The translation desk at Halvorsen & Wren, a literary press: chapter seven of a novel out of Spanish, the source and the English side by side, and every open decision in the margin beside the line it is about.
claude-brand is the third house style in this portfolio, after calm-practice and apex-brand, and the one that needed the least inventing. The other two had no vendor and no published values, so their generators solve a palette from a declared ground up. This style publishes everything: a twenty-one step ramp, thirteen named colours, fifteen type steps, a leading ladder, ten radii, four border widths, seventeen spacing rungs, six measures in ch, nine easings and three durations, all of it in its own stylesheet. There was nothing here to make up, which changes what the generator is for.
Where the style comes from
The reference is claude.com. Seven stylesheets carry the whole style (419,714 bytes, re-fetched 2026-09-16 across /, /pricing and /product/claude-code, which is what it takes to see all seven) and everything in the table below was measured out of those files rather than described from a look at the page.
| Decision | Measured |
|---|---|
| Faces | three, self-hosted: a sans (28 declarations), a serif (21), a mono (7), each shipping a roman and an italic |
| Ramp | 21 steps at 50-unit granularity, L 0.9818 down to 0.1908, plus pure white and pure black at the ends |
| Ramp hue | 19 of the 21 steps carry chroma, at hue 95–107; two of them (850, 900) leave for hue 68 |
| Named colours | 13: one brand terracotta with two darker rungs, and twelve illustration colours named after materials |
| Type ramp | 15 steps: 2 display, 6 headline, 2 large body, 3 body, a caption at 12px and a micro at 10px |
| Fluid | every step above 19px is a clamp(), and all of them are solved between 320px and 1440px |
| Leading | a named ladder of seven (1, 1.1, 1.2, 1.3, 1.5, 1.6, 1.7) where every reading step takes 1.6 and every display step takes 1.1 or 1.2 |
| Tracking | six distinct values in 373 kB, and the display steps are not among them: the largest type is set at normal tracking |
| Weights | 300 to 700, plus 480 for a button, which only a variable face can express |
| Radii | 2 · 4 · 6 · 8 · 12 · 16 · 24 · 32 · 48 · 64px, and the four in real use are 8, 12, 16 and 24 (21, 19, 21 and 16 declarations) |
| Borders | 0.5 · 1 · 1.5 · 2px |
| Spacing | 17 rungs, 2px to 200px |
| Section rhythm | three rungs, 64→96 · 96→128 · 128→200, plus an even half-step at 52→64. The 32→64 clamp beside them is the page's own side gutter, --container-margin, and is not a rung |
| Measures | six, in ch: 20 · 30 · 45 · 60 · 70 · 80 |
| Holds | 660 · 960 · 1192 · 1440px |
| Easings | nine, named by curve rather than by use, the extreme of them cubic-bezier(1, 0, 0, 1) |
| Durations | three: 60ms · 200ms · 450ms. .3s appears eighteen times as a literal and never as a declared rung |
| Shadows | three, and the largest is 16px of blur at 10% black |
The structural finding, and the reason this is a system rather than a palette. Every step of that ramp is warm. The lightest is a paper at hue 95 and the darkest is an ink at hue 107, with a real chroma on nineteen of the twenty-one steps, so a page in this style is not a grey page with a colour on it, it is paper. That is also why neither end of the ramp is white or black even though both are declared: the two pure values sit outside the family and are used for a label on a fill rather than for a ground.
The second finding is about what a published palette leaves out. A ramp is not a palette until something says which step is body text and which is a rule, and the reference's own semantic layer answers that for its pages, not for somebody else's. So claude_tokens.py solves the assignment: each role is the step nearest its limit that still clears its WCAG ratio against the worst ground in its theme rather than against the page background, which is the mistake that ships a secondary colour legible on a card and illegible in the well beside it. Fifty pairs, both themes, none failing.
The three things that came out of doing it that way
- The brand terracotta cannot be text on the paper. It measures 2.96:1, which clears neither the 4.5:1 body text owes nor the 3:1 a control boundary owes; its two darker published rungs reach 3.68:1 and 3.85:1 and stop. So a text rung is solved at the same hue and chroma (the only colour in the emitted file the reference does not publish) and the brand colour itself stays a fill and a mark.
