Strategy consultancy
A site that is almost entirely typography, which means nowhere to hide: every measure, rule and footnote had to land exactly where the studio drew it.
The studio's design was brave in the way that finds a build partner out: no imagery to speak of, two typefaces, and long essays set to a reading measure with sidenotes. When a site is only typography, every rounding error is on show. This is the project we point to when someone asks why we go on about tolerances.
Where it started
The studio warned us on the first call that two build partners had already turned the design down as too fussy for the money. We asked for the file before answering. What read as fussy turned out to be consistency: one spacing scale applied without exception and two typefaces doing everything, which is the easiest kind of design to build and the hardest to fake. We quoted in two days, fixed, and took the job partly to make the point.
The build ran type-first. Before a single block existed we set the ramp, the measure and the baseline grid in a bare template and had the partners proof real essay text in a browser, at their own text sizes, on the train home. Two weights got swapped as a result of that week, which cost an afternoon then and would have cost the schedule in week four.
The build
The type ramp arrived as tokens and left as CSS custom properties, with nobody's eye in between. Footnotes are entries-within-entries rather than markup pasted into a rich text field, so they renumber themselves when a partner reorders an argument, which, being strategy consultants, they do.
The case-study system is the part the client uses most. Each engagement is written to a fixed structure (the situation, the move, what changed) with the numbers held in fields rather than prose. The template refuses to publish a case study with an empty outcome. That sounds strict, and it has already improved the writing.
Eleven essays came across from the old site by hand, re-set into the new measure with their footnotes rebuilt as entries. We could have scripted the migration and chose not to. Eleven is below the point where a script pays for itself, and re-setting each one was the fastest way to find the edge cases, which turned out to be one essay with nested footnotes and one with a table nobody remembered.
Tolerances, measured
The overlay pass is where this build justified its fee. The first pass across the nine blocks logged 27 deviations of a pixel or more, mostly line-height rounding, plus one wrong optical margin on the pull quotes. The second logged three, the third none, and the studio's design director signed the staging link off in a morning.
The sidenotes sit on a CSS grid against the paragraph that owns them, and collapse to tappable end-notes on narrow screens rather than pretending a phone has a margin. The reading measure is clamped between 45 and 75 characters at every size, because the essays are the product and a 130-character line would undo the whole design.
Motion, mostly declined
The motion pass on this build was mostly an exercise in deletion. What survives is an underline that draws in on link hover and a soft highlight on a footnote when its reference is tapped, each under a fifth of a second, and both absent under reduced motion. Everything else was tried and cut. On a page that is only text, movement is the loudest thing in the room.
Under the typography the Craft work is small and exact: nine entry types, a set of Twig macros that keep the footnote markup in one place, and the type ramp arriving as design tokens and leaving as custom properties. The semantic pass took the accessibility work beyond the visual. Sidenotes are real list items linked to their references, so a screen reader offers them at the reference point rather than as a surprise at the end, and the heading order survived the designer's size choices, because the two are allowed to disagree.
The first build partner we tried sent back our spacing scale rounded to the nearest five pixels. These people found two errors in it that were ours.
After launch
The partners publish their own thinking without the studio in the loop, and the studio's design has survived two years of them doing it, because the system only offers choices that were designed. That was the brief, even if nobody phrased it that way. Forty-one case studies have gone through the fixed structure since launch, each with a filled-in outcome, because the template still refuses anything less.