/*
 * The landing page's design language (LANDING-02), and its styles.
 *
 * ## Why this file exists at all, before anything about design
 *
 * A separate same-origin file rather than an inline <style>, and that is a CSP
 * decision rather than a stylistic one.
 *
 * The edge policy is `style-src 'self' 'sha256-<the holding page's hash>'`. An
 * inline block here would need its **own** hash pinned in the Caddyfile — and
 * `deploy.sh` and `deploy-dev.sh` do not install Caddy config, so every future
 * edit to this file would need a manual edge change to be visible. With twelve
 * LANDING issues still to touch these styles, that is a trap that springs
 * repeatedly and silently: the page deploys, the CSS is blocked, and the deploy
 * reports success.
 *
 * `'self'` already allows this file, so it needs no CSP change now and none
 * ever again. The holding page's "zero external requests: no fonts, no CDN, no
 * analytics" property is kept — this is one same-origin file on the same
 * connection, not a third party.
 *
 * ## THE ONLY PLACE IN packages/web/landing THAT MAY CONTAIN A COLOUR VALUE
 *
 * The same rule `theme/tokens.css` carries for the client, for the same reason:
 * a stray literal is exactly the thing that survives review and then turns out
 * to be the one element nobody can read. `theme/landing-tokens.test.ts` scans
 * this directory — the HTML and the script included — and fails on a colour
 * anywhere but the `:root` block below.
 *
 * It cannot be `tokens.css` itself. This is a static document outside the Vite
 * client, so it can reach nothing the bundler produces. The tokens are therefore
 * **duplicated deliberately and pinned by test**: the values that carry identity
 * — the ground, the ink, the interaction accent, the brand amber — must equal
 * `theme/tokens.css`'s dark theme, and that test fails if either side moves.
 * A landing page that drifts from the product it sells is the failure worth
 * spending a test on.
 *
 * ## One composition, and it is dark
 *
 * The client has two themes; this page has one. The hero is a photograph of a
 * night world and the whole composition is built on it — there is no light
 * treatment of that image that is not a different page. So the tokens below are
 * a single set, `color-scheme` is pinned to dark, and the contrast figures
 * recorded beside them are the only ones there are. LANDING-10 holds them to AA;
 * these are measured, not asserted.
 */

:root {
  color-scheme: dark;

  /* ---- ground and surfaces -------------------------------------------
   *
   * `--lp-bg` is `tokens.css`'s `--bg-base`, and that is the reconciliation
   * LANDING-02 owed. There were two dark navies in this repository: the client's
   * #0b1017 and the holding page's #060a12, which this page had copied. They are
   * three points apart and nobody could tell them apart side by side, which is
   * precisely why the difference was not deliberate — it was a value typed twice.
   *
   * The front door now uses the product's ground, so a visitor who signs in
   * lands in the same world they were just looking at.
   *
   * The **holding page keeps #060a12**, and that is deliberate rather than
   * missed: its <style> is inline and pinned in the Caddyfile by hash, so
   * editing it is an edge rollout, and ADR-0028 retires the document at go-live
   * anyway. Two navies remain; one of them has an expiry date.
   */
  --lp-bg: #0b1017;
  --lp-bg-none: rgb(11 16 23 / 0%);
  --lp-surface: #131a24;
  --lp-surface-quiet: rgb(148 163 184 / 7%);
  --lp-border: rgb(148 163 184 / 14%);
  --lp-border-strong: rgb(148 163 184 / 28%);
  --lp-header-bg: rgb(11 16 23 / 86%);
  --lp-scrim: rgb(11 16 23 / 72%);
  --lp-scrim-strong: rgb(11 16 23 / 92%);

