/* ========================================================================== TRU72 — SHARED DESIGN TOKENS (Rev06 brand) -------------------------------------------------------------------------- ONE source of truth for colour, type, spacing and shape on EVERY Tru72 site. COLOUR IS THE APPROVED FIVE AND NOTHING ELSE. Bethany confirmed this palette on 2026-09-09; it is what the Rev06 brand guide records. It is NOT the North Carolina build's as-shipped colour set — NC drifted (B8952A, F5F0E8 and a rust family that the rebrand dropped). The layouts, structure, components and typography come from NC, which is correct. The colour comes from here. INK #3D3F3F SCENERY GREEN #696859 ANTIQUE GOLD #C39954 LINEN #CBC0AA PAPER #FFF6DC FIVE, PLUS ONE THAT DOES EXACTLY ONE JOB. (D83, 2026-09-22) GOLD-AGED #A8854A the italic display line, on PAPER, nowhere else D42 dropped GOLD-AGED and the palette ran at five. D83 puts it back, and deliberately puts it back NARROW: on the live T. Byrd's site the italic second half of a display headline is warm gold on cream, and that is the single most recognisable thing about the reference's typography. ANTIQUE GOLD cannot do that job — 2.43:1 on PAPER, a failure at every size, which no amount of making it bigger rescues. GOLD-AGED measures 3.18:1, which clears WCAG's 3.0 for LARGE text, so it does the job legitimately. It is NOT a general-purpose sixth colour, and the cage below is built so it cannot become one. There is still no rust, no B8952A and no F5F0E8, and there is no seventh colour. Identical on all four sites, no per-site variation. A contrast problem that cannot be solved inside these six gets REPORTED, not solved with another colour. --------------------------------------------------------------------------- THE GOLD-AGED CAGE — why it cannot be misused (D83) GOLD-AGED is declared ONCE, as --t72-gold-aged, and read ONCE, by --t72-display-italic-on-paper. Nothing else in any stylesheet on any of the four sites may name either one. tools/audit_phase5.py check 15 fails the build if anything does, and it also re-derives the two facts that make the colour legal: THE GROUND MUST BE PAPER. Measured on each light step — PAPER #FFF6DC 3.18:1 passes as large text PAPER-WARM #F6ECD3 2.92:1 FAILS, even as large text PAPER-DEEP #F0E7CE 2.78:1 FAILS, even as large text LINEN #CBC0AA 1.91:1 FAILS SCENERY GRN #696859 1.64:1 FAILS — and D83 says never on green INK #3D3F3F 3.09:1 passes, but ANTIQUE GOLD is 4.04:1 there and is already the display colour on ink, so ink keeps it The margin on PAPER is 0.18. A tint is enough to lose it. So the two LINEN-in-PAPER tints — .t72c-lite-warm and .t72c-lite-deep, which relevel() can put on ANY light band at build time — take the colour BACK to SCENERY GREEN, and so does every LINEN and green band. THE TYPE MUST BE LARGE. 3.18:1 clears AA for large text only: 24px, or 18.66px at weight 700. The italic line is Fraunces at weight 400, so the threshold is the full 24px. Only two rules in the system may read the role, and each is floored so its SMALLEST rendered size clears it: .t72c-h1 em .82em of clamp(40px, 4.6vw, 62px) -> 32.8px minimum .t72c-h em max(24px, .74em) of clamp(30px,…) -> 24px minimum Never body text, never a label, never an eyebrow, never a button fill, never a border or a rule — ANTIQUE GOLD keeps every decorative job. THIS BLOCK IS THE SWITCH. Every colour on all four sites resolves through the six declarations below. Changing Bethany's mind — for example adopting NC's as-built values instead — is editing this one block, not a rebuild. --------------------------------------------------------------------------- THE CONTRAST RULES THAT FOLLOW FROM SIX COLOURS (measured, WCAG 2.1) on PAPER INK 12.1:1 ok · SCENERY GREEN 5.2:1 ok ANTIQUE GOLD 2.4:1 FAILS at every size on INK PAPER 9.7:1 ok · LINEN 5.7:1 