Why 8 out of 10 is the right score, and 10 out of 10 is a warning

Every parent has watched a child sail through a worksheet and felt pleased. That feeling is the problem. A perfect score is information — and what it tells you is that nothing was learned.

Published 19 June 20266 min read4 published sources, caveats printedFor parents of 6–11s
The short version
1Learning is fastest just below always getting it right. One line of research puts the sweet spot near 85% correct; we hold children at about 80%.
210 out of 10 means the questions were too easy. Pleasant, and a wasted session.
35 out of 10 doesn’t read as “hard” to a 7-year-old. It reads as “I’m bad at this”, and that is much harder to undo.
4The 2 wrong answers are not a tax on learning. They are the learning — provided the child is not being graded on them.
01

There is an actual sweet spot

In 2019 a group of researchers asked a precise question: if a learner can control how hard the task is, what difficulty produces the fastest improvement? For a broad class of learning systems the answer came out at an error rate of about 15% — roughly 85% correct.

The result is elegant and it needs its caveat stated loudly: it was derived for machine-learning models and simple perceptual tasks, not for children doing logic puzzles. What it gives us is a direction, not a law. We hold 80% rather than 85% because children are not perceptrons and a little more slack is kinder.

The same 10 questions, three difficulty settings
10 right — too easynothing learned
8 right — about rightthe target
5 right — too hardchild gives up
Wilson et al., Nature Communications, 2019 — the “eighty-five percent rule”. Caveat: derived from machines and simple tasks; we treat 80% as a target to steer by, not a scientific constant.
02

Why easy feels productive and isn’t

There is a longer-standing idea behind this, from memory research: difficulties that feel unhelpful in the moment often produce better long-term learning. Re-reading feels effective and mostly is not. Being tested feels harder and works better.

That mismatch is why parents and children both reliably prefer the wrong thing. A session where everything went smoothly feels like progress. A session with three stuck moments feels like a bad day — and is usually the one that changed something.

Feels good
Fluent, smooth, quick

Recognition without retrieval. The child recognises the answer rather than producing it, and forgets it by Thursday.

Feels worse
Slow, stuck, effortful

Retrieval with friction. Uncomfortable in the moment, and the version that is still there next month.

The rule
Aim for mostly-right

Enough success to stay willing, enough failure to have something to think about. That is the whole design brief.

A perfect score is information, and what it tells you is that nothing was learned.
03

How to hold it without a spreadsheet

1
If they are getting everything right, make it harder — quietly

Do not announce it. Add one constraint, remove one clue. Children who know they have been “moved up” start performing rather than thinking.

2
If they are getting most of it wrong, remove a piece

Not a hint — a piece. Fewer variables, same puzzle. Difficulty should come from the thinking, not from the load.

3
Never let the 2 wrong ones cost them anything

No score, no comparison, no disappointed face. The moment mistakes have a price, children start choosing easy questions on purpose.

Questions parents actually search for

What is the ideal success rate for learning?

Research on optimal difficulty points to roughly 85% correct for fast learning, and practical systems for children often aim slightly lower, around 80%. The key idea is that both near-perfect and near-failure scores waste the session.

Should my child be getting things wrong?

Yes — about 2 in 10 is a healthy rate. Those are the questions that require actual thinking rather than recall, and they are where progress comes from, as long as being wrong carries no penalty.

What is the 85% rule?

A 2019 finding that, for a broad class of learning systems, improvement is fastest when about 15% of responses are errors. It was derived mathematically and tested on machines and simple perceptual tasks, so treat it as a direction for humans rather than a precise target.

Isn’t it demoralising to get questions wrong?

It depends entirely on the framing. Wrong answers that are scored, compared or sighed at are demoralising. Wrong answers that are simply interesting — “huh, why didn’t that work?” — are the opposite.

Everything this piece leans on
Wilson, Shenhav, Straccia & Cohen, Nature Communications, 2019
Derived an optimal training error rate of about 15.87% — the “eighty-five percent rule” — for a broad class of learning problems.
Caveat: mathematical derivation plus machine and simple perceptual tasks. Not a study of children doing logic puzzles.
Bjork & Bjork, on desirable difficulties
A long line of memory research showing that conditions which feel harder during study often produce better retention and transfer.
Caveat: mostly laboratory work with older learners; the principle generalises better than the specific numbers.
Roediger & Karpicke, Psychological Science, 2006
Retrieval practice beat repeated study for long-term retention, while learners predicted the reverse.
Caveat: verbal material, university students.
Vygotsky, on the zone of proximal development
The original framing of the idea that learning happens just beyond what a child can already do alone.
Caveat: a theoretical framework from the 1930s, valuable as a lens rather than as measured evidence.

One finding a week, in plain words, with its caveat attached

That is the whole newsletter. Two minutes on a Friday, nothing sold, unsubscribe in one click.

Get the weekly noteIt comes with a parent account — the only place we ask for an email, and it is covered by the privacy notice.
Keep reading