EIKIRA

Studio Eikira was a boutique brand and marketing agency with one name, one mark and a flat structure. Eikira is a software company that also does design — a parent with two arms, a product line, and a naming rule that puts the parent first. I ran the rebrand: architecture, marks, color, type, voice, values, the guidelines, and the site that replaced the old one.

The visual language survived almost entirely intact. The architecture around it changed completely. What follows is the before, the reasoning, and the after — with the live sites, not screenshots of them.

The old site, running. It is not on the web any more — studioeikira.com now redirects to eikira.com — so this is a frozen snapshot of the last build before the rebrand, served from this repository and sealed in its own frame.

One approach ran through the whole rebrand: write a rule, then audit it against itself, and fix it where it breaks on its own page. Most of what changed between V1 and V2 was not found by rereading the document. It was found by building something the document was supposed to describe, and watching the two disagree.

A guideline is a hypothesis until something is built against it. Three of those disagreements are worth showing, because in each one the thing that failed was not the design. Twice it was the measurement. Once it was a rule that was correct and still could not be used.

Choosing a mark for very small sizes needs a number, so I wrote one: the largest connected shape as a share of lit pixels at 16px. A low score means the mark has fragmented into specks. It ranked the candidates cleanly and retired the full seal from small sizes on the evidence.

Then it passed two things it should not have. The full-body cat scored 25% and was visually mush. The micro mark scored a perfect 100% — not because it was excellent, but because a solid silhouette is physically incapable of fragmenting. The first micro-mark draft also scored 100% and was a dark lump with two slits.

Fragmentation is one failure mode. Blobbing is the other, and the metric had nothing to say about it in either direction. The rule that went into the guidelines keeps the number and refuses to trust it alone.

The casing rule told you which strings take caps and which take title case. Its scope note read: this governs type that is set. That sentence is accurate. It is also unusable, because it asks you to decide whether something is set type — a term of art — before you can apply the rule at all.

It broke on a real string. The PDF cover’s signature line sits in a display slot, is arguably set type, and is unmistakably a sentence — and by the time anyone looked, the product’s own name was being set three different ways across four places in the repo, each one defensible from the rule as written. The replacement asks something anyone can settle without the vocabulary: is this a sentence, or is it a label?

A second audit, after the rule was rewritten, turned up the opposite failure. METIS, set in caps across four files — including the site title, which is the browser tab on every page and the name on every shared link. Metis is a product, and products are title case everywhere — in sentences and as labels alike — because case is the tier signal: a product set in caps would read as an arm. Nothing here was ambiguous. This was one error copied four times, rather than four readings of a hard rule.

Both belong to the same rule, in opposite directions. Where it left room, it produced a spread and every answer looked careful. Where it left none, nobody had read it.

V1's wordmark set STUDIO in caps over Eikira in title case. The casing was a decision — caps on the brand name so it read as the primary word, and because it simply looked better that way. What was not a decision was the size gap the casing produced. Nobody specified one; Cinzel Decorative renders uppercase noticeably larger than title case, and the lockup ended up accommodating that.

Both strings below are set at the same font size. The gap between them is the typeface's, not the designer's — so the most conspicuous feature of the wordmark was discovered rather than drawn, and could only be worked around. V2 keeps the intent and drops the mechanism: plain Cinzel, with that contrast rebuilt as scale someone actually sets.

Underneath the three threads, the structural change: a flat brand became a three-tier family. EIKIRA is the parent. EIKIRA STUDIO and EIKIRA TOOLS are sibling arms. Metis is the first product, with an AI persona inside it. The parent name always comes first, which is why “Studio Eikira” was retired as a written name in favor of “Eikira Studio” — both arms then share one pattern.

That produced the mark rule that replaced one mark doing every job: one mark type per tier. Parent takes the full-body cat. Arms take the head only. Products take the enso ring with a geometric object inside it. No overlap, so the mark alone tells you what level of the business you are looking at.

Casing carries the tier too — parent and arms in caps, products in title case — so every future product inherits the rule without a new decision being made.

The live site, embedded the same way. It is the real eikira.com, running now, in a sealed frame at both a desktop and a mobile viewport.

The V2 guidelines didn’t stop at a PDF. The brand has to hold when the person making the slides has never met me, so I built the guidelines into templates. A few rules are enforced by the layout itself: a fixed title box clips a long title, and a twelve-column grid has no room for a fourth column. Most are notes written into the placeholder copy, read at the moment they apply. Then I tested it by feeding it things the rules forbid.

Not every rule can be enforced by a layout. The demo shows the ones the template refuses and the one it can’t.

A repositioning this size did not require starting over. The cat, the seigaiha pattern and its overlay rule, EB Garamond and Inter in their existing roles, the origin story, and one voice principle carried over verbatim. The palette lost two colors mid-V1 and gained one promoted from working use.

What changed was the architecture around all of it — and the discovery, three times over, that the thing most likely to be wrong in a brand system is not the design but the number someone wrote down to justify it.