/* =============================================================
 * SUPERFLY — HOME SECTION 2 — THE CATEGORY GRID — looks
 * -------------------------------------------------------------
 * Built 2026-11-10 on the operator's instruction, to the second
 * section of sohub.digital. That section is measured value by
 * value in docs/research/2026-11-10-sohub-work-section.md; read it
 * before re-pitching anything here, because most of the numbers
 * below are theirs re-fitted to a dark page rather than guesses.
 *
 * Markup:   assets/html/category-grid.html (a readable copy — the
 *           live markup is in page 42's Elementor data)
 * Movement: assets/js/category-grid.js
 *
 * NAMESPACE. Everything is scoped under .sfcg. The prefixes already
 * taken in this folder are sf-acc, sf-bike(s), sf-cart, sf-cat(s),
 * sf-checkout, sf-dealers, sf-drawer, sf-filters, sf-hero, sf-intro,
 * sf-lights, sf-live, sf-livery, sf-login, sf-maskword, sf-mega,
 * sf-navbar, sf-net, sf-pdp, sf-pf, sf-sequence, sf-shop, sf-social,
 * sf-sort, sf-step, plus .sfc, .sfcl, .sfidx, .sfop, .sfrev, .sfsr
 * and .sfv — grep assets/css before adding another.
 *
 * BOX SIZING. Elementor's frontend CSS puts box-sizing:border-box on
 * `.elementor *` — every DESCENDANT, never the root. A section that
 * caps its own root therefore measures 80px narrower inside Elementor
 * than outside it (CLAUDE.md has the full story). This file states it
 * on the root ITSELF, so the two environments agree and no page-id
 * override is ever needed. Do not remove the first rule.
 *
 * COLOURS come from theme.json through var(--wp--preset--color--*).
 * accent-1 is #ec1e79 on the live theme and that is the brand magenta
 * (CLAUDE.md → Conventions). The fallbacks exist only so the local
 * preview renders before WordPress prints its variables; never
 * hardcode the hex where a theme variable would do.
 * ============================================================= */

.sfcg,
.sfcg *{ box-sizing: border-box }

.sfcg{
	--sfcg-accent: var(--wp--preset--color--accent-1, #ec1e79);
	--sfcg-bg:     var(--wp--preset--color--base, #171717);
	--sfcg-ink:    var(--wp--preset--color--contrast, #f2f2f4);
	--sfcg-dim:    rgba(242,242,244,.62);
	--sfcg-rule:   rgba(242,242,244,.14);

	/* INTER EVERYWHERE IN THIS SECTION — operator, 2026-11-10, an hour
	   after the headline alone went to Inter: the reference sets its
	   whole band in one family and this now does too. All three tokens
	   point at the same stack; they are kept as three names because
	   the rest of the file reads them by role, and because a future
	   band may want the house faces back in one of the roles without
	   touching the others.

	   THIS SECTION IS THE ONLY INTER ON THE SITE. Every other band
	   stays on Archivo Black / Archivo / Chakra Petch, so the seam is
	   between SECTIONS, not inside this one. If Inter spreads further
	   that is a decision for the whole page, not a copy of this line.

	   WEIGHT CARRIES WHAT THE FACE NO LONGER DOES. Archivo Black is a
	   single very heavy weight, so the card names were set at 400 and
	   still shouted; Inter at 400 in the same place reads as a caption.
	   Each rule below states its own weight and says so. */
	--sfcg-display: "Inter", system-ui, -apple-system, sans-serif;
	--sfcg-body:    "Inter", system-ui, -apple-system, sans-serif;
	--sfcg-tech:    "Inter", system-ui, -apple-system, sans-serif;

	/* INTER, FOR THE STATEMENT LINE ONLY — operator, 2026-11-10, after
	   spotting it in the reference's inspector: "change to another font
	   like sohub, i think that sohub is inter font". It is: their
	   paragraph computes `Inter, "Inter Fallback"` at weight 600.

	   THIS IS THE ONLY PLACE ON THE SITE THAT USES INTER. The card
	   names below, the eyebrow and every other band stay on Archivo
	   Black / Archivo / Chakra Petch — so the family is deliberately
	   scoped to this token rather than added to --sfcg-display, which
	   the names also read. If Inter is ever wanted more widely it
	   needs a decision, not a second copy of this line.

	   It is loaded on its OWN handle, superfly-inter, in
	   functions.php. Adding a family to superfly-hero-fonts does
	   nothing — that handle is registered 43 times and the first
	   registration wins (CLAUDE.md).

	   SUPERSEDED THE SAME HOUR: the whole section went to Inter, so
	   this token is no longer the only one that holds it. It stays
	   because the title is the one place whose weight and ramp are
	   copied from the reference, and keeping its family named
	   separately means re-pitching the title cannot silently move the
	   card names with it. */
	--sfcg-head:    "Inter", system-ui, -apple-system, sans-serif;

	/* FULL BLEED, GUTTER ONLY — no centred measure. It shipped capped
	   at 1560px and the operator rejected it on sight (2026-11-10):
	   on a 1900px screen the cap left 170px of dead margin on each
	   side ON TOP of the gutter, so the band read as a narrow column
	   in a wide page while every other section ran to the edge. The
	   reference does the same thing — sohub.digital's section is
	   `w-screen` with `xl:px-20` and nothing else.

	   The cost is that the cards grow without limit on a very wide
	   monitor, which is the reference's behaviour too. If that ever
	   needs reining in, cap the GRID rather than the wrap, or the
	   heading stops sharing the cards' left edge. */
	--sfcg-wrap: none;

	/* THE SIDE GUTTER IS THE NAVBAR LOGO'S LEFT EDGE, to the nearest
	   fluid value — the same literal category-index.css carries, for
	   the same reason and with the same warning. The floating pill
	   stops 24px short of the screen and holds up to 20px of padding
	   inside its glass, so the logo sits at 44px on any desktop width.
	   3.1vw caps at 44 from about 1420px up and falls away below.

	   NOT a var() read from navbar.css. A custom property substitutes
	   at computed-value time on the element where it is DECLARED, so
	   a cross-file :root read is the trap CLAUDE.md records on the
	   navbar. Two literals that say they belong together beat one
	   that fails silently. If --sf-navbar-gutter or
	   --sf-navbar-pad-x ever move, re-measure the logo and re-pitch
	   BOTH files. */
	/* 80px AT THE TOP END, WHICH IS THE REFERENCE'S, NOT THIS SITE'S.
	   The comment above describes the 44px this shipped with and why
	   44 is the house number. The operator looked at the two side by
	   side on 2026-11-10 and asked for sohub's measure — theirs is
	   `xl:px-20`, a flat 80, and at 44 the cards sat closer to the
	   screen edge than anything else on the page.

	   SO THIS SECTION NO LONGER STARTS ON THE NAVBAR LOGO'S LINE, and
	   that was a deliberate trade, not an oversight: 80 against the
	   logo's 44. Every other section that cares still uses 44 — do not
	   "align" them by changing one of them without asking.