ok ANTIQUE GOLD 4.0:1 — large type only (24px+, or 18.7px bold); never small type on SCENERY GREEN PAPER 5.2:1 ok LINEN 3.1:1 FAILS small · ANTIQUE GOLD 2.2:1 FAILS at every size. There is NO gold type on green, ever. on LINEN INK 5.9:1 ok SCENERY GREEN 3.1:1 FAILS small · ANTIQUE GOLD 1.5:1 and PAPER 1.7:1 FAIL at every size. On a LINEN band the type is INK, all of it. So: gold is DECORATION as TYPE — rules, keylines, borders, hairlines — and a FILL only where D46's size goes with it. INK on ANTIQUE GOLD is 4.04:1, which clears AA large text, so a gold button is legal at 24px, or at 18.66px bold, and illegal below that. PAPER on ANTIQUE GOLD is 2.6:1 and is never used. --------------------------------------------------------------------------- THE TWO TINTS, AND WHY THEY ARE NOT A SEVENTH COLOUR A six-flat-colour site has no way to say "this band is not the last band". Rather than introduce a hue, the two light surfaces below are LINEN mixed into PAPER with color-mix(), so they carry no pigment that is not already in the palette and they move automatically if PAPER or LINEN move. The hex on the line above each is the pre-computed fallback for an engine without color-mix, and is the same mix arithmetic. No tint is ever used to solve a contrast problem, and no type colour is a tint — type is always one of the six. ========================================================================== */ /* -------------------------------------------------------------------------- THE DISPLAY FACE — self-hosted, one weight -------------------------------------------------------------------------- Friz Quadrata Bold is the only weight that exists, so it is display and titles only; Fraunces carries everything smaller. Nothing here synthesises a lighter or heavier cut. `font-weight: 400 700` is not a claim that the file holds a regular — it is how you stop a browser faking one. A page asking for weight 400 in this family matches the descriptor, so the real Bold outlines are drawn instead of a synthesised thin. `font-synthesis: none` on the body is the belt to that braces. Full face, not subset: the .otf already carries only 209 glyphs (Latin plus the punctuation the sites use) and the woff2 is 17 KB, so subsetting would save a rounding error and risk dropping a character. -------------------------------------------------------------------------- */ /* THE OPTICAL SIZE CORRECTION — size-adjust (Phase 20) Bethany: our headlines render bigger and bolder than the reference's. They do, and it is not a size token, letter-spacing or a synthesised bold — BOTH h1s compute to 58.88px / weight 700 / .005em / line-height 55.936px. It is the FACE, and the reason is that the reference never had ours. The reference site ships NO font file and NO @font-face at all. Its display stack is `'Friz Quadrata Bold', Cinzel, serif`; it loads Cinzel (plus Fraunces and Geist Mono) from Google Fonts, Friz Quadrata resolves to nothing, and every display heading on it is therefore CINZEL. Ours is the real licensed face, self-hosted. Measured in the browser, `HAMBURGEFONS` at 700/59px, ink pixels counted on a canvas: face cap height width ink px Friz Quadrata Bold (ours) 46px 547px 13,099 Cinzel 700 (what the ref draws) 43px 528px 8,000 Fraunces 700 (our fallback) 43px 517px 9,870 7% taller and 64% more ink: "bigger and bolder", exactly. `size-adjust: 93.5%` (= 43/46) scales the outlines so the cap height lands on 43px, the reference's, and the measure comes to 512px against its 528px. The remaining stroke weight is the typeface and cannot be scaled away. Friz Quadrata is the brand display face (D43) and Cinzel is not a Tru72 face, so Cinzel is NOT being added to the stack to chase the last of it. IT ALSO FIXES THE SWAP. With `font-display: swap` the fallback paints first; before this the webfont arrived 7% bigger than