Somewhere around the 40th icon in a set, consistency starts to loosen.
It usually isn’t the shapes. It’s the values that look completely normal one at a time: stroke width, corner radius, cap style, grid alignment, visual weight.
Why eyes aren’t reliable
On any single icon, each of these deviations is within tolerance.
A 1.75 stroke versus a 2, a corner radius of 1 versus 1.5, a path that ends at 12.03 instead of 12. On its own, none of them is a problem. Lined up together, they make the set feel a bit loose, and by then it’s hard to say which icons are responsible.
Besides, by the 40th icon you’ve looked at the same batch hundreds of times.
Let a script measure what can be measured
We now review icons with a script that checks three things:
- Geometry: stroke width, corner radius, caps and grid alignment all have to land on the values listed in the spec
- Source: every shape has to come from the existing icon library, nothing drawn by hand on the spot
- Small sizes: whether the details blur together at the smallest size
Anything that fails doesn’t go into the library.
What’s left for people
A script can’t judge whether the metaphor is right, whether this icon and that one are saying the same thing, or whether you can still recognize it once it’s small.
We still look at those one icon at a time. Stroke widths, we don’t.