	   One happy side effect: --sfcg-rail-clear below is a max()
	   against this gutter, so at 80 it computes to 0 and the heading
	   needs no indent at all on a desktop. It still earns its keep on
	   a phone, where the gutter is 20. */
	--sfcg-gutter: var(--sf-gutter);

	/* THE FIXED SOCIAL RAIL SITS ON THE HEADLINE, and at the gutter
	   above it is a 2px overlap that looks like a clipped letter.
	   .sfsr is pinned to the middle of the left edge of every page at
	   z-index 90 and is 46px wide (40 below 782px) — measured on the
	   live page, rail [0 … 46] against the title's second and third
	   lines starting at x=44, with the chips plainly over the A of AND
	   and the R of RACING.

	   58px is the house clearance: .sfidx uses the same number for its
	   row numbers and .sfop's desktop title lands at 56 for the same
	   reason. A max() against this section's own gutter rather than a
	   media query, so it is 14px at 1440, ~26 at 1024 and 38 on a
	   phone, and would disappear on its own if the gutter ever grew
	   past 58.

	   IT GOES ON THE HEAD ONLY. The cards keep the 44px gutter: a
	   photograph passing under the rail is what the rail does on every
	   page, and their labels are already inside the card's own padding
	   at 68. Indenting the whole wrap would take the cards off the
	   navbar logo's line, which is the one thing the gutter above is
	   for. */
	--sfcg-rail-clear: max(0px, calc(58px - var(--sfcg-gutter)));

	/* The card corner. 12px EVERYWHERE, which is not this section's
	   decision to make: the navbar pill and the hero video panel were
	   standardised on 12px on 2026-11-10 and a third radius 24px
	   below them would read as a mistake. --sf-navbar-bar-radius and
	   --sfop-radius are the other two. Change all three or none. */
	--sfcg-radius: 12px;

	/* THE NEUMORPHIC EMBOSS — operator, 2026-11-12: "i want the image
	   card is using neumorphism like card in section 5". These are
	   .sfdw's values exactly: light edge up-left, dark edge
	   down-right, no border, so the light IS the edge. Section 5
	   takes them in turn from the footer's carved SUPERFLY wordmark
	   (dealer-wall.css says so at length — read it before re-pitching
	   either).

	   TWO LITERALS RATHER THAN ONE SHARED TOKEN, the same decision
	   this file already records for --sfcg-radius against
	   --sf-navbar-bar-radius: a custom property substitutes at
	   computed-value time on the element where it is DECLARED, so a
	   :root token declared in dealer-wall.css and read from here
	   would depend on load order and fail silently. **Change both
	   files or neither.**

	   WHAT IS DIFFERENT HERE, AND IT IS THE WHOLE REASON THIS IS ONLY
	   HALF A PORT: section 5's card has no content, so its FILL does
	   the carving — #171717 on #171717, a flat panel pressed out of
	   the weave. This card is a photograph bleeding to all four
	   edges. There is no face to colour, so only the two shadows come
	   across and the picture is what sits inside them. Do not copy
	   --sfdw-card over as well; it would paint under the image and
	   show nothing.

	   DEPLOYED, THEN LOST, THEN REDEPLOYED on 2026-11-12: a later
	   save from the box carried the gap change on top of a copy that
	   predated this, and took the emboss with it. See CLAUDE.md. */
	--sfcg-lift: rgba(255,255,255,.10);
	--sfcg-sink: rgba(0,0,0,.70);

	/* The arrow box INCLUDES the gap to the word (padding-right), so
	   one length both sizes the arrow and sets how far the name is
	   pushed. See "the label" below — the whole hover is built on
	   this being a single number. */
	--sfcg-arrow-w:   clamp(26px, 2.5vw, 38px);
	--sfcg-arrow-gap: clamp(8px, .9vw, 14px);
	--sfcg-shift:     calc(var(--sfcg-arrow-w) + var(--sfcg-arrow-gap));

	/* Per-character stagger. 14ms, not the reference's 15 — "Suspensions"
	   is 11 characters against their longest at 7, and the same trade is
	   already recorded on the navbar's text roll. At 14ms the last
	   letter starts 140ms after the first, which is still inside the
	   0.4s move. */
	--sfcg-step: 14ms;
	--sfcg-move: .4s;

	/* The one easing. A long tail is what makes the card feel like it
	   settles rather than stops. The lift uses the house spring
	   instead — see .sfcg__card. */
	--sfcg-ease: cubic-bezier(.16, 1, .3, 1);

	position: relative;
	z-index: 1;

	/* CUT BACK ON THE OPERATOR'S CALL, 2026-11-10 — "why that section
	   too many padding". It shipped at 120/140 and measured 100 above
	   the eyebrow at 1440, which is more air than anything else on the
	   page opens with. 72/88 now. The reference is 96 above and 128
	   below, but theirs sits on a white page where space reads as
	   lighter; on carbon the same gap reads as a gap. */
	/* THE REFERENCE'S OWN, at last: 96px above and 128px below, which
	   is their `mt-24` and `pb-32`. Asked for on 2026-11-10 ("added a
	   padding for that section"), the shorter cards having left the
	   band sitting tight against its neighbours.

	   It has been 120/140 → 72/88 → 96/128. The middle step was a
	   reaction to the 1560px wrap, which left 170px of dead margin at
	   the sides and made the whole band read as over-padded; with the
	   wrap gone and the cards back at the reference's proportion, the
	   reference's spacing is right too. */
	/* The `+ 3px` is literal — operator, 2026-11-10, "increase another
	   3px more", on top of the reference's own values. It is written
	   as a calc rather than folded into the clamps so that the
	   reference's numbers stay readable and the deliberate 3px over
	   them stays visible; fold it in only if the clamps are re-pitched
	   from scratch. */
	padding: calc(clamp(56px, 5.6vw, 96px) + 3px) 0
	         calc(clamp(72px, 7.5vw, 128px) + 3px);

	color: var(--sfcg-ink);
	font-family: var(--sfcg-body);

	/* THE HOUSE CARBON WEAVE. Every band on the home page sits on ONE
	   surface or the seam between two of them shows — settled by the
	   operator 2026-10-19 and re-based to #171717 on 2026-11-10. Both
	   diagonals are needed: the light one at 45deg is the sheen on the
	   tow, the dark one at -45deg the shadow under it, and with one
	   alone it reads as corduroy. Copy BOTH halves into any new band. */
	background-color: var(--sfcg-bg);
	background-image:
		repeating-linear-gradient( 45deg, rgba(255,255,255,.025) 0 2px, transparent 2px 4px),
		repeating-linear-gradient(-45deg, rgba(0,0,0,.35)        0 2px, transparent 2px 4px);
}

.sfcg__wrap{
	max-width: var(--sfcg-wrap);
	margin-inline: auto;
	padding-inline: var(--sfcg-gutter);
}