Fraunces and the headline jumped. Now webfont (cap 43 / 512px) and fallback (cap 43 / 517px) are within 1%, so a late or failed webfont degrades to the same size — which is also what the reference's stack does, since its two entries never differ. Every line-height in the system is a unitless multiple of font-size, so no line box moves. Chrome 92+, Firefox 92+, Safari 17+; older engines ignore the descriptor and get today's rendering. */ /* -------------------------------------------------------------------------- THE DISPLAY FACE — CINZEL, self-hosted Bethany, 2026-09-28 -------------------------------------------------------------------------- This is what tbyrdhvac.com has rendered all along. Its CSS asks for `"Friz Quadrata Bold", Cinzel, serif` and ships NO font file, so every headline on the site Bethany has been matching is Cinzel, served from Google. We shipped Friz, so we rendered what T. Byrd's only ever asked for — 22% more ink in 10% less width, which is what she kept calling too thick. Self-hosted, so the estate still makes ZERO external font requests. Cinzel is SIL OFL; the licence ships beside it as cinzel-OFL.txt. One variable file, 400-900, latin subset, 25.9 KB. IT IS AN ALL-CAPS FACE — lowercase renders as small caps. That is safe here and checked, not assumed: 21 of 24 headings on a sample Oregon page already compute `text-transform: uppercase`, as do all three on the reference. The three that do not are the Fraunces italic accent phrases, which take the body face anyway. */ @font-face { font-family: 'Cinzel'; src: url('../fonts/cinzel.woff2') format('woff2'); font-weight: 400 900; font-style: normal; font-display: swap; } @font-face { font-family: 'Friz Quadrata Bold'; src: url('../fonts/friz-quadrata-bold.woff2') format('woff2'); font-weight: 400 700; font-style: normal; font-display: swap; size-adjust: 93.5%; } /* -------------------------------------------------------------------------- THE LABEL FACE — GEIST MONO, self-hosted (D95, Phase 21) -------------------------------------------------------------------------- D43 said two brand faces and only two, and the label voice — the wide, tracked, uppercase small type on nav, the utility strip, eyebrows, numbered labels and buttons — was Fraunces with tracking standing in for a mono. It does not stand in. Measured on the reference at 1280px, every one of those roles is 'Geist Mono': `.utility` 10.5px/.18em, `.nav-links` 12px/.2em/500, `.eyebrow` 11px/.22em/500, `.intent-num` 10px/.26em/600, `.btn` 10.5px/.16em/600. Our sheets already carry those exact sizes and that exact tracking — only the FACE was wrong, so a serif was being asked to do a monospaced job and the texture never arrived. Bethany, Phase 21: "Geist Mono is a THIRD typeface and supersedes the two-face rule (D43) for this UI role." It is a THIRD FACE WITH ONE ROLE, not a free-for-all. Display stays Friz Quadrata Bold; prose, decks, italics and card bodies stay Fraunces. Geist Mono is reached only through --t72-font-label / --t72-font-mono, which is the name every label rule in the system already uses. SELF-HOSTED, like the display face, and for the same reason: the reference pulls it from fonts.googleapis.com and we do not add a third-party font request. Geist Mono 1.401 is SIL OFL 1.1 (openfontlicense.org), which permits redistribution; the file is the upstream variable woff2 subset to Latin-1 plus the marks the sites use (·, ★, →, —, curly quotes) — 31 KB, one weight axis 100-900, so 400/500/600/700 are all real cuts and nothing is synthesised. Source of record: assets/_system/fonts/geist-mono.woff2. -------------------------------------------------------------------------- */ @font-face { font-family: 'Geist Mono'; src: url('../fonts/geist-mono.woff2') format('woff2'); font-weight: 100 900; font-style: normal; font-display: swap; } /* -------------------------------------------------------------------------- THE BODY FACE - FRAUNCES, self-hosted (D98, Phase 22) -------------------------------------------------------------------------- Fraunces is the prose voice: decks, card bodies, paragraphs, the gold italic display line, and the fallback behind Friz Quadrata. It was the last face still fetched from a third party - all 208 source pages linked fonts.googleapis.com for it, with a preconnect pair to match. Three external requests per page, on every one of the five builds, for a face we are licensed to host. Fraunces is SIL OFL 1.1 (assets/_system/fonts/fraunces-OFL.txt, shipped beside the files as the licence requires). Geist Mono's licence, which Phase 21 named but never shipped, is now beside it as geist-mono-OFL.txt. BUILT TO MATCH WHAT GOOGLE WAS SERVING, EXACTLY. The link these pages carried was family=Fraunces:opsz,ital,wght@9..144,0,300..900;9..144,1,300..900 so what the browser actually received was the upstream variable font with SOFT and WONK pinned at their defaults (SOFT 0, WONK 1 - Fraunces ships `wonky` ON) and opsz 9..144 plus wght 100..900 left live, declared to CSS as `font-weight: 300 900`. These two files are the same upstream source, instanced the same way, subset to Latin-1 plus the punctuation the sites use, and they declare the same `300 900`, so no weight that used to resolve to a real cut now clamps to a different one. VERIFIED, not asserted. Both files were diffed against the woff2 that fonts.gstatic.com was serving for that URL: 1,176 glyph outlines each (28 characters x 7 weights 300-900 x 6 optical sizes 9-144), 0 differing, plus 435 advance widths each and every vertical metric - upem, ascender, descender, line gap, win ascent/descent, x-height, cap height - identical. The opsz axis is KEPT, so `font-optical-sizing: auto` still reshapes the face by size the way it did; pinning it would have been smaller and wrong. TWO FILES, because we use true italics. The gold display line, the pull quotes and every `em` in a deck are Fraunces italic, and a single roman file would have let the browser synthesise a slant instead of drawing the real italic - a different face, not a tilted one. 70 KB roman + 82 KB italic, against the 66 KB + 80 KB Google was sending for the same coverage. -------------------------------------------------------------------------- */ @font-face { font-family: 'Fraunces'; src: url('../fonts/fraunces-roman.woff2') format('woff2'); font-weight: 300 900; font-style: normal; font-display: swap; } @font-face { font-family: 'Fraunces'; src: url('../fonts/fraunces-italic.woff2') format('woff2'); font-weight: 300 900; font-style: italic; font-display: swap; } :root { /* ======================= THE APPROVED FIVE ============================ */ --t72-paper: #FFF6DC; /* PAPER — the page */ --t72-ink: #3D3F3F; /* INK — type, dark surfaces */ --t72-green: #696859; /* SCENERY GREEN — muted type, mid band */ --t72-linen: #CBC0AA; /* LINEN — type on ink, hairlines */ --t72-gold: #C39954; /* ANTIQUE GOLD — decoration only */ /* ====================================================================== */ /* ---- THE SIXTH, CAGED (D83) ------------------------------------------- Declared here, read on the next line, and named nowhere else on any of the four sites. Do not reach for --t72-gold-aged in a stylesheet; reach for the role, and only from a rule that is on PAPER and 24px or larger. Audit check 15 enforces both halves of that sentence. */ --t72-gold-aged: #A8854A; /* GOLD-AGED — 3.18:1 on PAPER, large only */ --t72-display-italic-on-paper: var(--t72-gold-aged); /* Surface tints — LINEN in PAPER. No new pigment. Never used for type. */ /* PAPER WARM IS NOW AN ALIAS OF PAPER DEEP, not a colour of its own. Measured: #F6ECD3 on #F0E7CE is 1.049:1, which is no seam at all -- the