  /* ---- ink ------------------------------------------------------------
   *
   * Measured against the two grounds it is used on (AA needs 4.5:1 for body
   * text, 3:1 for large text and meaningful graphics):
   *
   *   --lp-text        on --lp-bg          16.38    on --lp-surface   15.01
   *   --lp-text-secondary on --lp-bg        8.19    on --lp-surface    7.51
   *   --lp-text-secondary on the quiet fill 6.73
   *
   * There is deliberately **no third, quieter ink**. `tokens.css`'s
   * `--text-muted` (#6d7c93) was the obvious import and it measures **4.13:1 on
   * --lp-surface** — under AA, on the cards, for copy. It sits at exactly 4.50
   * on the page ground, which is the kind of pass that becomes a fail the first
   * time a surface moves. So the quiet register on this page is
   * `--lp-text-secondary` at a smaller size, not a dimmer colour.
   */
  --lp-text: #e9eef6;
  --lp-text-secondary: #9dabc0;
  --lp-ink-on-accent: #04121f;

  /* ---- the two roles colour is allowed to carry -----------------------
   *
   * Blue is interaction. Amber is the mark. Nothing on this page is both, and
   * `landing-tokens.test.ts` fails if a rule paints an interactive selector
   * with `--lp-brand`. See `theme/tokens.css`, which carries the rule this
   * qualifies.
   *
   *   --lp-accent on --lp-bg (link text)          8.88
   *   --lp-ink-on-accent on --lp-accent (the CTA) 8.80
   *   --lp-brand on --lp-bg (mark, icons)        11.10
   *   --lp-brand on the hero wash                10.02
   */
  --lp-accent: #5eb8ff;
  --lp-accent-hover: #8ccdff;
  --lp-brand: #ffb84d;

  /* ---- the one status this page can show -------------------------------
   *
   * A refused sign-in, and nothing else. The client's whole status set does
   * not belong here — there is no flight to be late — but this one does,
   * because the OAuth callback can land a visitor back on this page having
   * failed, and until LANDING-04 it said nothing at all.
   *
   * The client's `--status-cancelled`, pinned by test like every other shared
   * value: a refusal should look the same on the front door as it does inside.
   * And paired with a glyph, per H.4 — meaning is never carried by hue alone,
   * which matters most in the one place a visitor is already confused.
   *
   *   --lp-status-refused on --lp-surface   6.32
   */
  --lp-status-refused: #f87171;
  --lp-status-refused-quiet: rgb(248 113 113 / 16%);

  /* ---- the provider marks ---------------------------------------------
   *
   * Google's and Discord's own colours, which is the whole point of using them:
   * a sign-in button that does not look like the service it opens is the one
   * people hesitate over.
   *
   * **Google's published button is not AA at body size, and this is the fix.**
   * White on #4285F4 measures **3.56:1** — under 4.5 for normal text, and
   * Google's branding guidelines specify that exact pairing, so honouring the
   * guideline and honouring AA are in direct conflict. WCAG's large-text
   * threshold is 3:1 at 18.66px bold, so `--lp-text-provider` is 1.1875rem
   * (19px) at weight 700 and the pair conforms *as large text* rather than by
   * waiver. Discord's white on #5865F2 is 4.61 and Twitch's on #9146FF is 4.64;
   * both would have passed either way, and both get the same treatment so the
   * three buttons match.
   *
   * The size is not decoration. Change it down and the Google button stops
   * being accessible — the test records the requirement next to the token.
   */
  --lp-google: #4285f4;
  --lp-google-field: #ffffff;
  --lp-discord: #5865f2;
  --lp-twitch: #9146ff;
  --lp-provider-ink: #ffffff;

  /* ---- the hero photograph ---------------------------------------------
   *
   * `--lp-hero-wash` is the map image's own deep navy, so the tint over it
   * belongs to the picture rather than flattening it toward page black. The
   * alpha steps are tokens rather than inline `rgb(… / …)` for the same reason
   * every other colour here is: they are the gradient, and a gradient stop typed
   * by hand in four places drifts.
   */
  --lp-hero-wash: #0b1a38;
  --lp-hero-wash-strong: rgb(11 26 56 / 88%);
  --lp-hero-wash-mid: rgb(11 26 56 / 60%);
  --lp-hero-wash-soft: rgb(11 26 56 / 22%);
  --lp-hero-wash-none: rgb(11 26 56 / 0%);
  --lp-hero-veil: rgb(11 26 56 / 82%);
  --lp-hero-veil-deep: rgb(9 18 38 / 94%);