/* ── the heading ───────────────────────────────────────────── */

.sfcg__head{
	/* 64px is the reference's `mt-16`, asked for by name on
	   2026-11-10 after the first cut took it to 48 and the headline
	   sat on top of the cards. */
	margin-bottom: clamp(32px, 3.6vw, 64px);
	padding-left: var(--sfcg-rail-clear);
}

.sfcg__eyebrow{
	margin: 0 0 clamp(14px, 1.6vw, 24px);
	font-family: var(--sfcg-tech);
	font-size: clamp(11px, .95vw, 13px);
	font-weight: 700;
	letter-spacing: .16em;
	text-transform: uppercase;
	color: var(--sfcg-dim);

	opacity: var(--sfcg-in, 1);
	transform: translateY(calc((1 - var(--sfcg-in, 1)) * 18px));
	transition:
		opacity   .7s var(--sfcg-ease),
		transform .7s var(--sfcg-ease);
}

/* ─────────────────────────────────────────────────────────────────
   THE STATEMENT LINE — SET TO SOHUB.DIGITAL'S NUMBERS EXACTLY.

   Operator, 2026-11-10: "i want the title size font and full width
   same like in sohub section ... also you need to follow letter
   spacing and line spacing, i want exactly same".

   So the four values below are not pitched, they are COPIED, and
   they were read off their live page rather than off their class
   names — Tailwind's `2xl:` in particular is not where the default
   config puts it (see the ladder). Measured on
   https://sohub.digital/, the `.home-projects` paragraph:

       viewport   font-size   line-height   letter-spacing
       291        30          30            -2.25
       768        48          48            -3.60
       1024       60          60            -4.50
       1280       72          72            -5.40
       1440       72          72            -5.40
       1536       72          72            -5.40
       1850       72          72            -5.40
       1900       112         112           -8.40

   line-height always EQUALS font-size (Tailwind `leading-none`, so
   1), and their letter-spacing is always exactly -0.075em of it
   (`tracking-7`).

   THE TABLE IS NOW OURS AS WELL AS THEIRS — every value in it is
   what this file sets. It did not get here in one go: the size, the
   tracking and the leading were each reverted at some point during
   2026-11-10 and all three were put back the same day, after the
   operator compared the two headlines side by side and said "please
   make it same". The earlier reasoning is kept in the notes on each
   rule below, because it was sound for the face this title USED to be
   set in (Archivo Black) and would apply again if it ever changes
   back. Against Inter 600 — which is now the face, as it is theirs —
   none of it holds, and the values are simply theirs.

   THE ONLY DIFFERENCE LEFT IS COLOUR, and it is not a type decision:
   their page is near-black on near-white, this one is near-white on
   carbon. The sheen word inverts with it.

   THE 112 STEP IS NOT AT TAILWIND'S DEFAULT 1536. Measured: 1536,
   1600, 1700, 1792, 1800 and 1850 all give 72, and 1900 gives 112,
   so their `2xl` is somewhere in 1851–1900. 1900 is the width the
   change was actually observed at and is what this file uses; if it
   ever matters, bisect that bracket rather than assuming 1536.

   NO max-width. Theirs computes `max-width: none` and the paragraph
   fills the section — 1740px of a 1900px window. So nothing here
   controls the line count: the words break where the full width runs
   out, exactly as they do on the reference. Nothing downstream
   depends on a particular number of lines.
   ───────────────────────────────────────────────────────────────── */
.sfcg__title{
	margin: 0;

	/* FULL WIDTH, AS THEIRS IS. A 22ch cap lived here for part of
	   2026-11-10, to force the reference's three-line SHAPE — at full
	   width and the smaller size the sentence broke as two long lines
	   and an orphan ("here." alone). The cap went when the size and
	   the tracking were restored, because at 112px/-0.075em the words
	   fall differently; see the note above the rule and the line
	   shapes recorded in CLAUDE.md.

	   IF AN ORPHAN EVER COMES BACK, the cap is the tool — measure it
	   in `ch` so it holds the same character measure at every step of
	   the ladder, and put it on the TITLE, not on the section: padding
	   the section moves the cards, and full-bleed cards on the gutter
	   are their own instruction. */
	max-width: none;

	/* Inter — the reference's own face. Weight matters as much as the
	   family here: Inter at 400 beside Archivo Black card names reads
	   as a caption, not a heading.

	   **THE WEIGHT IS 700, NOT THE REFERENCE'S 600** (operator,
	   2026-11-12: "now i want you make section 1 , 2 and section 4
	   using weight 700"). It followed an audit of the whole page
	   which found section 3 alone at 700 and everything else at 600;
	   the answer was to bring the others UP rather than take that one
	   down. **Every "600" in the prose around this band is the older
	   state** — the band is one weight throughout, and that weight is
	   now 700: title, card names, eyebrow and the placeholder tag.

	   THE FOUR FILES THAT SHARE THESE LITERALS ARE IN STEP AGAIN —
	   home-opening.css, this file, bike-ribbon.css and
	   video-panel.css all carry 700. Change them together or not at
	   all; the note at `.sfop__title` says why they are literals. */
	font-family: var(--sfcg-head);
	font-weight: 700;

	font-size: 30px;

	/* `leading-none`. It was .95 for part of 2026-11-10, which was
	   right while this title was set in Archivo Black: that face's
	   descenders ride into the next line's caps and a multi-line block
	   interlocks rather than stacks. The face is Inter 600 now, as the
	   reference's is, so the reference's own value applies.

	   Line-height does not affect WRAPPING, so it never changes a line
	   COUNT — only the height of the block. */
	line-height: 1;

	/* UPPERCASE — operator, 2026-11-12: "section 1 title is upper
	   case , section 2 also upper case". This REVERSES the 2026-11-10
	   ruling recorded above and in CLAUDE.md ("for first section and
	   second section, i want the title is titlecase not uppercase"),
	   which is theirs to reverse — but the reasoning that went with it
	   still applies, so read it before touching the leading or the
	   measure:

	   CASE IS NOT COSMETIC ON A HEADLINE THAT WRAPS. Caps are about
	   21% wider than title case in Inter 600 — measured live on this
	   page, 4.992x the font size against 4.131x for the same word — so
	   the same sentence takes more lines at every width. The phone-
	   taking-seven-lines problem this file records under the old
	   uppercase version is the thing to watch if the sentence is ever
	   lengthened again.

	   IT IS A TRANSFORM RATHER THAN RE-CASED MARKUP, unlike .sfbw and
	   .sfvp where the case carries meaning (a make against a model
	   badge). The sentence stays readable in the widget, the screen
	   reader keeps its natural text, and putting it back is this one
	   declaration — no database write.