two were one tint with two names. Bethany dropped it from the palette on 2026-09-29. The NAME survives because roughly forty rules across four stylesheets read it, and repointing it is safer than editing all of them; every one of those rules now paints the single tint. Do not give this a value of its own again. */ --t72-paper-warm: #F0E7CE; --t72-paper-warm: var(--t72-paper-deep); /* alias — see above */ /* 28%, not 34%. At 34% this tint measured #EDE4CB, and SCENERY GREEN type on it is 4.45:1 — under the 4.5:1 AA needs for normal text, by a hair. The fix stays inside the five (D42): the same two colours, mixed a little lighter. At 28% the pair measures 4.57:1 and INK measures 8.59:1. Do not darken it again without re-measuring green on it. */ --t72-paper-deep: #F0E7CE; --t72-paper-deep: color-mix(in srgb, var(--t72-linen) 28%, var(--t72-paper)); /* Rules, drawn from INK and the gold at low alpha. */ --t72-rule: rgba(61,63,63,.14); --t72-rule-strong: rgba(61,63,63,.32); --t72-rule-on-ink: rgba(203,192,170,.24); --t72-rule-gold: rgba(195,153,84,.55); /* the hairline, ANTIQUE GOLD */ --t72-rule-gold-soft: rgba(195,153,84,.22); /* the nested keyline */ /* A PANEL NESTED ON A DARK SURFACE. (Phase 14) The reference site's "pick your market" cards are a TRANSLUCENT dark laid over the SCENERY GREEN booking card at about a third opacity, and the translucency is part of the effect: the panel reads as a recess in the green rather than as a second, unrelated block. A solid fill does not look the same. It is reproduced with INK at 34%, rather than with the second charcoal the reference reaches for, so it adds no pigment that is not already in the five. Over SCENERY GREEN it resolves to about #5A5A50, on which PAPER measures 6.46:1 — better than PAPER on flat green (5.23:1), so nesting the panel improves legibility rather than costing it. */ --t72-panel-on-dark: rgba(61,63,63,.34); --t72-panel-on-dark: color-mix(in srgb, var(--t72-ink) 34%, transparent); /* Semantic roles — what to reach for, so no page has to re-derive the contrast table above. */ --t72-type-on-paper: var(--t72-ink); /* 12.1:1 */ --t72-type-on-paper-muted: var(--t72-green); /* 5.2:1 */ --t72-type-on-ink: var(--t72-linen); /* 5.7:1 */ --t72-type-on-ink-strong: var(--t72-paper); /* 9.7:1 */ --t72-type-on-green: var(--t72-paper); /* 5.2:1 — the ONLY one */ --t72-display-on-ink: var(--t72-gold); /* 4.0:1 — 24px+ only */ /* ---- Type ------------------------------------------------------------- */ /* TWO BRAND FACES, AND ONLY TWO. Friz Quadrata for display and titles, Fraunces for accents and secondary. The two decorative faces Oregon used to load locally are retired permanently, and so are the two non-brand webfonts the NC build reached for. None of them is a Tru72 face, and none of them is loaded anywhere any more. docs/PHASE3B_OR_RESKIN_REPORT.md names them. Friz Quadrata Bold is licensed for web use and SELF-HOSTED from the site's own assets — no third-party font CDN. The @font-face below serves the woff2 built from the supplied .otf; the .otf stays in assets/fonts/ as the source of record. The .ttf copy in that folder is NOT used: it is missing the middot and the accented glyphs and renders empty boxes. THREE, since D95. The label voice — the wide, uppercase, tracked small type on nav, the utility strip, eyebrows, numbered labels and buttons — is GEIST MONO, self-hosted, and it is the third face. It used to be Fraunces with tracking; the reference has always drawn it in a mono and the substitution was visible at every size. See the @font-face above for why this supersedes D43 and how narrowly it is scoped. */ /* PHASE 26 — THE DISPLAY FACE IS FRAUNCES 600, AND IT IS ONE SWITCH. Bethany: "the title font is still thicker than tbyrds". It is, and the reason is that tbyrdhvac.com asks for `'Friz Quadrata