- White on terracotta is 4.05:1, and the reference puts white on terracotta. The label on the fill is the near-black here instead, at 4.55:1. Darkening the brand colour until white cleared was the other option and it was refused: this style publishes that colour, and moving it to keep a white label spends the thing the style is for. The reference's own primary button token is the far end of the ramp rather than the brand colour, which is the same conclusion reached from the other direction, so the filled control on this page is the ink at 17.50:1.
- The illustration palette is theme-independent. All twelve of those colours are lighter than L 0.84, so the near-black is the only label either theme can put on them (11.42:1 to 13.31:1), and a note keeps its colour when the page inverts. They are grounds for a mark, never ink: every one of them is under 1.6:1 as text on the paper, which
--self-testasserts in both directions so that spending one as a colour is impossible rather than discouraged.
The direction
- Archetype. Split. Two documents in one grid, so a source paragraph and its translation are the same row and cannot drift apart while somebody scrolls.
splitis taken once elsewhere in this portfolio, bysap-fiori, and that page spendstable-first. - Stance. Prose-first. The heading is the book, the sentence under it is what the chapter is waiting on, and the counts are evidence beneath the sentence rather than four tiles across the top. The product is two columns of prose and a decision about a word; anything tile-first here would be a dashboard about a novel.
- Density. Comfortable, and read rather than scanned: 17px at 1.6, a forty-five character measure in each column, and a fourteen-rem margin between them that is a column of its own rather than an overlay.
- The bold thing. No toolbar. Every control that acts on the translation sits in the margin at the segment it changes, and the one warm colour does exactly one job per screen: it marks the segment waiting on a person. Three controls sit outside that rule and none of them touches the book: the way back to the portfolio, the balance of the two columns, and the theme. Balance and theme are the same kind of thing, which is the line the page draws.
The hard state
CONTRACT.md §3b asks for the state the archetype makes hardest, and for a split that is non-correspondence: the shape pretends the two sides pair up one to one, and a real translation does not. Three of the six segments on the page are that, drawn as ordinary states of a bilingual document rather than as errors:
- 28: one paragraph into three. Her sentence turns twice on semicolons and English will not hold both turns in one breath, so the pair is deliberately not one to one and the margin says so.
- 29: a sentence with no source. The translator's own clause, added for a reader who does not know what a conventillo was. The source half is a written-out absence rather than an empty cell.
- 31: held for rights. Ocampo-Salas quotes a stanza from 1911. The Spanish is out of copyright and the 1978 English translation the press wants is not, so the segment cannot ship and the chapter goes without it. The control that would quote it is present, disabled, styled as disabled, and carries the reason beside it.
And 30 is the one waiting on a person: a term with three defensible renderings and no right answer, each with what it costs, on the page's only ground of its own. Choosing one writes it into the translation, settles the segment, and takes one off the count the chapter head and the delivery panel both report.
Generated
python3 tools/claude_tokens.py --both > systems/claude-brand/tokens.css
python3 tools/claude_tokens.py --check # 50 pairs, both palettes
python3 tools/claude_tokens.py --self-test
Measured
python3 tools/claude_tokens.py --check: 50 pairs across both palettes, 0 failing. The pairs the page leans on hardest:
| What sits on what | Carries | Ratio |
|---|---|---|
#141413 on #faf9f5 | the source, the translation, every heading | 17.50:1 |
#4d4c48 on #f0eee6 | every margin line, on the worst ground | 7.40:1 |
#5e5d59 on #f0eee6 | the quietest label, on the worst ground | 5.67:1 |
#141413 on #c6613f | the label on the terracotta mark | 4.55:1 |
#ad5030 on #f0eee6 | every link, on the worst ground | 4.56:1 |
#87867f on #f0eee6 | every control boundary, on the worst ground | 3.15:1 |
#141413 on #e3dacc | the marked term, both themes | 13.31:1 |
Dark theme, the same roles: 16.66:1, 6.82:1, 5.38:1, 5.90:1, 4.51:1, 3.14:1 and 13.31:1.