	   CAPS HAVE NO DESCENDERS, so line-height: 1 has more air in it
	   than it looks. Do not close it to match the old mixed-case block
	   without measuring; the reference value is the value. */
	text-transform: uppercase;

	/* `tracking-7`, -0.075em, theirs exactly. Also off for part of the
	   same day, for the same reason as the leading: -0.075em closes
	   Archivo Black's counters up. On Inter 600 at these sizes it is
	   the gesture the reference is built on, and the whole look of the
	   headline depends on it. */
	letter-spacing: -.075em;

	opacity: var(--sfcg-in, 1);
	transform: translateY(calc((1 - var(--sfcg-in, 1)) * 26px));
	transition:
		opacity   .8s var(--sfcg-ease) .08s,
		transform .8s var(--sfcg-ease) .08s;
}

/* THE GRADIENT WORD. This is NOT a scroll effect and NOT a second
   colour — it is one span with a gradient clipped to the text,
   sweeping for ever. Three things in it are load-bearing, all of them
   taken from sohub.digital's .h-word-light and re-pitched for carbon:
   the gradient RETURNS TO THE PAGE INK at both ends, so the word never
   reads as a different colour, only as a sheen passing through it;
   the background is oversized (220%) and the sweep moves on BOTH axes,
   which is what stops it looking like a straight left-to-right wipe;
   and the negative-margin / positive-padding pair gives the last glyph
   room, because at -0.03em the background box clips the side bearing
   and cuts the end of the word off.

   background-clip:text MUST be written as both longhands. Blink does
   not clip from the shorthand alone — the gradient then paints as a
   solid magenta band with invisible type inside it. reviews.css
   records the same failure on its headline. */
.sfcg__word{
	display: inline;
	margin-right: -.12em;
	padding-right: .12em;

	color: transparent;
	background-image: linear-gradient(90deg,
		var(--sfcg-ink)    0%,
		var(--sfcg-accent) 48%,
		var(--sfcg-ink)   100%);
	background-size: 220% 220%;
	-webkit-background-clip: text;
	background-clip: text;
	animation: sfcgSheen 5s ease infinite;
}

@keyframes sfcgSheen{
	0%   { background-position: 10% 0 }
	50%  { background-position: 91% 100% }
	100% { background-position: 10% 0 }
}

/* The size ladder, stepped rather than fluid, because the reference is
   stepped. line-height and letter-spacing follow from the size above
   and are not repeated.

   THE REFERENCE'S VALUES EXACTLY — 30 / 48 / 60 / 72 / 112. They ran
   about 11% under for part of 2026-11-10 ("reduce a bit size font",
   so 28 / 44 / 56 / 64 / 96) and were put back the same day when the
   operator compared the two headlines and asked for them to match.

   A SIZE CHANGE RESHAPES THE HEADLINE — it is not a free dial, because
   it moves where the words break. Swept at the smaller ladder: at 1900
   three lines held 84–112 and 80 tipped to two; at 1440 they held
   64–72 and 60 tipped to two; at 375, 30 and 28 both gave five and 26
   gave four. Re-sweep before touching any of these, and read the line
   shapes at the current values in CLAUDE.md. */
@media (min-width:  768px){ .sfcg__title{ font-size:  48px } }
@media (min-width: 1024px){ .sfcg__title{ font-size:  60px } }
@media (min-width: 1280px){ .sfcg__title{ font-size:  72px } }
@media (min-width: 1900px){ .sfcg__title{ font-size: 112px } }

/* THE HEADING IS HANDED TO GSAP, and this is the handover. The script
   adds .is-gsap once it has SplitText, ScrollTrigger and a heading to
   split; from that moment opacity and y belong to the tween and the
   CSS reveal above must get out of the way, or a scrub fights a
   0.8s transition and arrives a third of a second late on every
   frame. The script also pins --sfcg-in open so the declarations
   above compute to the finished state.

   Without .is-gsap — no GSAP, or prefers-reduced-motion — none of
   this applies and the heading keeps the plain CSS reveal. */
.sfcg.is-gsap .sfcg__eyebrow,
.sfcg.is-gsap .sfcg__title,
.sfcg.is-gsap .sfcg__line{
	transition: none;
}

/* SplitText writes `display:block` and its own position on each line;
   this only adds the hint, and the overflow guard that keeps a
   descender from being clipped while a line sits 4rem low. */
.sfcg__line{
	will-change: transform, opacity;
}

/* ── the grid ──────────────────────────────────────────────── */

.sfcg__grid{
	display: grid;
	grid-template-columns: repeat(2, minmax(0, 1fr));
	/* THE MIDDLE TERM IS THE ONE THAT MOVES THE BAND, 2.2vw -> 4.2vw
	   (2026-11-12, operator: "why the gap still not change ? i want
	   added more gap").

	   The ceiling went 40 -> 60 -> 80 over two messages and the grid
	   did not move a pixel, because a clamp returns its MIDDLE term
	   until that term passes the ceiling — and 2.2vw only reaches 80px
	   at 3636px of viewport. At 4.2vw the ceiling binds from about
	   1905px, which is what asking for 80 meant.

	   SO THIS IS NO LONGER THE REFERENCE'S VALUE. sohub is `gap-6` /
	   `lg:gap-10` — 24 and 40 — which is exactly what the first two
	   numbers were, asked for by name on 2026-11-10 ("have gap") after
	   the square cards sat too close together. The floor is still
	   theirs; the ramp is ours.

	   EACH CARD PAYS HALF OF ANY INCREASE, because the two tracks
	   split what the gap does not use. Measured live at 1368: gap 30.1
	   -> 57.9 and the cards 612 -> 597 wide. It is also the ROW gap,
	   so the band grows by the same amount between its two rows. */
	/* CEILING 80 -> 60 (2026-11-12, operator: "reduce a gap maybe
	   60px"), an hour after 4.2vw finally made 80 reachable. It now
	   binds from about 1430px, so a desktop is a flat 60 and only the
	   narrow end still ramps: 60 at 1440 unchanged, 79.8 -> 60 at 1900,
	   24 on a phone. */
	/* 4.2vw/60 -> 3vw/44 (2026-11-12, operator: "reduceeeeeeee ,oreee
	   thee gap , cause the gap is too much even already change 60px").

	   BOTH TERMS, because the ceiling alone is what wasted three
	   deploys earlier the same day — see the note above. 3vw reaches
	   the 44 ceiling at about 1470px, so the ramp still does the work
	   on everything narrower and a desktop sits at a flat 44.

	   Where it has been at 1440: 31.7 (2.2vw, which read as "not
	   changing") -> 60 ("too much") -> 42.8 -> 50 ("i think 50px
	   should be okayy"). 3.5vw reaches the 50 ceiling at about 1429px,
	   so a desktop is a flat 50. The floor is still
	   sohub's 24. */
/* 	gap: clamp(24px, 3.5vw, 50px); */
	   gap: clamp(24px, 2vw, 35px);
}

/* ── the card ──────────────────────────────────────────────── */

.sfcg__card{
	position: relative;
	display: block;

	/* 16:9, THE REFERENCE'S OWN `aspect-video`. It has been 16:9 → 3:2
	   → 1:1 → 16:9 in one afternoon, and the round trip is worth
	   understanding before anyone changes it again.

