Original data · recomputed every build

What 22 award-tier websites are actually made of.

Across 22 published sites cut into 83 catalogued pieces, the median site is made of 4 parts — not the dozens people assume. The largest runs to 8; the smallest is a single unbroken composition. 64% carry at least one reusable effect. Everything below is measured from our own build registry, not surveyed or estimated.

A site is fewer parts than you think

Pieces are cut at scroll boundaries and never mid-mechanic — a pinned sequence is one part however long it runs. That is why the counts are lower than a component inventory would suggest: this measures the journey, not the DOM.

Median
4

parts per site

Mean
3.8

parts per site

Most
8

the biggest build

Fewest
0

one composition

What those parts are

section26
component25
hero17
effect15

Most common shapes

cursor8
gallery8
table6
about6
stats4
card4
configurator4
text-effect4

Themes, by how often they appear

dark12
editorial11
minimal10
luxury6
tech5
surreal5
brutalist4
gaming3
automotive2
sports2

Themes that travel together

Which pairs actually co-occur on the same build. Useful mostly as a warning: the common pairs are common because they are safe, and the rare ones are where a site stops looking like the others.

  • editorial + minimal6 sites
  • luxury + minimal5 sites
  • dark + editorial4 sites
  • dark + tech4 sites
  • editorial + luxury4 sites
  • dark + minimal3 sites

Method, so you can argue with it

The corpus is every site we have published and can show a stranger — hidden, archived and client-owned builds are excluded, which is why the number here matches the catalog exactly rather than the number we have built. Pieces are counted only from reachable sites: the parts of a hidden site are hidden with it. Builds span 2026-07-15 to 2026-07-29. The page is regenerated on every deploy, so it cannot go stale.