The decorative rule is held deliberately below the 3:1 a control boundary owes (2.11:1 light, 2.04:1 dark), so substituting it for a stroke has to be somebody's decision. --self-test asserts that, and asserts that the step one rung past each solved limit fails, without which every pass above is uninformative.
The highlight under segment 30 is solved as a ground rather than as a colour: the faintest wash of the brand colour into the paper that is still visibly another ground, at ΔEOK 0.02, which lands at 6% and carries the body ink at 16.51:1 light and 15.83:1 dark. Solved the other way (the strongest wash the ink still clears), it came out at 100%, which is the brand colour itself, and a highlight the same colour as the mark beside it is one signal doing two jobs.
Browser, both themes: a11y-probe.mjs clean at 1280px with every run catching its control probe; shots.mjs clean at 1440px and 375px, document 1440/1440 and 375/375; drive-probe.mjs drove thirteen controls without a throw, an emptied region or a lost focus. Driven by hand as well: choosing a rendering moves the counts 3 → 2 and 40 → 41 and back on reopen, the balance writes the column header and the segment list to the same three tracks (428/224/428 even, 548/224/308 source-wide), and interrupting the swap mid-exit still lands the last choice, which is the transitioncancel path.
Deviations, and their reasons
- The three faces are not the reference's. They are proprietary and self-hosted, and CONTRACT.md §2 admits Google Fonts only, so each is substituted by an open-licensed variable face doing the same job: Instrument Sans where the interface speaks, Source Serif 4 where the book does, JetBrains Mono for the numbers. What is kept is the division of labour (a serif that reads and a sans that operates), which is the part of the decision that belongs to the style rather than to the foundry.
- The fluid sizes keep the measured endpoints and not the measured form. The reference's clamps carry a px constant plus a vw term (
28.857px + 1.607vw), which between the two bounds ignores the reader's own font-size setting. Every step here is re-solved through the same two widths inrem, whichtypography's own solver warns about and is the reason it is not a transcription.
- Four literal colours, in the two places a
var()cannot reach. The favicon'sdata:URI and the twotheme-colormetas are parsed outside CSS. Each value is transcribed fromtokens.cssand the comment above them names which token it is.
- The caption step is 12px, and APCA licenses no body text that small at any contrast. Measured: the near-black on paper is Lc 101 and still fails APCA's font table at 12px/400, as does every other pair on the page, because the table itself starts at 15px. It is the published step and this page keeps it, at weight 500 or above, for labels and margin notes rather than for anything read at length; the two state marks that sit lowest on APCA (Lc 36 light, Lc 45 dark) each carry a word and a shape as well as a colour. The gate this portfolio holds is WCAG 2.2 AA, which every pair clears.
Findings kept, with reasons
slop_scan·default-ai-palette: cream and terracotta is named as one of the three palettes generated design converges on, "legitimate if the brief asked for it". The brief is this palette: every value is the reference's own.direction_check --ledger·hero-undecided, andtype scale flat/shape unset:centaurireports the same hero warning for the same reason (a working surface has no first screen to decide about), and the twounsetlines are the known blind spot recorded intools/distinctness.sh: a fully tokenised page, which §1 mandates, declares no literal sizes or radii for that tool to count.copy_lint·jargon 'input'×3: matching the<input type="radio">attribute, not the copy.title-case×1: it is a book's title.label-period×20: full sentences of prose, not labels.craft_lint·disabled-unstyledand twohover-no-transition: the disabled state is.btn[disabled]in a stylesheet rather than adisabled:utility, and the two modifier selectors inherit their transition from.btn.keyboard_probe·unreachable-control×2 per width: the two unchecked radios in the balance group. A radio group is one tab stop and the arrow keys move inside it, which is the native behaviour the page is relying on.audit_page·apple-touch-icon-missing,manifest-missing: a specimen page is not an installable application.