  /* ---- type ------------------------------------------------------------
   *
   * The same system stack `tokens.css` chose, and for its reason: the CSP has no
   * `font-src`, so a web font would be blocked by `default-src 'self'` — and a
   * marketing page whose headline arrives late is the one LANDING-11 is trying
   * to prevent.
   *
   * **The text steps are the client's**, so the front door and the product set
   * type at the same sizes. `--lp-text-md` and `--lp-text-provider` are the two
   * additions, and both have a stated job.
   *
   * **The display steps are this page's alone**, and they are not in
   * `tokens.css` on purpose. Nothing in the client sets 4.4rem type; a display
   * scale sitting unused in the shared token file is how `--rail-width-collapsed`
   * came to be defined-and-unused, which LANDING-10 names as a smell. They live
   * where they are used.
   *
   * Every display step is a `clamp()` rather than a media-query staircase, so
   * the headline reflows continuously and survives 200% text scaling — the
   * failure LANDING-10 tests for is a display headline overflowing, and a clamp
   * with a `rem` floor scales with the user's text size where a `vw`-only rule
   * would not.
   */
  --lp-font:
    ui-sans-serif, system-ui, -apple-system, 'Segoe UI', Roboto, Helvetica, Arial, sans-serif;

  --lp-text-xs: 0.6875rem;
  --lp-text-sm: 0.8125rem;
  --lp-text-base: 0.9375rem;
  --lp-text-md: 1.0625rem;
  --lp-text-lg: 1.125rem;
  --lp-text-provider: 1.1875rem;

  --lp-display-1: clamp(2.4rem, 6.4vw, 4.4rem);
  --lp-display-2: clamp(1.5rem, 3vw, 2.1rem);
  --lp-display-3: clamp(1.25rem, 2.4vw, 1.7rem);
  --lp-display-4: clamp(1.3rem, 2.6vw, 1.75rem);
  --lp-lede-size: clamp(1rem, 1.4vw, 1.0625rem);

  --lp-weight-regular: 400;
  --lp-weight-semibold: 600;
  --lp-weight-bold: 700;
  --lp-weight-display: 800;

  --lp-leading-display: 0.98;
  --lp-leading-tight: 1.15;
  --lp-leading-snug: 1.3;
  --lp-leading-normal: 1.5;

  --lp-tracking-display: -0.02em;
  --lp-tracking-title: -0.01em;
  --lp-tracking-label: 0.06em;
  --lp-tracking-wordmark: 0.22em;

  /* ---- space ----------------------------------------------------------
   *
   * A rhythm, not a pile. The first draft of this page used twenty-one distinct
   * gap values between 0.4rem and 2.75rem, most of them within a hair of each
   * other; they are snapped onto these steps, which moves a few edges by a pixel
   * or two and makes the page's vertical rhythm something a later section can
   * match rather than guess at.
   */
  --lp-space-1: 0.25rem;
  --lp-space-2: 0.5rem;
  --lp-space-3: 0.75rem;
  --lp-space-4: 1rem;
  --lp-space-5: 1.25rem;
  --lp-space-6: 1.5rem;
  --lp-space-7: 1.75rem;
  --lp-space-8: 2rem;
  --lp-space-10: 2.75rem;

  --lp-radius-sm: 0.375rem;
  --lp-radius-md: 0.55rem;
  --lp-radius-lg: 0.85rem;
  --lp-radius-round: 50%;

  /* ---- layout ----------------------------------------------------------
   *
   * **Four bands, three boundaries** — mobile, tablet, laptop, wide desktop —
   * which is what LANDING-10 needs and what the client's single `--bp-mobile`
   * could not express. The laptop boundary is the one that matters most: a
   * two-column hero with a bleeding map has an awkward middle, and that band is
   * also the app's own known gap (UX-09).
   *
   * **A media query cannot read a custom property.** `@media (min-width:
   * var(--lp-bp-laptop))` is not valid CSS and fails silently — the query never
   * matches and the layout below it simply never arrives. So the literals stay
   * in the queries and `landing-tokens.test.ts` asserts that every width in
   * every media query is one of these three values. The token is the source of
   * truth; the test is what keeps the literal honest.
   *
   * Everything below is written mobile-first, `min-width` only, so no boundary
   * needs an off-by-one companion (`max-width: 47.99rem`) that a token cannot
   * express either.
   */
  --lp-bp-tablet: 48rem;
  --lp-bp-laptop: 64rem;
  --lp-bp-desktop: 90rem;