	   `object-fit: cover` scales to the card's WIDTH, so the card's
	   shape alone fixes how much of a photograph you see:

	       card    square photo   the portrait wheel
	       16:9    56%            40%   <- current
	       3:2     67%            48%
	       4:3     75%            54%
	       1:1     100%           71%

	   THE WHEEL IS CROPPED AT THIS SHAPE AND THERE IS NO CSS FIX. It
	   is the only portrait file (1142x1600) and its subject spans 66%
	   of the image height, so at 16:9 it cannot be whole: 40% is all
	   there is. 1:1 was tried for exactly that reason and the operator
	   rejected the height it cost ("no, should be the size card is
	   only like this"), choosing the reference's proportion over a
	   whole rim.

	   **The fix, if the crop is ever raised again, is the FILE** — cut
	   cat-wheels.webp to 16:9 around the rim and the subject fills the
	   card at this shape with nothing lost. Do not reach for a taller
	   card or for `contain`; both have been tried and both cost more
	   than they are worth here.

	   The per-card --sfcg-pos below is derived from THIS ratio. Change
	   one and re-derive the other. */
	aspect-ratio: 2 / 1;
	overflow: hidden;
	border-radius: var(--sfcg-radius);
	text-decoration: none;
	color: var(--sfcg-ink);
	background: rgba(0,0,0,.45);

	/* The reveal. 0 closed, 1 open; the script's whole job is to set
	   it to 1 and everything here reads it. One source of truth for
	   what "open" looks like, and no tween — the same arrangement
	   category-index.css uses. */
	--sfcg-in: 1;
	opacity: var(--sfcg-in);
	transform: translateY(calc((1 - var(--sfcg-in)) * 34px)) scale(calc(.985 + var(--sfcg-in) * .015));

	transition:
		opacity   .8s var(--sfcg-ease),
		transform .45s cubic-bezier(.22,.68,0,1.5);
}

/* THE EMBOSS NEEDS TWO CLASSES, AND THE LOCAL HARNESS CANNOT SHOW
   YOU WHY. Elementor's frontend.min.css carries `.elementor a{
   box-shadow: none }` — specificity 0,1,1, and this card is an <a>,
   so a plain `.sfcg__card` at 0,1,0 LOSES. The rule is present, the
   file is correct, the computed value is `none`.

   This file already records the identical trap one layer down, where
   `.elementor img{ height:auto }` at 0,1,1 beat `.sfcg__img` and
   silently undid every object-fit crop. Same cause, same fix, same
   blind spot: docs/home-concepts loads no Elementor CSS, so both
   looked perfect locally and were wrong live. **Measure the computed
   value on the live page, not in the harness.**

   Raised, exactly as .sfdw__card has it. NO BORDER anywhere on this
   card and none is wanted: a neumorphic card's shape is made only by
   the light falling on it, and a hairline flattens the whole thing
   into a rectangle with a glow.

   BOTH HALVES OR NEITHER. The lift is what stops it reading as a flat
   rectangle; the sink is what lifts it off the weave. Cutting one
   alone breaks it — return-refund.css and dealer-wall.css both say
   the same in the same words.

   IT DOES NOT CHANGE ON HOVER, which is section 5's rule and the
   footer's before it. The 2% lift below scales the shadow with the
   card for free, because a transform scales what it paints;
   deepening the shadow as well would be two effects where there is
   one. */
.sfcg .sfcg__card{
	box-shadow:
		-4px -4px  8px var(--sfcg-lift),
		 5px  5px 12px var(--sfcg-sink);
}

/* The staggered arrival. Four cards, so four delays written out
   rather than calculated — shorter than the calc and obvious at a
   glance. They apply to the OPENING only; the hover lift below
   overrides the transform transition and must not inherit a delay. */
.sfcg__card:nth-child(1){ transition-delay: 0s,    0s }
.sfcg__card:nth-child(2){ transition-delay: .08s,  0s }
.sfcg__card:nth-child(3){ transition-delay: .16s,  0s }
.sfcg__card:nth-child(4){ transition-delay: .24s,  0s }

/* The 2% lift, on the house spring — the overshoot is doing the work.
   Scoped to a real pointer: on a touch screen :hover sticks after a
   tap and the card stays lifted until something else is tapped. */
@media (hover: hover){
	.sfcg__card:hover,
	.sfcg__card:focus-visible{
		transform: translateY(0) scale(1.02);
	}
}

.sfcg__card:focus-visible{
	outline: 2px solid var(--sfcg-accent);
	outline-offset: 3px;
}

.sfcg__media{
	position: absolute;
	inset: 0;
	display: block;
	z-index: 0;
}

/* THE DARKENING OVERLAY (operator, 2026-11-11: "added overlay, cause
   i want the image is a bit dark"), added in the same pass as the four
   new photographs.

   IT IS NOT THE SAME THING AS .sfcg__veil. The veil is a gradient over
   the bottom 62% that exists so the card's name has something to sit
   on; this is flat across the whole picture and exists so the pictures
   sit back from the page. Both are needed — the veil alone leaves the
   top of a bright photograph as loud as it was, and this alone does
   not give the name enough contrast at the foot.

   Inside .sfcg__media, so it paints above the img and below the veil
   (z-index 1) and the label (2) without any z-index of its own.

   --sfcg-overlay is the dial. Raise it and check the two BRIGHT cards
   first — the caliper and the group shot — because the dark ones can
   take more than they can. */
.sfcg{
	--sfcg-overlay: .34;
}

.sfcg__media::after{
	content: "";
	position: absolute;
	inset: 0;
	background: rgba(0, 0, 0, var(--sfcg-overlay));
	pointer-events: none;
}

/* TWO CLASSES, NOT ONE, AND THAT IS THE WHOLE RULE'S POINT.
   Elementor's frontend CSS carries `.elementor img { height: auto;
   max-width: 100% }` — specificity 0,1,1, which BEATS a single class.
   With `height:auto` winning, the square photograph is laid out 888px
   tall inside a 592px card, `object-fit` has nothing left to do, and
   the card simply shows the TOP of the picture, clipped by the card's
   own overflow. Every aim below silently does nothing.

   It is invisible in the local harness, which loads no Elementor CSS,
   and invisible to the obvious check: the computed `object-position`
   still reads back exactly what you set. Measured on the live page
   2026-11-10 — img box 888x888 inside a media box of 888x592.

   Read the HEIGHT, not the object-position, when a crop looks wrong. */
.sfcg .sfcg__img{
	width: 100%;
	height: 100%;
	object-fit: cover;
	display: block;

	/* EVERY CARD AIMS ITS OWN CROP AGAIN, now that the card is 16:9
	   and no photograph fits whole. Each `--sfcg-pos` is inline in the
	   markup, because it is a fact about one picture, not about the
	   section.