Bold', Cinzel, serif` and SHIPS NO FRIZ FILE — so it has always drawn Cinzel. We ship the real Friz, so we draw what it only ever asked for. Measured in the browser, "Heritage. Standard. Method." at 48px, ink pixels on canvas: Fraunces 400 5,677 ink 615px wide Fraunces 500 6,498 ink 628px Fraunces 600 7,356 ink 640px Cinzel 700 (the ref) 8,076 ink 759px Fraunces 700 9,175 ink 665px Friz Quadrata Bold 9,866 ink 686px Friz is 22% more ink than Cinzel in 10% less width — ~35% denser per unit of measure. It is a single bold master; there is no lighter cut to pick and `font-synthesis: none` (correctly) forbids faking one. Cinzel is not an approved Tru72 face, so the closest shippable match to what the reference actually renders is FRAUNCES 600 — an approved face, already self-hosted as a 300-900 variable, so weight 600 costs no new request. Friz Quadrata Bold stays in the stack and the file still ships. */ /* CINZEL, at the reference's own weight. Friz and Fraunces stay in the stack and both files still ship, so either is one edit away. */ --t72-display-weight: 700; --t72-font-display: 'Cinzel', 'Friz Quadrata Bold', 'Fraunces', Georgia, 'Iowan Old Style', 'Times New Roman', 'DejaVu Serif', serif; /* THE ACCENT'S DEFAULT GOLD, declared here because a HAND-WRITTEN band declares no ground and therefore no `--t72c-gold-ok`. Ten display italics on the ported NC pages resolved to nothing and fell back to their heading's own colour -- ink on cream, paper on dark -- while every component heading went gold. Cream is the default ground, so the default gold is the one legal on cream: GOLD-AGED at 3.18:1. Bands that paint something else restate it (see tru72-components.css). Declared in THIS sheet, which is the one place D83's cage exempts from its reader count. */ --t72c-gold-ok: var(--t72-display-italic-on-paper); --t72-font-body: 'Fraunces', Georgia, 'Iowan Old Style', 'Times New Roman', 'DejaVu Serif', serif; /* The mono fallbacks are the platform monos, NOT the serif stack: if the webfont is slow or blocked the label voice must still degrade to a monospace texture, not back to Fraunces, or the swap moves every label. */ --t72-font-label: 'Geist Mono', ui-monospace, 'SFMono-Regular', 'Menlo', 'Consolas', 'DejaVu Sans Mono', monospace; --t72-font-mono: var(--t72-font-label); /* the name is right again */ --t72-track-display: .005em; /* inscriptional faces want POSITIVE */ --t72-track-mono: .2em; --t72-track-mono-tight: .16em; /* ---- Spacing + shape --------------------------------------------------- */ /* THE SHEET AND ITS GUTTERS, IN ONE PLACE. (Phase 18) Measured off the live reference: `.page { max-width: 1440px }` on a dark body, and every band `padding: 56px`, so the measure is 1440-112=1328. At 1280px of viewport (1265 once the scrollbar is off) that resolves to 1153, which is what the reference computes — and now what we compute too. --t72-page-max is the SHEET; --t72-page-measure is what is left of it inside the gutters. Change the page width here and the sheet, every band's gutter and every component's measure follow. */ --t72-page-max: 1440px; --t72-page-measure: 1328px; /* 1440 - two 56px gutters */ --t72-gutter: 56px; --t72-gutter-mobile: 22px; --t72-radius: 0; /* the system is square-cornered */ --t72-shadow-soft: 0 6px 28px rgba(61,63,63,.10); --t72-shadow-deep: 0 12px 48px rgba(61,63,63,.18); --t72-shadow-drop: 0 28px 64px -24px rgba(61,63,63,.35); } /* THE MOBILE GUTTER IS A REDEFINITION OF THE SAME TOKEN, not a second rule per band. The reference drops every band from 56px to 22px at ONE breakpoint, 720px — `.utility, .nav, .hero, .trust-band, … { padding-left: 22px; padding-right: 22px }` — so the token moves and everything that reads it moves with it: the shared bands, the brand hub's `.band` family, and the Oregon/Nevada `.container`. One gutter, one breakpoint, and no band