Why your odd one out score will not match another site's

You scored 14 here and something quite different somewhere else. Both can be right, and here is exactly where the difference comes from.

Version 1.0 · Published 5 September 2026 · Last reviewed 5 September 2026 · About the odd one out test

1. The rules are not the same rules

Our test uses four kinds of difference: fill, shape, orientation, and a combination where two features vary at once and neither alone identifies the odd tile. That mix is a design decision, and it is published on the test page.

A test built mostly from matrix completion, or from sequence continuation, or from spotting a rotated figure, is asking you to do something related but distinct. Change which rules appear and in what proportion, and the same person's score moves — without anything about that person having changed.

2. The item count changes what a point is worth

We use eighteen grids, so one item is worth about six per cent of the total. A test with ten items makes each one worth ten per cent, and a test with sixty makes each worth under two.

Short tests are noisier, and that cuts both ways. On a ten-item test, two lucky guesses move you a long way. On ours, a blind guess is right about one time in nine, so chance alone produces roughly two correct out of eighteen — which we publish precisely so you can subtract it mentally rather than being sold the whole number as skill.

3. Timing changes the task, not just the difficulty

Our test has no timer at all. That is deliberate: it makes the task about what you notice rather than how fast, and it means nothing about your score depends on your device, your connection or whether the tab was in the background.

A timed test is measuring something genuinely different — noticing under time pressure, which recruits speed as well as discrimination. Neither is wrong. But a timed score and an untimed score are not two attempts at one measurement, and treating them as comparable is the mistake.

We learned this one the hard way on our other instrument, and published the correction: browsers slow timers down in background tabs, so a timed task can quietly become an easier task without anyone being told.

4. Most tests will not tell you their scoring rule

This is the one that makes the other three impossible to reason about. If a site does not publish how it turned your performance into a number — what counted, what was excluded, whether items were weighted, whether retries were allowed — then there is no way to work out why your two scores differ, and no way to tell whether either was computed sensibly.

Ours is on the test page in full, including the exclusions and the parts that are unflattering. That is not generosity; it is the only thing that makes a comparison like this possible at all.

What would have to be true for two scores to be comparable

• The same task, built from the same kinds of rule in the same proportions.
• The same number of items, or a scoring method that accounts for the difference.
• The same timing conditions — both timed the same way, or both untimed.
• Both scoring rules published, so the arithmetic can actually be checked.
• Both norm-referenced against a described sample, if either is going to make a claim about where you sit relative to other people.

Almost no pair of online tests satisfies the first four. Very few satisfy the last one, and we are not among them — our reference sample is 0 people, which is stated on the instrument register along with what would change it.

So which number should you believe?

Neither, as a statement about you. Each is a description of what you did on one specific task under specific conditions. Our number is a count of grids out of eighteen, and that is all it ever claims to be.

The one comparison that survives all of this is you against yourself — the same task, conditions you can name, a note of what was different. That is available to you here for free, and it does not require us to have measured anybody.