	   The lit subject in each file was measured by sampling it on a
	   canvas and taking the bounding box of everything brighter than
	   luma 42 — the subject against the black — as a percentage of
	   image height. The aim puts the middle of that band in the middle
	   of the card's window:

	       file          subject    window   aim
	       wheels        23 → 89    40.1%    center 60%   cropped
	       suspensions   29 → 99    56.3%    center 83%   cropped
	       brakes        45 → 99    56.3%    center bottom  FITS

	   Only the brake pump fits its window; the other two are trimmed
	   at both ends, the wheel heavily. That is the price of the card's
	   shape and it is recorded on the card rule above.

	   THE AIM IS A FUNCTION OF THE SUBJECT **AND** THE CARD'S RATIO.
	   Re-run the sampling when a picture is replaced, and re-derive
	   every number here if the ratio changes:

	       visible    = cardRatio⁻¹ ÷ (imageH ÷ imageW)
	       windowTop  = pos × (1 − visible)
	       aim (pos)  = (subjectCentre − visible ÷ 2) ÷ (1 − visible) */
	object-position: var(--sfcg-pos, center);

	transform: scale(1);
	transition: transform .6s var(--sfcg-ease);
}

@media (hover: hover){
	.sfcg__card:hover .sfcg__img{ transform: scale(1.04) }
}

/* Legibility under the label — a plain overlay, not a filter. Half
   the card, black at .55 fading out, which is heavier than the
   reference's .4 because these photographs are darker than theirs
   and the type is white on both. */
.sfcg__veil{
	position: absolute;
	z-index: 1;
	left: 0; right: 0; bottom: 0;
	height: 62%;
	display: block;
	pointer-events: none;
	background: linear-gradient(to top,
		rgba(0,0,0,.72) 0%,
		rgba(0,0,0,.34) 46%,
		rgba(0,0,0,0)  100%);
}

/* ── the label, and the whole hover ────────────────────────── */

/* THE ARROW IS NEVER FADED. It is parked one box-width to the left
   and this row's overflow:hidden is the only thing hiding it — the
   same as the reference. Nothing in this effect touches opacity, and
   adding one would flatten it. */
.sfcg__label{
	position: absolute;
	z-index: 2;
	left: var(--sfcg-pad, clamp(16px, 1.7vw, 30px));
	right: var(--sfcg-pad, clamp(16px, 1.7vw, 30px));
	bottom: var(--sfcg-pad, clamp(16px, 1.7vw, 30px));

	display: flex;
	align-items: center;
	overflow: hidden;

	/* The descender guard, the same pair navbar.css carries on its
	   text roll: the padding opens the clipped box enough that it does
	   not shave the tails off letters sitting still, and the margin
	   takes the same amount back out of the layout. */
	padding-block: .18em;
	margin-block: -.18em;
}

.sfcg__arrow{
	flex: 0 0 auto;
	width: var(--sfcg-shift);
	padding-right: var(--sfcg-arrow-gap);
	color: var(--sfcg-ink);

	/* parked: exactly its own box to the left, so the gap travels with
	   it and the word can sit flush against the label's left edge */
	transform: translateX(-100%);
	transition: transform var(--sfcg-move) var(--sfcg-ease) calc(var(--sfcg-step) * 0);
}

.sfcg__arrow svg{
	display: block;
	width: 100%;
	height: auto;
}

.sfcg__name{
	font-family: var(--sfcg-display);

	/* 600 AND MIXED CASE — operator, 2026-11-10: "for title each card
	   also i want titlecase not uppercase". The names render as the
	   markup writes them (Wheels / Suspensions / Brakes / Others), so
	   capitalisation is editable without touching CSS.

	   THE WEIGHT MOVED WITH THE CASE, 700 → 600. 700 was chosen an hour
	   earlier for a reason the case change retired: this type was
	   uppercase and smaller than the heading, and caps at 600 over a
	   photograph read thin. The reference's card titles are 600 and are
	   mixed case, which is now exactly this situation — and 600 also
	   matches both big titles, so the whole band is one weight.

	   If a name ever has to hold against a brighter photograph, raise
	   THIS to 700 rather than reaching for the veil; the veil is
	   already heavier than the reference's.

	   THEY WERE UPPERCASE FOR AN HOUR on 2026-11-12, when both big
	   titles went up ("also uppercase"), and came back the same hour
	   ("make it the title for each card is titlecase not uppercase").
	   Nothing of it is left, and that is the point of doing it as a
	   transform: the words in the markup were always mixed case, so
	   the change was one declaration and the revert was deleting it —
	   no database write either way. THE TWO BIG TITLES ARE STILL
	   UPPERCASE; this band is deliberately mixed against them now. */
	font-weight: 700;
	/* DOWN A QUARTER (2026-11-12, operator: "reduce size font for title
	   each card"), in the same breath as the gap: 2.9vw/44 -> 2.2vw/34,
	   so 41.8 -> 31.7 at 1440, 44 -> 34 at 1900 and 24 -> 20 on a phone.

	   THE FLOOR CAME DOWN WITH THE RAMP. Leaving it at 24 would have
	   made a phone louder than a 1024px tablet, which is the shape of
	   bug the hero records at length — check the narrow end of a ramp,
	   not just the wide one.

	   The name is `white-space: nowrap` inside a card that clips, so
	   smaller only ever buys room: SUSPENSIONS had the least and it is
	   the one to measure if this ever goes back up. */
	font-size: clamp(20px, 2.2vw, 34px);
	line-height: 1.05;
	letter-spacing: -.02em;
	white-space: nowrap;
}


/* Before the script splits it — and for ever if the script never
   runs — the whole word moves as one block. The split version below
   replaces this with the per-character stagger. */
.sfcg__name:not(.is-split){
	display: inline-block;
	transform: translateX(calc(-1 * var(--sfcg-shift)));
	transition: transform var(--sfcg-move) var(--sfcg-ease);
}

/* EVERY CHARACTER IS ITS OWN INLINE-BLOCK, SPACES INCLUDED, and an
   inline-block holding only a space collapses to zero without
   white-space:pre — "Brake Systems" would render as "BrakeSystems".
   navbar.css records the same trap on its text roll. */
.sfcg__char{
	display: inline-block;
	white-space: pre;
	transform: translateX(calc(-1 * var(--sfcg-shift)));

	/* THE RESTING STATE CARRIES THE LEAVING DELAYS and the hover rule
	   carries the entering ones, which is how one transition plays
	   differently in each direction with no JavaScript at all. On the
	   way out the arrow goes first and the word follows; on the way in
	   the word is pushed first and the arrow lands in the gap it has
	   opened. --sfcg-i is the character's index from the END of the
	   word, written on by the script. */
	transition: transform var(--sfcg-move) var(--sfcg-ease) calc(var(--sfcg-step) * 3);
}

@media (hover: hover){
	.sfcg__card:hover .sfcg__arrow,
	.sfcg__card:focus-visible .sfcg__arrow{
		transform: translateX(0);
		/* lands after the stagger has run — --sfcg-n is the word's
		   length, written on by the script, with a floor so a short
		   word does not fire it instantly */
		transition-delay: max(100ms, calc(var(--sfcg-step) * (var(--sfcg-n, 6) - 1)));
	}

