How it works
What counts as an outlier, and why it matters
The most useful thing in a dataset is often the one row that does not fit. Here is the plain-language version of how to find it, and why a flag beats an eyeball.
Omar Haddad, Co-founder and CTO · May 20, 2026 · 5 minute read
Averages hide things. A month where revenue was flat can contain a week that fell off a cliff and a week that made up for it, and the average will tell you none of that. The interesting number is usually the one that does not sit near the others.
A number, not a hunch
You can eyeball an outlier in a small chart. You cannot eyeball one across forty columns, and you will miss the one that matters because it is on a chart you did not draw. So Plotwell computes it. For every numeric column it works out the usual range, the middle half of the values, and flags anything that falls well outside it.
The usual range
Take the value a quarter of the way up your sorted numbers and the value three quarters of the way up. The gap between them is where most of your data lives. A point far past either end is worth a look.
That is a standard statistical test, the interquartile fence, and its virtue is that it is boring. It does not need a threshold you invented, it does not care about the shape of the data, and it flags the row it came from so you can go and look at what happened that day.
Why the flag has to be honest
An outlier flag is only useful if the value behind it is real. So the value, the row and the range all come from your actual data, computed in code. The model is allowed to explain an outlier in a sentence. It is never allowed to decide there is one, or to change the number.
The point of a flag is not to be dramatic. It is to put your eye on the one row you would otherwise have scrolled past.
Omar Haddad, CTO