  /*
   * A **height** boundary, which is the axis the four width bands cannot
   * express and the one the hero actually runs out of.
   *
   * A 1280x720 laptop is wide and short: every width-driven rule pays out its
   * maximum there, and the display headline alone took 207px of a ~700px
   * budget. Below this line the hero uses a smaller display treatment — which
   * is not a compromise so much as the correct type size for the screen.
   *
   * 50rem (800px) sits above every common laptop height (720, 768) and below
   * every desktop one (900, 1080) and the tall phones (844, 852), so the
   * block catches exactly the band that needs it.
   */
  --lp-bp-short: 50rem;

  --lp-page-max: 120rem;
  --lp-gutter: clamp(1rem, 3.5vw, 3rem);
  --lp-header-h: 4.5rem;
  /*
   * How much of the next section shows under the hero.
   *
   * The hero used to be exactly `100svh - header`, so the first screen ended
   * on a clean edge and gave a visitor no reason to believe anything followed
   * it. A sliver of the next section is the cheapest possible affordance and
   * the only one that needs no chrome — no bouncing chevron, no 'scroll'
   * label, nothing to translate or animate.
   *
   * Small on purpose: it comes out of the CTA's budget, and the CTA being
   * above the fold is the criterion that outranks this one.
   */
  --lp-peek: 3.5rem;
  /*
   * The hero's own vertical rhythm, which answers to height rather than the
   * page's spacing scale.
   *
   * Everything else on this page is spaced on the `--lp-space-*` steps, and
   * the hero cannot be, because it is the one block with a hard vertical
   * budget: headline, lede, four pillars and the sign-in card have to fit
   * above the fold together. On a 1280x720 laptop the fixed steps put the CTA
   * 58px below it — measured, not guessed.
   *
   * So the three gaps in the stack breathe with the viewport and stop at the
   * scale's own values on a tall screen. It is the smallest change that keeps
   * the composition intact; the alternative was shrinking the headline, which
   * is the one element the hero exists to show.
   */
  --lp-hero-gap: clamp(0.625rem, 2.4svh, 1.75rem);
  --lp-hero-gap-tight: clamp(0.5rem, 1.8svh, 1.5rem);
  --lp-measure: 34rem;
  --lp-measure-lede: 32rem;

  /* ---- motion ----------------------------------------------------------
   *
   * Zeroed at the root under `prefers-reduced-motion`, exactly as `tokens.css`
   * does it, so **no component below needs its own media query** — which is the
   * property that keeps the rule true as LANDING-04…09 add sections. The
   * carousel's slide is the one long duration on the page and it is a token, so
   * reduced motion turns it off without the carousel knowing.
   */
  --lp-motion-fast: 150ms;
  --lp-motion-slide: 620ms;
  --lp-ease: cubic-bezier(0.2, 0, 0.2, 1);
  --lp-ease-slide: cubic-bezier(0.22, 0.61, 0.36, 1);
}

/*
 * The page's only `prefers-reduced-motion` block, and it is root-level.
 *
 * Somebody who asked their system not to animate did not mean "except when the
 * page scrolls itself", so `scroll-behavior` is here too — it is the one motion
 * that is not a duration and therefore cannot be a token.
 */
@media (prefers-reduced-motion: reduce) {
  :root {
    --lp-motion-fast: 0ms;
    --lp-motion-slide: 0ms;
  }

  html {
    scroll-behavior: auto;
  }
}

/* ---- base ------------------------------------------------------------- */

*,
*::before,
*::after {
  box-sizing: border-box;
}

html {
  scroll-behavior: smooth;
}

body {
  margin: 0;
  background: var(--lp-bg);
  color: var(--lp-text);
  font-family: var(--lp-font);
  font-size: var(--lp-text-base);
  line-height: var(--lp-leading-normal);
  -webkit-font-smoothing: antialiased;
}

a {
  color: inherit;
}

:focus-visible {
  outline: 2px solid var(--lp-accent-hover);
  outline-offset: 2px;
}