left holding a stale value. --t72-gutter-mobile stays because the hub and the component sheet still name it in places, and the two must agree. */ @media (max-width: 720px) { :root { --t72-gutter: var(--t72-gutter-mobile); } } /* =========================================================================== THE PAGE-FAMILY HEADING SCALE (Phase 25a) =========================================================================== STOP ASSIGNING ONE HERO SIZE GLOBALLY. `.t72c-h1` used to be a single clamp(40px, 4.6vw, 62px) for every page that renders the shared hero, which put 58.9px on 41 of North Carolina's 49 routes. That is the HOMEPAGE size. It is not what the reference does and it is why a financing page shouted as loudly as the front door. EVERY NUMBER BELOW IS READ OFF TByrds_Launch, not guessed. The declarations were extracted from the reference's own page sources and then CONFIRMED by loading both trees in Chromium and reading getComputedStyle at four viewports. Reference

, measured: family 390px 1024px 1280px 1440px reference declaration home 42.0 47.1 58.9 64.0 clamp(42px, 4.6vw, 64px) city-landing 48.0 55.3 69.1 77.8 clamp(48px, 5.4vw, 80px) equipment 34.0 51.2 64.0 68.0 clamp(44px, 5vw, 68px) seasonal 32.0 47.1 58.9 60.0 clamp(40px, 4.6vw, 60px) membership 38.0 45.1 56.3 60.0 clamp(38px, 4.4vw, 60px) schedule 36.0 46.1 57.6 60.0 clamp(36px, 4.5vw, 60px) repair-metro 31.2 43.0 53.8 58.0 clamp(36px, 4.2vw, 58px) careers 36.0 45.1 56.0 56.0 clamp(36px, 4.4vw, 56px) sitemap 40.0 45.1 56.0 56.0 clamp(40px, 4.4vw, 56px) service-index 30.0 38.9 48.6 52.0 clamp(34px, 3.8vw, 52px) faqs 31.2 36.9 46.1 48.0 clamp(32px, 3.6vw, 48px) suburb-landing 28.0 36.9 46.1 48.0 clamp(32px, 3.6vw, 48px) repair-suburb 31.2 36.9 46.1 48.0 clamp(32px, 3.6vw, 48px) about 27.3 34.8 43.5 44.0 clamp(30px, 3.4vw, 44px) contact 31.2 34.8 43.5 44.0 clamp(30px, 3.4vw, 44px) service-areas 27.3 34.8 43.5 44.0 clamp(30px, 3.4vw, 44px) legal 31.2 34.8 43.5 44.0 clamp(32px, 3.4vw, 44px) program 27.3 32.8 41.0 42.0 clamp(28px, 3.2vw, 42px) Bethany's three cited figures are correct to the tenth: About 43.5px, financing 41.0px, the services index 48.6px, all at 1280px. The 390px column is a SECOND clamp the reference sets under its own 720px breakpoint — the mobile heads are not just the desktop clamp's minimum. It is reproduced below. The family comes from `pages//pages.json` and is stamped onto by build.py's ensure_family_class(). A page with no family keeps the old default. PHASE 25B: brand, or and nv now declare one too, so all 258 pages of all five builds take a family scale and NO page is left on the 62px global default. The default below is therefore dead code that exists only so a newly added route renders while its family is being decided — and audit check 21 fails the build if one is ever left that way. */ body.t72-fam-home { --t72c-h1-size: clamp(42px, 4.6vw, 64px); } body.t72-fam-city-landing { --t72c-h1-size: clamp(48px, 5.4vw, 80px); } body.t72-fam-equipment { --t72c-h1-size: clamp(44px, 5vw, 68px); } body.t72-fam-seasonal { --t72c-h1-size: clamp(40px, 4.6vw, 60px); } body.t72-fam-membership { --t72c-h1-size: clamp(38px, 4.4vw, 60px); } body.t72-fam-schedule { --t72c-h1-size: clamp(36px, 4.5vw, 60px); } body.t72-fam-repair-metro { --t72c-h1-size: clamp(36px, 4.2vw, 58px); } body.t72-fam-careers { --t72c-h1-size: clamp(36px, 4.4vw, 56px); } body.t72-fam-sitemap { --t72c-h1-size: clamp(40px, 4.4vw, 56px); } body.t72-fam-service-index { --t72c-h1-size: clamp(34px, 3.8vw, 52px); } body.t72-fam-faqs { --t72c-h1-size: clamp(32px, 3.6vw, 48px); } body.t72-fam-suburb-landing { --t72c-h1-size: clamp(32px, 3.6vw, 48px); } body.t72-fam-repair-suburb { --t72c-h1-size: clamp(32px, 3.6vw, 48px); } body.t72-fam-legal { --t72c-h1-size: clamp(32px, 3.4vw, 44px); } body.t72-fam-about { --t72c-h1-size: clamp(30px, 