screen, so it often gets filled in by whoever happened to build the paper.

After: a blank mark is not a zero

Once the exam is over the marks table fills with blank cells, and they do not mean the same thing. Some students did not hand in. Some handed in and their written answers are waiting to be marked. Some had a paper that was not due yet. Some were never given it at all.

Only one of the four deserves to be treated as a zero, and only then if your school decides so. The difference is covered separately in Four reasons a mark is blank, because it determines the number that ends up on the report.

After: checking the paper, not only the students

Results get read as a report on the students. Half of it is a report on the questions you wrote: items nobody reached because the paper ran long, items almost everyone got wrong because the key was set to the wrong option, items everyone got right that therefore told you nothing.

All three are only recognisable while you still remember why each question was written that way, and that memory does not survive until the next paper is built.

The checklist

Before your school's first online exam, make sure somebody has answered these eight things — in whichever software you use:

1. Whether candidates are named as a class or as a list, and who is responsible for updating it when a student moves.
2. How many written questions there are, and who marks them inside the promised deadline.
3. The working time and the exam window, and what happens when they collide.
4. Whether question order is shuffled, option order is shuffled, or both.
5. When students can see the answer key — and whether everyone sits the paper at the same time.
6. How many retakes are allowed and which attempt counts.
7. Whether blank cells count as zero, and which blank cells.
8. Who may see a whole class's marks, and who may only see the subject.

None of those eight needs software to answer. But answered before the day, the software almost never turns out to be the problem — and left unanswered, no application rescues it.