The logo-swap test, and why its number is high here
tools/distinct-one.sh claude-brand: 22 pairs, worst 0.74 against a ceiling of 0.75, which passes by one hundredth and is worth explaining rather than celebrating. Every point of it is the instrument. Against luxury-hospitality, the pair that scores highest, five of the seven dimensions come back 1.00: radii and type_steps because CONTRACT.md §1 forbids a literal and both pages are fully tokenised, so the tool compares two empty sets and scores them identical; hero, card_density and bleed because both pages have none of the thing being counted. The one dimension with real content in it, grid_shapes, is 0.17: a two-column bilingual document with a margin track and a photography-led hospitality page do not share a shape. The same holds for centauri at 0.73, where grid_shapes is 0.10.
That is the floor tools/distinctness.sh's own header warns about, and this page sits closer to the ceiling than any other in the set because it declares no literals at all and has no hero, no cards and no bleed to be counted against. Reading the number without the breakdown would report a resemblance that does not exist.
What I am least sure about
Whether the margin should hold the decision at all. The bold thing says every control sits beside the line it changes, and the three renderings with their costs are the most valuable thing on the page; a fourteen-rem column is where they fit worst. They were three cards at twelve rem and every label broke inside its own border; the boxes came off and the margin went to fourteen, which is legible now and still the narrowest column on the screen. The alternative is a panel that opens over the translation, and I think it is worse (it covers the text the decision is about at the moment somebody is deciding), but it would let the three costs be read side by side, and somebody could reasonably prefer that trade.
Uplift, 2026-10-29
The archetype, the people and the tokens are unchanged. What changed, in the order a visitor meets it:
- A title page for the first fold. The Spanish title at the display step in the serif's display optical size, the English title under it, a colophon hanging on a rule, and the sentence about what the chapter waits on in the large body step. Labels are real small capitals, tracked.
- The ledger. Forty-four marks on a rule, one per segment: short for settled, tall and warm with a dot for waiting, tall and ink with a bar for held. The sum under it (40 + 3 + 1 = 44) is live and is written from one function with the sentence under the title and the list of what the chapter owes.
- Paired phrases. Ten phrase pairs, keyed in both columns. Pointing at or focusing a phrase washes both sides; selecting it opens a translator's note in the margin that names the move (borrowing, tense shift, addition). The English phrases are links to their notes, so a keyboard walks them. The wash is not the accent, which stays with the segment waiting on a person.
- The decision follows the term. Choosing a rendering for sereno rewrites segments 27 and 30, not 30 alone, because the chapter needs the word to agree with itself. Counts, sentence, ledger mark and owed list move together.
- Setting. Reading size is body-1 at 1.6, old-style figures in prose, hyphenation by language, a two-line initial and small capitals on the scene opening, one paragraph gap (half a line) shared by both columns, true apostrophes, en dashes in ranges. The spread uses the 1440 hold at wide widths with a fluid margin of 14 to 17rem.
- Phone. State above the text, the note and decision below it, and the balance as a thumb-height pill at the bottom of the screen.
- Dates agree. The page's today is Thursday 29 October 2026; the chapter is due Friday 6 November, in 8 days.
Findings kept: boson's focus-not-visible is an artefact of its copying only index.html (it vanishes with the tokens inlined); overlapping-controls on adjacent inline phrase links is the bounding boxes of wrapped lines; fixed-height-clipped-text on the 64px title is the style's published 1.1 display leading, nothing is clipped; target-too-small on the radios is a 44px label.
Round four
- The first-screen rule is now keyed: the three counts under it carry the same three marks (short rule, warm dot, flat bar) in the same colours, and one line says what each means and where the numbers sit. Nothing on the rule itself moved.
- The decision's radios are drawn (ring plus inner dot from the ink tokens) instead of the native control. In the dark theme the native one took one pale ink for ring and dot, so chosen and unchosen were the same disc; now unchosen is a ring in the third ink and chosen is a full-ink ring with a dot. Same drawing in the light theme and for the balance control's inputs, which are visually hidden and unaffected.
