Why do I keep making silly mistakes in maths exams?

Because they are not silly and not random. A 'silly mistake' is a habit — a specific place where your working skips a step, your handwriting is ambiguous, or your attention drops. Everyone has four or five recurring shapes, such as sign errors when expanding brackets or dropping units in the final line, and they repeat because nobody ever wrote them down and looked at the pattern.

How do I stop making careless mistakes in exams?

Stop treating it as a willpower problem. Log every careless mark you lose for two weeks with a one-line description, then count the categories. You will find a handful of repeating shapes. Build a short, specific checking routine that targets those exact shapes — for example 'check every sign after expanding' — because a targeted five-second check works where a general instruction to 'be careful' does not.

Do silly mistakes matter if I understand the content?

They are usually the difference between two grade bands at the top end. Once content gaps are closed, nearly all remaining lost marks are execution errors, so an hour spent on how you write working returns more marks than an hour of extra revision. Method marks limit the damage but they do not remove it — a sign error in line two costs every accuracy mark after it.

What is the best way to check your work in a maths exam?

Do not re-read your working — you will read what you meant to write. Instead check by a different route: substitute your answer back into the original equation, sanity-check magnitude and units, and re-read the question's final line to confirm you answered what was asked. Reserve the last ten minutes for this rather than for one more attempt at a question you have already abandoned.

How to stop making silly mistakes in maths exams

The most expensive marks you lose

Ask a strong student where their marks went and you get some version of: "I knew all of it, I just made stupid errors." They mean it as an excuse. It is actually a diagnosis, and a good one — it says the content is in place and the losses are all in execution.

That is worth taking seriously, because execution errors are cheap to fix and content gaps are not. Learning a topic you have never understood takes hours. Stopping yourself dropping a negative when you expand brackets takes a fortnight of noticing. At the top of the grade range, almost every remaining mark is in the second category.

What it costs. In a method-marks subject, a sign error in line two of a six-mark question typically keeps the method marks and loses every accuracy mark after it. Four of those across a paper is a grade band.

Why "be more careful" cannot work

Because it is not an instruction. It has no trigger, no object and no test — there is no moment in the exam where you can tell whether you are currently complying with it. Compare: "after every expansion, check the sign of each term". That one has a moment attached, takes three seconds, and you can tell whether you did it.

The second reason it fails: you cannot proofread your own working by reading it. Reading your own line of algebra shows you what you intended to write. This is the same reason you cannot spot typos in your own essay — the brain reconstructs from intention, not from marks on the page. Any checking method that amounts to "look at it again" is close to worthless. Effective checks always go by a different route than the original working.

The six shapes silly mistakes come in

Across maths, physics and chemistry papers, near enough all careless losses fall into six categories. Almost nobody makes all six — most people have two or three that account for the overwhelming majority of theirs.

Shape	Looks like	The check that catches it
Sign / sequence	Lost minus expanding brackets, flipped inequality when multiplying by a negative	Re-derive the one line, don't re-read it; substitute a number in
Transcription	Copying 3.47 as 3.74 from line to line, or from calculator to page	Check every value against the question, not the previous line
Units & form	cm vs m, no units on the final answer, wrong significant figures, exact form asked	Re-read the final line of the question, every question, at the end
Misread instruction	"Hence" ignored, "correct to 2 d.p." ignored, answered part (b) for part (c)	Underline command verbs and constraints before starting
Premature rounding	Rounding mid-calculation, answer off in the third digit	Keep full precision in the calculator; round once, at the end
Unfinished	Correct working, never actually stated the conclusion the marks were for	Every answer ends with a sentence or a boxed value

Notice that only one of these is arithmetic. "Silly mistake" gets used as if it means can't-do-sums; in practice most of the losses are reading and finishing problems, which is why drilling times tables does nothing for them.

Find yours: the two-week count

This is the whole method, and it is boring in a way that makes people skip it.

For two weeks, every time you lose a mark to something that was not a content gap, write one line: what you did, in which topic. Not "silly mistake" — that is the word that hides the information. Write "dropped the negative expanding (x−3)(x+5)" or "gave answer in cm, question wanted m".

At the end of a fortnight you will have somewhere between fifteen and forty lines. Sort them by shape and count. What comes out is nearly always lopsided: one shape accounts for a third or more of everything, and two or three cover most of the rest. That top shape is now a named opponent instead of a vague character flaw, and you can build a check for it specifically.

What a real count looks like

14 × sign errors expanding or rearranging · 6 × units missing on final line · 5 × misread "exact value" · 3 × transcription from calculator · 2 × premature rounding

→ Two targeted checks (signs after every rearrange, units on every final line) address 20 of 30.

Building the checking routine

A checking routine should be short, specific to your count, and attached to a moment. Three or four items, no more — a long checklist is a checklist you abandon at minute 90 when you are behind on time.

Per-question, before starting: underline the command verb and any constraint (exact form, 2 d.p., units, "hence"). Ten seconds.
Per-question, at the end: re-read only the final sentence of the question and confirm your answer is that thing, in those units. This one check kills two whole shapes.
Your top shape, at its trigger: if signs are your shape, every rearrange gets a sign scan. If it is transcription, every copied number gets checked against the question paper.
Last ten minutes: substitute answers back where possible, and sanity check magnitudes. A speed of 4×10⁸ m s⁻¹ is wrong and you know it without any working.

Practise the routine during practice papers, not for the first time in the exam. It costs maybe four minutes across a paper, which is why the last of the four items has to come out of time you were previously spending on a question you had already lost.

Why the log is the whole thing

Every part of this depends on one unglamorous act: writing down what went wrong, in enough detail to be countable, every single time. Not the score. Not "revise trig again". The specific move you made.

That record is an error log, and slips are exactly where it earns its keep — a content gap you at least remember, whereas a careless error is embarrassing, feels like nothing, and is forgotten within an hour. Unlogged, it happens again in three weeks and feels like the first time. The repetition is not bad luck; it is what happens to any pattern nobody is tracking.

The reason most people do not keep one is the typing, at night, after a paper. That is the specific job errorboard does: photograph your working, and it reads the page, identifies what went wrong, sorts it into a category, and files it by topic without you writing anything. Then it shows you the counts — so the fortnight of counting above happens automatically, and your top shape is just sitting there on the progress page waiting to be checked for.

The one-paragraph version

Careless errors are habits with shapes. Yours are a short list, not an infinite fog of carelessness. Log every one for two weeks, count them, find your top two shapes, and build a three-item check that targets those specifically at a defined moment in every question. That is a fortnight of mild admin for, typically, several marks a paper — the cheapest marks available to anyone who already knows the content.