"He knows it, he just makes silly mistakes." That sentence is said in millions of homes, in every language, after every exam — and it is almost never true.
We understand why it's said. It's a kind sentence. It protects the child ("he knows it") and it protects the evening ("nothing to fix here, just be more careful next time"). But it quietly costs your child the most valuable thing they produce: information. Wrong answers are not noise. They are the most information-dense thing your child brings home.
Because here is what decades of error research keep finding: children's mistakes are overwhelmingly systematic, not random. A "careless" child is usually making the same mistake, in the same situation, every single time. And a mistake that repeats is not a character flaw. It's a pattern — diagnosable, nameable, and fixable.
Errors have fingerprints
The classic work here comes from Brown and Burton, who analysed thousands of children's subtraction errors and showed that most were produced by "buggy algorithms" — procedures learned slightly wrong and then executed with perfect consistency. VanLehn extended this research and catalogued whole families of stable, repeatable subtraction bugs.
Read that again: executed with perfect consistency. The child everyone calls careless is often the opposite — meticulously, faithfully running a procedure that happens to contain one wrong step. Telling that child to "be more careful" is like telling someone with a wrong recipe to stir harder.
"Careless" is what a pattern looks like before anyone has counted it.
So let's count it. Below are the six patterns that account for most of what gets marked in red, each with what it looks like and what it actually is. As you read, hold a specific child's exam paper in your mind — say a child in Grade 4 (Class 4 in India, Year 5 in England) — and see which ones you recognise.
The six patterns
1. Place-value slips
Looks like: 34 + 28 = 512. Bizarre, almost random — how do you get a three-digit answer from two small numbers?
Actually is: perfectly logical, once you see it. The child added each column blind: 3 + 2 = 5, and 4 + 8 = 12, then wrote both results side by side — 5, then 12, making 512. Nothing careless happened. Place value never consolidated, so the columns are just digits sitting next to each other, not tens and ones.
The fix: rebuild place value with materials the child can hold — bundling, base-ten blocks, expanded form (34 is 30 + 4) — before touching column arithmetic again.
2. Fact-retrieval failures
Looks like: 7 × 8 = 54. Again. The same wrong answer for the same fact, week after week — while harder problems come back correct.
Actually is: a fluency gap wearing a careless costume. The fact was never truly automatised, so the child reconstructs or half-remembers it each time — and a wrong version has started to feel familiar. The reliability of the error is the giveaway: careless answers scatter; retrieval failures repeat.
The fix: short daily retrieval practice on the handful of facts that misfire — not the whole table, just the offenders. This is also the pattern behind many children labelled slow; if that word has been used about yours, read is my child bad at maths, or just bad at speed?
3. Buggy algorithms
Looks like: 52 − 38 = 26. And 63 − 47 = 24. And 81 − 25 = 64. Every subtraction with borrowing comes back wrong, yet the child works diligently and shows every step.
Actually is: the most famous bug in the research literature — smaller-from-larger. In each column the child subtracts the smaller digit from the larger one, regardless of which is on top: for 52 − 38, tens give 5 − 3 = 2 and ones give 8 − 2 = 6, hence 26 instead of 14. The procedure itself was learned wrong, and it will repeat forever until it is re-taught. No amount of "check your work" helps, because when the child checks, they re-run the same buggy procedure and confirm their own answer.
The fix: re-teach the procedure from meaning — regrouping with base-ten materials — rather than repeating the rule louder.
4. Working-memory overload
Looks like: loses the carried 1 — but only in 3-digit problems, never 2-digit ones. So 56 + 27 comes back as a correct 83, while 356 + 287 comes back as 633 instead of 643: the carry from the ones column evaporated somewhere around the hundreds.
Actually is: not attention — capacity. Ashcraft's research on working memory and arithmetic shows that when basic facts aren't automatic, each retrieval consumes working memory, and multi-step problems overflow it — producing errors precisely and only on the longer problems.
That "only on long problems" signature is the tell.
The fix: counterintuitive but well-established — don't drill the long problems. Automatise the underlying facts so each step gets cheaper and memory frees up. This is the same missing-prerequisite story that explains most maths struggling, in miniature.
5. Misreading under pressure
Looks like: the working is flawless — for a different question. Asked "how many more does Priya have than Tom," the child adds the two amounts, beautifully.
Actually is: answering the question they expected rather than the one asked. Part exam craft (no habit of underlining what's actually being asked), and part anxiety — a stressed child rushes to be done rather than to be right. If exams reliably produce a different child than homework does, maths anxiety deserves a look of its own.
The fix: a pre-answer ritual — underline the question word, say what's being asked before computing — plus lowering the stakes until reading feels affordable again.
6. Unit and format slips
Looks like: the right number in the wrong clothes. A correct calculation reported as 4 metres when the question asked for centimetres. Or ½ written as 1.2 — the child converting "one over two" into the only decimal-looking thing those digits suggest.
Actually is: a representation gap. The child can compute but hasn't securely linked the ways a quantity can be written — fraction, decimal, unit, and word forms are separate islands, and ½ = 0.5 is a ferry they've never ridden.
The fix: matching work, not more computation: pair fractions with decimals with pictures with measurements until the representations fuse.
The parent protocol: count the pattern
Here's the whole method, and it takes one evening. Collect ten recent wrong answers — from exams, homework, anywhere. Sort them into the six buckets above. Then watch what happens to the word "careless."
Almost always, it dissolves. Ten scattered "silly mistakes" turn out to be one or two named patterns — six place-value slips and three misreads, say — and a named pattern is a to-do list, not a verdict.
One mother we spoke to was certain her daughter "knows it but rushes." She sat down anyway and sorted twelve red-marked answers from two exams. Nine of the twelve were the same smaller-from-larger subtraction bug. One evening of re-teaching borrowing — with base-ten blocks, at age 10, and who cares — removed three-quarters of the "carelessness" before the next test. Her daughter's comment afterwards: "Why did no one tell me I was only making one mistake?"
Why "be more careful" is the least effective sentence in maths parenting
Look back at the six patterns and notice something: carefulness fixes none of them. The place-value slip is careful. The buggy algorithm is extremely careful — that's why it's so consistent. The working-memory overflow happens to careful children mid-carry. "Be more careful" addresses a seventh pattern, random inattention, which turns out to be the rarest of the lot.
Worse, the sentence teaches the child that their errors are moral rather than mechanical — that they fail because of who they are, not because of one wrong step that nobody has found yet. Said often enough, that belief hardens into "I'm just bad at maths," and now there are two problems instead of one.
From one evening to every session
The protocol above is powerful, and you should genuinely run it once. Its only limitation is dosage: you'll sort ten answers on one evening, and the patterns drift as your child learns.
Pattern-spotting across hundreds of answers, every day, is precisely what CREST Champs' engine was built to do. Our free daily adaptive practice reads every wrong answer the way you just learned to — is this retrieval, procedure, overload, representation? — then quietly routes the next questions at the actual cause, and tells you in plain skill terms what it found. Think of this article as the manual version. The automatic one takes about ten minutes a day.