/* ---- the component family --------------------------------------------
 *
 * Every class on this page is `lp-`. The client's own families — `admin__*`,
 * `login__*`, the console's — are not reachable from here and must not be
 * recreated here either; a landing page borrowing `.login__button` would be the
 * same mistake UX-11 exists to fix, in a new place.
 *
 * The prefix is not only hygiene. `.card`, `.panel` and `.stat` — which this
 * page used before LANDING-02 — are all live class names in `dashboard.css`, so
 * the moment ADR-0028's deferred "static shell that hydrates" arrives, an
 * unprefixed family collides with the product on its first day.
 *
 * `docs/landing-design-language.md` is the catalogue LANDING-04…09 compose from.
 */

.lp-wrap {
  width: 100%;
  max-width: var(--lp-page-max);
  margin-inline: auto;
  padding-inline: var(--lp-gutter);
}

/* Keyboard users get to skip the nav; LANDING-10 owns the full pass. */
.lp-skip {
  position: absolute;
  left: -9999px;
}

.lp-skip:focus {
  left: var(--lp-space-4);
  top: var(--lp-space-4);
  z-index: 10;
  padding: var(--lp-space-3) var(--lp-space-4);
  background: var(--lp-surface);
  border: 1px solid var(--lp-border);
  border-radius: var(--lp-radius-md);
}

.lp-sr-only {
  position: absolute;
  width: 1px;
  height: 1px;
  margin: -1px;
  padding: 0;
  overflow: hidden;
  clip-path: inset(50%);
  white-space: nowrap;
  border: 0;
}

/* ---- header ----------------------------------------------------------- */

.lp-header {
  border-bottom: 1px solid var(--lp-border);
  background: var(--lp-header-bg);
}

.lp-header__inner {
  display: flex;
  align-items: center;
  gap: var(--lp-space-6);
  padding-block: var(--lp-space-4);
}

.lp-brand {
  display: flex;
  align-items: center;
  gap: var(--lp-space-3);
  font-weight: var(--lp-weight-bold);
  letter-spacing: var(--lp-tracking-wordmark);
  text-decoration: none;
  font-size: var(--lp-text-base);
}

/*
 * The tail fin, in the brand amber.
 *
 * Decorative and not interaction, even though it sits inside a link: what makes
 * the wordmark clickable is the word beside it, in `--lp-text`. This is the
 * exact case the amber rule is qualified for.
 */
.lp-brand__mark {
  width: 1.6rem;
  height: 1.6rem;
  color: var(--lp-brand);
}

.lp-header__nav {
  margin-left: auto;
  display: flex;
  align-items: center;
  gap: var(--lp-space-7);
}

/*
 * The in-page links are hidden on a phone, where the CTA is the only thing worth
 * the width. Mobile-first: they arrive at the tablet boundary.
 */
.lp-header__nav a:not(.lp-cta) {
  display: none;
  color: var(--lp-text-secondary);
  text-decoration: none;
  font-size: var(--lp-text-base);
  transition: color var(--lp-motion-fast) var(--lp-ease);
}

.lp-header__nav a:not(.lp-cta):hover {
  color: var(--lp-text);
}

@media (min-width: 48rem) {
  .lp-header__nav a:not(.lp-cta) {
    display: inline;
  }
}

.lp-cta {
  display: inline-flex;
  align-items: center;
  gap: var(--lp-space-2);
  padding: var(--lp-space-3) var(--lp-space-5);
  border-radius: var(--lp-radius-md);
  background: var(--lp-accent);
  color: var(--lp-ink-on-accent);
  font-weight: var(--lp-weight-semibold);
  text-decoration: none;
  font-size: var(--lp-text-base);
  transition: background var(--lp-motion-fast) var(--lp-ease);
}

