The writing and agent indexes were mostly plain lists, and the open source page was still a long stack of prose. Rather than settle on the first card layout, we made a lab with ten versions using the actual page content, then expanded it to thirty. It gave us somewhere to try the display type, depth effects, HUD pieces, metrics and project cards together, and to see where photos or screenshots added something useful.
The first pass was all inside the normal content measure. That's useful for comparing lists, but it made every hero feel like another box on the page. We added a full-width collection, then let some backgrounds reach right up to the top of the viewport behind the primary navigation. Section now has an explicit option to remove its gutters for those bands, while the text inside can still follow the site's content measure. Large photo backgrounds will need that on the landing page too.