Round eight
- The sticky Verso/Recto head ran past the spread. Real. The head was a sibling of the segment list inside the page's main container, so its sticky range was that whole container and it stayed over the delivery list and the design notes. The head, the list and the two "settled" rules now sit in one
.spreadwrapper, so the head leaves with the last segment. Its ground was already opaque; what showed beside it were the decision panel's tinted corners in the gutters, so the ground now runs to both screen edges (a shadow spread and a clipped box). The critic's "line half-cut under it" is any line scrolling under a hard edge; an earlier capture also caught the page mid smooth-scroll, which draws a button's lower edge inside the head. Checked at instant scroll: the head covers it. - The floating pill sat over the ENGLISH label (390). Real, in the sense that a pill floating over a column leaves text visible at both sides of it. On a phone the balance control is now a bar across the foot of the screen with its own ground and a top rule, so lines cut cleanly at its edge. The three choices and the 40px targets are unchanged; the end-of-page padding already cleared it.
- Declined: nothing.
Round nine
Scored against the ten heuristics before and after (0 to 4): visibility 2 to 3, real world 4 to 4, control 3 to 3, consistency 3 to 3, error prevention 2 to 3, recognition 2 to 3, flexibility 2 to 2, aesthetic 3 to 4, error recovery 2 to 3, help 3 to 3. Total 26 to 31. What stays below 3: flexibility (no way to jump to the next waiting segment from the keyboard) and consistency (segments 36 and 41 say "Waiting on you" and carry a note but, unlike 30, no control).
- The page claimed 33 to 44 were settled, and two of them were not. The ledger and the list of what the chapter owes both said 36 and 41 were waiting, and neither could be reached. The two are now drawn (a pronoun of address, and a pun on hora), each with a note and a paired phrase, and the fold is split into 1-26, 33-35, 37-40 and 42-44. Counted from the page: eight segments drawn, thirty-six folded, forty-four in all; forty settled, three waiting, one held. The bracket on the rule now covers 27 to 41, its label sits above it, and the eight drawn marks are the dark ones.
- The delivery sentence went stale. Choosing a rendering moved the sum and the sentence under the title but left "Four segments are not yet settled, and three of them..." as it was. It is now written from the same function.
- Four buttons did nothing. Chase the clearance, Translate it fresh, Send chapter to Tom and Export the open decisions each answer beside themselves and to the status region. Send asks twice while a decision is open and says so; Export downloads a plain list of the open items.
- Focus under the sticky head. Walking backwards by keyboard landed a link under the Verso and Recto head. Anything focusable in a segment, and the list of what is owed, now scrolls clear by 80px on both sides.
- Phone. The colophon is a label and name table below 40rem, so no name breaks across two lines at 360; the sum fits on one line at 360; every control is 44px under a coarse pointer (the small buttons, the theme control, the portfolio mark, the balance pills, the links in the owed list); the balance bar steps away once the spread's last line is in the top third of the screen instead of sitting over the delivery list and the notes.
- Two rules that stacked at each fold are now one.
Checked and left: the 15px caption line in the margin sits at the published caption step; the three inline phrase links under 44px are inline in a sentence, which the target-size rule exempts.
Round ten
A cold second read, scored harder than round nine. Before and after (0 to 4): visibility 3 to 3, real world 4 to 4, control 2 to 3, consistency 2 to 3, error prevention 2 to 3, recognition 3 to 3, flexibility 2 to 3, aesthetic 4 to 4, error recovery 2 to 3, help 3 to 3. Total 27 to 32 of 40. Still below 3: none; visibility, recognition and help sit at 3 because the ledger marks are not themselves clickable (a ruler of 44 marks reads as a jump bar and is not one).
- Send had no way back (severity 3). One press, then a dead button and no undo. Sent now offers Recall from Tom, which returns the chapter to a draft and puts focus on Send; the pressed button was disabled and dropped focus to the page, so focus now moves to Recall.
- A decision after sending left Tom with the old chapter (3). Choosing, keeping or reopening anything after Send now says so (a line under the title, a line in the delivery text, a status announcement) and turns the button into "Send the update to Tom".