	.sfcg__card:hover .sfcg__name:not(.is-split),
	.sfcg__card:focus-visible .sfcg__name:not(.is-split){
		transform: translateX(0);
	}

	.sfcg__card:hover .sfcg__char,
	.sfcg__card:focus-visible .sfcg__char{
		transform: translateX(0);
		transition-delay: calc(var(--sfcg-step) * var(--sfcg-i, 0));
	}
}

/* NO HOVER MEANS NO ARROW, so on a touch screen the card would read
   as a photograph with a word on it and nothing to say it is a link.
   Below, the finished state IS the resting state. */
@media (hover: none){
	.sfcg__arrow{ transform: translateX(0) }
	.sfcg__char,
	.sfcg__name:not(.is-split){ transform: translateX(0) }
}

/* ── the card with no photograph ───────────────────────────── */

/* Others has no art yet. Rather than leave a black rectangle beside
   three photographs, it gets the house carbon with a magenta bloom —
   a deliberate surface, not a hole — and prints the same PLACEHOLDER
   tag the review cards use. Drop an <img class="sfcg__img"> into its
   .sfcg__media, delete data-placeholder, and all of this falls away.
   WHILE THE TAG IS VISIBLE THE SECTION IS NOT FINISHED. */
.sfcg__card[data-placeholder]{
	background-color: #101014;
	background-image:
		radial-gradient(120% 90% at 78% 112%, rgba(236,30,121,.42) 0%, rgba(236,30,121,0) 58%),
		repeating-linear-gradient( 45deg, rgba(255,255,255,.03) 0 2px, transparent 2px 4px),
		repeating-linear-gradient(-45deg, rgba(0,0,0,.40)       0 2px, transparent 2px 4px);
}

.sfcg__card[data-placeholder] .sfcg__veil{
	background: linear-gradient(to top, rgba(0,0,0,.55) 0%, rgba(0,0,0,0) 100%);
}

.sfcg__card[data-placeholder]::after{
	content: "PLACEHOLDER — ART TO COME";
	position: absolute;
	z-index: 3;
	top: clamp(14px, 1.5vw, 24px);
	left: clamp(16px, 1.7vw, 30px);
	padding: 5px 10px;
	border: 1px solid var(--sfcg-accent);
	border-radius: 999px;
	font-family: var(--sfcg-tech);
	font-size: 10px;
	font-weight: 700;
	letter-spacing: .14em;
	color: var(--sfcg-accent);
	pointer-events: none;
}

/* ── phones ────────────────────────────────────────────────── */

@media (max-width: 860px){
	.sfcg__grid{ grid-template-columns: minmax(0, 1fr) }

	/* The reference keeps `aspect-video` at every width and so does
	   this. On a 375px phone that is a 335x188 card, which is short
	   for a photograph with a word across the bottom — but the shape
	   was chosen deliberately over the taller alternatives (see the
	   card rule), and a phone-only exception would make the band two
	   different arrangements. */
}

/* THE FIXED SOCIAL RAIL SITS ON THE LABELS. .sfsr is pinned to the
   middle of the left edge of every page at z-index 90, 46px wide (40
   below 782px), and because it is FIXED every card's label passes
   behind it as the page scrolls. The same trap is recorded on
   .sfidx's row numbers and on .sfop's phone title.

   The indent goes on the LABEL, not on the card, so the photograph
   still bleeds to the card's own edge. It is a max() against this
   section's own gutter rather than a media query, so it is 0 on a
   desktop (where the 44px gutter already clears the rail) and ~38px
   on a phone, and disappears on its own as the window grows.

   CHECK THIS BY MEASURING THE TEXT, NOT THE ELEMENT — the shift is
   inside the label's box on the children, so a getBoundingClientRect
   on .sfcg__label still reads as a collision after the fix. Use a
   Range over the name's contents. */
@media (max-width: 1023px){
	.sfcg__label{
		padding-left: var(--sfcg-rail-clear);
	}
}

/* ── reduced motion ────────────────────────────────────────── */

/* Everything arrives, nothing moves. The reveal is forced open so a
   reader whose observer never fires still sees four cards, and the
   arrow is parked in its finished position so the affordance stays. */
@media (prefers-reduced-motion: reduce){
	.sfcg__card,
	.sfcg__eyebrow,
	.sfcg__title{
		--sfcg-in: 1 !important;
		opacity: 1 !important;
		transform: none !important;
		transition: none !important;
	}

	.sfcg__img,
	.sfcg__arrow,
	.sfcg__char,
	.sfcg__name:not(.is-split){
		transform: none !important;
		transition: none !important;
	}

	.sfcg__word{ animation: none }
}

/* ═══ SECTION 4 — SHOP BY BIKE (.sfcg--bikes), added 2026-11-11 ═══
   The home page carries a SECOND instance of this band: four model
   groups instead of four categories, same markup, same stylesheet,
   same script. Operator: "okay for section 4, i want like section 2
   .. make it the title like section 2, but for section 4 its related
   with shop by bike".

   THE MODIFIER IS WHERE ANYTHING THAT HAS TO DIFFER GOES. Everything
   above this line is shared on purpose — that is what "like section
   2" means, and it is why a re-pitch of the padding, the type or the
   cards lands on both bands in one pass. Do not fork this file.

   THE BAND NOW DARKENS EXACTLY AS SECTION 2 DOES, AND THAT IS THE
   POINT. It carried a filter() of its own for a few hours on
   2026-11-11, when its cards showed the catalogue's studio shots on
   white and needed taming. The operator supplied real bike photography
   the same day and asked for "overlay a bit dark i want like section
   2", so the filter is gone and the shared --sfcg-overlay is all that
   either band uses.

   Put a filter back only if a picture arrives that the overlay cannot
   hold — and if you do, remember the two multiply: brightness(.86)
   under a .34 overlay was measurably too dark, which is why that pair
   was softened to .95 before being dropped altogether.

   SO NOTHING IN THIS FILE IS BIKE-SPECIFIC EXCEPT THE PULL-UP BELOW,
   which is spacing, not looks. */

/* THE BAND SITS UP CLOSER TO THE VIDEO ABOVE IT (2026-11-11, "from
   section 3 to section 4, why too many space"). Measured at 1700, the
   exit from section 3 was 225px of empty carbon: 128 of its bottom
   padding and 98 of the top padding this band inherits from section 2.

   ** THE TOP PADDING IS NOT OURS TO TRIM. ** It is .sfcg's own, which
   is sohub's value plus the 3px the operator asked for, and section 2
   reads the same line — cutting it there would move the category band
   as well. So this band is pulled up instead, and section 3's bottom
   padding came down to 64 in video-panel.css to meet it.

