Sequence Memory Percentiles
See what level reaches the 25th, 50th, 75th and 95th percentile on a sequence memory test, and why all-or-nothing scoring widens the distribution.
Level 8 sits at roughly the 50th percentile. Level 10 is near the 70th, level 13 near the 85th, and level 19 or above is the top few percent.
The distribution is wider than most memory tests because a single mistake ends the run outright.
Level by percentile
Percentiles describe MeasureHuman's modelled distribution and will be revised as real results accumulate. Use them to locate yourself, not as a precise rank.
| Band | Score | What it usually means |
|---|---|---|
| 5th percentile | Level 4 | Below most first runs |
| 25th percentile | Level 6 | Lower half, usually tile-by-tile recall |
| 50th percentile | Level 8 | The median result |
| 70th percentile | Level 10 | Clearly above average |
| 85th percentile | Level 13 | Consistent path-based encoding |
| 97th percentile | Level 19 | Top few percent of all runs |
Why this distribution is wider than digit span
Number memory produces a tight distribution: almost everyone lands between 4 and 15 digits. Sequence memory spreads much further, and the reason is the scoring rule rather than the underlying ability. One wrong tile ends the run, so a moment of inattention at level 14 produces the same number as genuinely failing at level 5.
That makes any single run a lower bound rather than a measurement. The percentile that describes you is the one attached to your best of several attempts, not to whichever run you happened to take first. This is a real difference from tests that average across attempts, and it is worth knowing before you read too much into a low number.
It also means the top of the distribution is thinner than it looks. Reaching level 19 once is achievable for a lot of people with enough attempts; reaching it reliably is what the 97th percentile actually describes.
What all-or-nothing scoring does to the distribution
Most benchmarks degrade gracefully. A slow reaction time is still a reaction time; a low WPM is still a measurement. Sequence memory does not work that way — one wrong tile ends the run, so the score records where you stopped rather than how well you were doing.
That produces a distribution with a long tail downward that has little to do with ability. Someone capable of level fourteen who loses attention at level five records a five. Nothing in the number distinguishes that from someone who genuinely could not go further.
The consequence for reading your own percentile is straightforward: use your best result, and take enough attempts for your best to mean something. Three runs is usually enough to get within a level or two of what you can actually do. One run frequently is not.
Why other sites quote different numbers
Sequence memory tests vary more than they appear to. Some use a four-button layout rather than a nine-tile grid, which changes the difficulty substantially — fewer options means fewer ways to go wrong. Some regenerate the sequence each level instead of extending it, which removes the accumulating memory load entirely. Some allow a life or two before ending the run.
Each of those choices moves the average by several levels, so a figure quoted on one site is rarely comparable with a figure from another. A level twelve on a four-button test that regenerates each round is a different achievement from a level twelve here.
When comparing, check three things: how many tiles are in play, whether the sequence extends or regenerates, and whether a mistake ends the run. If those match, the numbers are broadly comparable. If they do not, they are not.
The same caution applies to the audience behind any published average. Everyone quoting a figure is describing people who chose to seek out a memory test, which is not a representative sample of anyone. Sites with a competitive following report higher averages than sites with a casual one, using identical mechanics, purely because of who shows up. A percentile is a reliable way to locate yourself within one population; it is not a measurement of the general public, here or anywhere else.
Frequently Asked Questions
How Averages Are Estimated
Percentiles are modelled from MeasureHuman's sequence memory distribution, calibrated so the median sits at the typical adult result. They assume a best-of-several score rather than a first attempt.
Measurement Limitations
The distribution has a long lower tail produced by attention lapses rather than ability, because a single wrong tile ends a run. Percentiles also describe self-selected test takers rather than a representative adult sample.