- 36 and 41 said "Waiting on you" and offered nothing to do (3). Each has a "Keep the draft" control and a Reopen, driven by the same count function as 30. Counts, ledger mark, owed list, the sentence under the title and the delivery sentence are all written from the segments, so 0 to 3 waiting all read correctly ("No decision waits on Isabel now").
- The result landed out of sight (2). On a phone the changed counts are a screen and a half above the decision. The confirming line now carries the count ("41 of 44 settled, 2 waiting, 1 held") and is scrolled into view.
- No way to the next waiting segment (2, flexibility). Each decision margin ends with "Next waiting: segment N" (or "On to what the chapter owes"), and N jumps to the next one from anywhere, wrapping, landing on its control. The key is stated in the line under the ledger.
- Buttons that answered and changed nothing (2). Chase the clearance now disables and reads "Reminder sent". Translate it fresh toggles (pressed state, the English side says it is open, a second press undoes it). Export lists only what is still unsettled.
Invented, since the brief gives no rule: Recall from Tom (undoes a send; the page does not say whether Tom has opened it); "Keep the draft" as the decision for 36 and 41 (the draft is already on the page, so keeping it settles the segment); "Translate it fresh" leaves 31 held until a version exists, so the held count stays 1; "Send the update to Tom" after a post-send change.
Left: the ledger marks and numbers are not links (severity 1, they are a decorative aria-hidden rule); 31 has no way to leave the held state because the brief gives no rule for what clears it.
Round eleven (2026-11-10)
- No tappable way to the first waiting segment on a phone, and a key hint shown on a screen with no keyboard (severity 2). Reproduced at 390 dark: segment 30 sits 3,215px down, and the only links above it were inside the first segment. Fixed: under the ledger sum a "Waiting on you" line holds one link per segment still waiting (Segment 30, 36, 41), 44px tall under a coarse pointer, landing on the segment's control exactly as N does and writing the same status line. A link leaves the line when its segment is settled and comes back on Reopen; the whole line is hidden at none waiting. The how-to now carries two sentences, one per input: with a fine pointer "Select a warm number on the rule, or press N anywhere, to go to a segment waiting on you"; under a coarse pointer "Tap a segment under Waiting on you to go to it; each decision then links on to the next one waiting".
- The warm pins on the rule looked like jump marks and did nothing (severity 2). Reproduced at 1440 light: a click on the 36 label left scroll at 0, hash empty, cursor default. Fixed by making them do what they look like: a click on a waiting number or its dot goes to that segment (same landing as the new links). Only a segment still waiting is clickable; a settled one loses its number and its pointer cursor. The rule stays aria-hidden, so this is a pointer shortcut and the "Waiting on you" links are the keyboard and screen-reader route to the same places.
- Invented: the label "Waiting on you" and the link text "Segment N" (both from the page's own vocabulary); the choice of pointer type, not screen width, to pick the how-to sentence. Nothing in the brief's facts changed.
Round thirteen (2026-11-10)
- The phone how-to quotes a key (reviewer, severity 1). Not reproduced: at 390 dark and 360 light (isMobile, hasTouch, deviceScaleFactor 3,
(pointer: coarse)true) the how-to under the ledger already reads only "Tap a segment under Waiting on you to go to it; each decision then links on to the next one waiting", and at 1440 and 1920 it reads only "Select a warm number on the rule, or press N anywhere, to go to a segment waiting on you". A phone-width window with a fine primary pointer (no touch emulation) still shows the key sentence by design. Not edited for it. - Audit of every key reference the page can show. The only other one was the status and reply line when Send is pressed with decisions open, which ended "Press again to send as it stands." on every device. Fixed: it now reads "Tap again to send as it stands." under
(pointer: coarse)and keeps "Press again" otherwise, tested when the line is written. The pointer-only how-to sentence "Dotted phrases ... Select one" names no key and stays. The N key, Escape and Space have no text of their own beyond the how-to sentence. - Invented: nothing new; "Tap again" is the verb the touch how-to already uses.