   What it overlaps is section 3's padding band: pure background, same
   base colour and weave both sides, nothing drawn in it. The pull-up
   is capped under that padding so the video is never reached — 44px
   at 1700 against its 61, 20px at 375 against its 40.

   ** THIS RULE AND video-panel.css's BOTTOM PADDING ARE A PAIR. **
   Change one and check the other, or the two bands will touch. */
.sfcg--bikes{
	margin-top: calc(-1 * clamp(20px, 2.6vw, 44px));
}


/* ═══════════════════════════════════════════════════════════════
 * THE COUNT BADGE, AND THE SHOP CURSOR          (added 2026-11-13)
 * ---------------------------------------------------------------
 * Both asked for by the operator against a screenshot of the
 * localhost proposal (docs/home-concepts/proposal-2026-11-13.html).
 *
 * WHAT THE BADGE IS FOR. The four cards are photographs with a name
 * on them, and nothing on the page tells a visitor whether Brakes
 * holds two parts or two hundred. The badge turns four pictures
 * into four shelves with stock on them.
 *
 * ** THE FIGURES ARE TYPED INTO THE MARKUP, NOT QUERIED. ** Read
 * from the live store on 2026-11-13 over wp-cli, published products
 * with children included, minimum price over products with a price:
 *
 *     wheels          10   from RM 80
 *     suspensions     13   from RM 60
 *     brake-systems   17   from RM 17
 *     /shop/ (Others) 104  from RM 17     <- the whole catalogue
 *
 * They are typed for the same reason .sfidx's are: this band is an
 * Elementor HTML widget with no PHP behind it, and reading them live
 * would mean putting a query on the front page to print four
 * integers. ** RE-READ THEM WHEN STOCK LANDS. ** The reconciliation
 * CLAUDE.md describes has not happened, so every one of these moves.
 *
 * IT INHERITS --sfcg-in, so it arrives with its own card and needs
 * no second observer. The .12s delay lands it just behind the card
 * rather than with it, which is what stops the two reading as one
 * block sliding in.
 *
 * WHY NOT BOTTOM-LEFT, BESIDE THE NAME. The name slides right by
 * --sfcg-shift on hover and the arrow takes its place; anything else
 * on that line has to be moved out of the way or collide. Top right
 * is the only corner of this card that nothing else uses.
 * ═══════════════════════════════════════════════════════════════ */

.sfcg__badge{
	position: absolute;
	z-index: 3;                      /* media 0, veil 1, label 2 */
	top:   clamp(12px, 1.6vw, 20px);
	right: clamp(12px, 1.6vw, 20px);

	display: inline-flex;
	align-items: center;
	gap: .4em;

	padding: .64em 1.05em;
	border-radius: 999px;

	font-family: var(--sfcg-tech);
	font-size: clamp(9px, .78vw, 11px);
	font-weight: 700;
	line-height: 1;
	letter-spacing: .14em;
	text-transform: uppercase;
	white-space: nowrap;

	color: var(--sfcg-ink);

	/* A GLASS PILL, NOT A SOLID ONE. The cards carry four different
	   photographs and three of them are bright at the top, which is
	   the corner this sits in — a flat fill would read as a sticker
	   on one card and vanish on another. The blur plus the hairline
	   holds on all four. The emboss is deliberately NOT reused here:
	   it is the card's own device and a second embossed object
	   inside the card flattens both. */
	background: rgba(0,0,0,.5);
	-webkit-backdrop-filter: blur(8px);
	        backdrop-filter: blur(8px);
	border: 1px solid rgba(255,255,255,.16);

	opacity: var(--sfcg-in);
	transform: translateY(calc((1 - var(--sfcg-in)) * -10px));
	transition:
		opacity   .8s var(--sfcg-ease) .12s,
		transform .6s var(--sfcg-ease) .12s;
}

/* The price half in the accent. The count is the fact, the price is
   the hook, and splitting them is what stops the pill reading as one
   long label. */
.sfcg__badge b{
	font-weight: 700;
	color: var(--sfcg-accent);
}

/* ── the SHOP cursor ───────────────────────────────────────────
 * A magenta ring with the word in it, replacing the arrow pointer
 * over a card. Built by category-grid.js and appended to <body>.
 *
 * ** BEHIND (hover: hover) and (pointer: fine), AND THE SCRIPT USES
 * THE SAME QUERY. ** If the two ever disagree a touch device gets
 * `cursor: none` with nothing drawn in its place — an invisible
 * pointer. bike-wall.css records the same pairing for the same
 * reason; keep the two strings identical.
 *
 * TWO ELEMENTS, NOT ONE. The script writes a translate on the outer
 * span every frame, so the show/hide scale has to live on a child or
 * one would overwrite the other. The independent `translate:` and
 * `scale:` properties would avoid that and are NOT used: they are
 * inert in Claude's browser pane, so nothing here could be verified.
 *
 * THE COLOUR IS WRITTEN AGAINST THE WP PRESET, NOT --sfcg-accent.
 * The element lives on <body>, outside this band, so the --sfcg-*
 * tokens do not inherit to it. The preset is declared on :root and
 * always does.
 * ───────────────────────────────────────────────────────────── */

@media (hover: hover) and (pointer: fine){

	/* Two classes, like the emboss above: Elementor's frontend CSS
	   carries rules on `.elementor a` at 0,1,1 and this card is an
	   <a>. It does not set `cursor` today — this is a guard against
	   the day it does, and it costs nothing. */
	.sfcg .sfcg__card{ cursor: none }

	.sfcg__cursor{
		position: fixed;
		left: 0;
		top:  0;
		z-index: 95;              /* over .sfsr at 90, under the navbar */
		pointer-events: none;
		will-change: transform;
		transform: translate3d(-300px, -300px, 0);
	}

	.sfcg__cursor i{
		display: grid;
		place-items: center;

		width:  84px;
		height: 84px;
		margin: -42px 0 0 -42px;   /* centred on the pointer */
		border-radius: 50%;

		border: 1.5px solid var(--wp--preset--color--accent-1, #ec1e79);
		background: rgba(236, 30, 121, .16);
		-webkit-backdrop-filter: blur(2px);
		        backdrop-filter: blur(2px);

		opacity: 0;
		transform: scale(.35);
		transition:
			opacity   .3s  cubic-bezier(.16, 1, .3, 1),
			transform .45s cubic-bezier(.16, 1, .3, 1);
	}

	.sfcg__cursor.is-on i{
		opacity: 1;
		transform: none;
	}

	.sfcg__cursor b{
		font-family: "Inter", system-ui, -apple-system, sans-serif;
		font-size: 10px;
		font-weight: 700;
		line-height: 1;
		letter-spacing: .16em;
		text-transform: uppercase;
		color: var(--wp--preset--color--contrast, #f2f2f4);
	}
}

/* Reduced motion keeps both, and stops them moving. The badge is
   information, so it is never hidden; it simply arrives. */
@media (prefers-reduced-motion: reduce){
	.sfcg__badge{ transition: none }
	.sfcg__cursor i{ transition: none }
}