3.4vw, 44px); } body.t72-fam-contact { --t72c-h1-size: clamp(30px, 3.4vw, 44px); } body.t72-fam-service-areas { --t72c-h1-size: clamp(30px, 3.4vw, 44px); } body.t72-fam-program { --t72c-h1-size: clamp(28px, 3.2vw, 42px); } /* The 404. Its own family because it IS its own reference page: TByrds_Launch ships `404.html` at 51.2px at 1280, between the sitemap and the legal pages, and North Carolina's `.error-hero h1` already spells exactly that clamp inline. Oregon's and Nevada's 404s were taking the legal family's 43.5px only because nothing had given them a family of their own. */ body.t72-fam-error { --t72c-h1-size: clamp(34px, 4vw, 52px); } /* --------------------------------------------------------------------------- OREGON AND NEVADA'S OWN FAMILIES (Phase 25b) --------------------------------------------------------------------------- Six kinds of page North Carolina does not have. NONE of them gets a size of its own invention: each takes the scale of the nearest reference family, paired by ROLE and recorded in tools/assign_families.py. ours nearest reference page its 1280px

neighborhood hvac-kannapolis-nc 46.1 (suburb-landing) manual-j faqs 46.1 (faqs) diagnostic-first about-charlotte 43.5 (about) reviews about-charlotte 43.5 (about) rebates savings-charlotte 41.0 (program) pricing savings-charlotte 41.0 (program) `neighborhood` deserves the note. Oregon's neighbourhood leaf and its service-area leaf are the SAME DOCUMENT — the same components in the same order, the same "HVAC service in ." heading — and the reference has no third geographic rung under its suburb landing. Giving the neighbourhood a smaller step would be inventing a size the reference does not have, so both leaves take the leaf's size and the difference between them stays in the copy, which is where it belongs. */ body.t72-fam-neighborhood { --t72c-h1-size: clamp(32px, 3.6vw, 48px); } body.t72-fam-manual-j { --t72c-h1-size: clamp(32px, 3.6vw, 48px); } body.t72-fam-diagnostic-first { --t72c-h1-size: clamp(30px, 3.4vw, 44px); } body.t72-fam-reviews { --t72c-h1-size: clamp(30px, 3.4vw, 44px); } body.t72-fam-rebates { --t72c-h1-size: clamp(28px, 3.2vw, 42px); } body.t72-fam-pricing { --t72c-h1-size: clamp(28px, 3.2vw, 42px); } /* The reference's own mobile step, at its own 720px breakpoint. Families the reference does not step (home, city-landing, membership, careers, schedule, sitemap) are absent on purpose — their desktop clamp's minimum IS their mobile size, which is what the reference measures at 390px. */ @media (max-width: 720px) { body.t72-fam-equipment { --t72c-h1-size: clamp(34px, 8vw, 44px); } body.t72-fam-seasonal { --t72c-h1-size: clamp(32px, 8vw, 42px); } body.t72-fam-repair-metro { --t72c-h1-size: clamp(30px, 8vw, 42px); } body.t72-fam-repair-suburb { --t72c-h1-size: clamp(30px, 8vw, 42px); } body.t72-fam-service-index { --t72c-h1-size: clamp(30px, 7.5vw, 38px); } body.t72-fam-faqs { --t72c-h1-size: clamp(26px, 8vw, 34px); } body.t72-fam-suburb-landing { --t72c-h1-size: clamp(28px, 7vw, 38px); } body.t72-fam-legal { --t72c-h1-size: clamp(26px, 8vw, 34px); } body.t72-fam-contact { --t72c-h1-size: clamp(26px, 8vw, 34px); } body.t72-fam-about { --t72c-h1-size: clamp(26px, 7vw, 34px); } body.t72-fam-service-areas { --t72c-h1-size: clamp(24px, 7vw, 32px); } body.t72-fam-program { --t72c-h1-size: clamp(24px, 7vw, 32px); } /* Phase 25b — each one steps exactly as the family it was paired with. */ body.t72-fam-neighborhood { --t72c-h1-size: clamp(28px, 7vw, 38px); } body.t72-fam-manual-j { --t72c-h1-size: clamp(26px, 8vw, 34px); } body.t72-fam-diagnostic-first { --t72c-h1-size: clamp(26px, 7vw, 34px); } body.t72-fam-reviews { --t72c-h1-size: clamp(26px, 7vw, 34px); } body.t72-fam-rebates { --t72c-h1-size: clamp(24px, 7vw, 32px); } body.t72-fam-pricing { --t72c-h1-size: clamp(24px, 7vw, 32px); } }