Teachomatic what to automate, and what not to

What You Learn From Marking

Ask a teacher what they got from an evening's marking and they will tell you about the marks. Ask them what they will do differently on Monday and you get something else entirely — a picture of the class that was assembled while they worked and that appears nowhere in the grades.

That picture is the actual product of marking. It is also the thing that vanishes silently when the marking is delegated, because nothing in the output records that it is missing.

For a workplace comparison of the underlying concept, Monitask’s discussion of chronemics shows how the same behaviour is framed outside education.

What the picture contains

Five things, none of which is in a mark scheme.

Which explanation failed. Not that eleven students got question four wrong, but that they all made the same wrong move, which traces back to a specific thing you said on Tuesday that was ambiguous. You recognise it because you remember saying it.

For an external perspective on teaching, assessment, and feedback, see EdSurge.

Which errors are interesting. Some wrong answers are more sophisticated than the right ones — a student applied a real principle in the wrong place. That is a completely different situation from not knowing, and it is invisible to anything scoring correctness.

Who changed. A student who was scraping through and has started thinking. A student who has quietly stopped. Both matter and neither shows up as a big score movement, because change begins small.

What the class can now do that it could not. The reason to move on, or not, and it is felt rather than calculated.

Where the question itself failed. Sometimes a whole class does badly because the question was ambiguous. You can usually tell from the pattern of misreadings, and only if you read the misreadings.

Why a summary cannot replace it

The obvious objection is that a tool can report common errors, and it can.

The problem is that a summary answers questions you thought to ask. The five things above mostly arrive unbidden. You were not looking for the student who quietly stopped; you noticed because their handwriting was different, or the work was shorter, or they left the last question blank when they never do.

That kind of noticing is what attention produces as a by-product, and it cannot be requested in advance. A report of common errors is a genuinely useful artefact and it is a summary of the answers, not of the class.

There is a second problem. Your picture of the class is cumulative. It builds across pieces of work over a term, and each reading updates it. Delegate three assignments in a row and the picture does not degrade sharply — it just stops being updated, and you carry on with a model of your class that is two months old and feels current.

What this does not argue for

Not that all marking should be done by hand. That position ignores the actual conditions of the job, and refusing help at ninety scripts produces worse feedback, not better, because fatigue degrades what you write, unevenly and in a predictable order.

The argument is narrower: the diagnosis is separable from the scoring, and it is the part worth protecting. You can automate scoring extensively and keep the diagnosis, as long as you do it deliberately.

How to keep it

Read a sample properly, always. Fifteen scripts, before any marks are awarded. This is the whole defence and it costs half an hour.

Read for the unexpected, not for errors. Different intention, different result. You are looking for anything that surprises you.

Write five lines the same day. What they did not get, what surprised you, what you will reteach. The picture fades by Tuesday, and five lines preserve enough of it to act on. Keeping your own record deliberately is worth more than any report a tool produces.

Sample across the range, not the top. The middle is where movement happens and where it is least visible.

Vary who you read. If you always read the first fifteen, you always see the same students.

The reason this is hard to argue for

Because nothing goes wrong.

If you delegate all of it, the scripts get marked, the grades are defensible, the students get feedback that reads well. There is no incident. What happens instead is that over a term your lessons become slightly less well-aimed, you reteach the wrong thing, you miss the student who stopped, and you find out in an exam.

No single decision caused it and no alarm sounded. That is the shape of this failure everywhere on this site, and it is why the countermeasure has to be a habit rather than a judgement — a step you perform every time, whose absence you would notice.

Half an hour of unassisted reading per assessment. That is the price of continuing to know what your class understands.

The short version