.lp-cta:hover {
  background: var(--lp-accent-hover);
}

/* ---- hero ------------------------------------------------------------- */

/*
 * The hero fills the first screen, and the world sits behind it.
 *
 * `100svh` — the *small* viewport height — rather than `100vh`: on mobile,
 * `100vh` is the height with the browser chrome hidden, so a `100vh` hero is
 * taller than the screen at rest and pushes the sign-in button below the fold
 * on the one device where that matters most. `svh` is the conservative
 * measurement, and the header is subtracted so header + hero come to exactly one
 * screen. `max()` keeps it sane on a short landscape window.
 */
.lp-hero {
  position: relative;
  overflow: clip;
  display: grid;
  align-items: center;
  min-height: max(34rem, calc(100svh - var(--lp-header-h) - var(--lp-peek)));
  /*
   * Height-based, not width-based, and that was a measured correction.
   *
   * This was `clamp(2rem, 5vw, 4rem)`, which is the wrong axis: what runs out in
   * a hero is vertical room, and a wide-but-short laptop is exactly the viewport
   * where a width-driven clamp pays out its *maximum* padding. At 1280x720 the
   * sign-in card's bottom measured 778px against a 720px fold — the CTA, the one
   * thing the hero exists for, was below it.
   *
   * `svh` matches the min-height above, so both react to the same measurement of
   * the viewport rather than disagreeing while the browser chrome slides away.
   */
  padding-block: clamp(1.25rem, 4svh, 4rem);
  background-color: var(--lp-bg);
}

/*
 * The map, as its own layer rather than a background on the section.
 *
 * A pseudo-element because the map is **inset to the right** on wide screens
 * rather than covering the section: measured off the mock, it occupies about the
 * right 68% of the hero, which is what puts the Americas clear of the headline
 * instead of underneath it. `background-size: cover` cannot express that — cover
 * fills the box, so the only way to move the map right is to crop the Americas
 * off the left, which is the opposite of what is wanted.
 *
 * Its own element also means the wash below can sit on top of the map without
 * also washing the section, and the map can be positioned without fighting the
 * gradient's own `background-position`.
 *
 * Mobile first: full width and quiet. Inset to the right on a phone would leave
 * a strip too small to read as a map. LANDING-10 owns the purpose-made narrow
 * crop that should eventually replace the scaled wide one here.
 */
.lp-hero::before {
  content: '';
  position: absolute;
  inset: 0;
  background-image: url('/landing-hero.webp');
  background-repeat: no-repeat;
  background-position: center;
  /*
   * `100% auto`, not `cover`. Cover fills the box, and this box is taller than
   * the map is proportioned — so cover scales to the height and crops the sides,
   * which took the Americas off the left edge. Fitting the width shows the whole
   * world and letterboxes vertically instead.
   */
  background-size: 100% auto;
  /*
   * The letterbox is why this mask exists. The map carries its own deep-navy
   * background, and where that meets the page's near-black it draws a horizontal
   * line across the hero. Feathering the top and bottom dissolves the seam, so an
   * inset map reads as part of the page rather than as a pasted rectangle.
   */
  mask-image: linear-gradient(to bottom, transparent 0%, #000 16%, #000 84%, transparent 100%);
  pointer-events: none;
}

/*
 * The wash over it, and the reason it is blue rather than black.
 *
 * The headline needs a floor to sit on, but painting it in the page's near-black
 * flattens the map to nothing exactly where the Americas are — the part the mock
 * deliberately keeps legible. A **blue** tint at lower opacity does the same job
 * for contrast while leaving the coastline and city lights visible through it, so
 * the left of the hero still reads as ocean rather than as a black panel.
 */
.lp-hero::after {
  content: '';
  position: absolute;
  inset: 0;
  pointer-events: none;
  background-image: linear-gradient(
    to bottom,
    var(--lp-hero-veil) 0%,
    var(--lp-hero-veil-deep) 58%,
    var(--lp-bg) 100%
  );
}

@media (min-width: 48rem) {
  .lp-hero::before {
    inset: 0 0 0 auto;
    width: 68%;
  }

  .lp-hero::after {
    background-image:
      linear-gradient(
        100deg,
        var(--lp-hero-wash) 0%,
        var(--lp-hero-wash-strong) 24%,
        var(--lp-hero-wash-mid) 40%,
        var(--lp-hero-wash-soft) 58%,
        var(--lp-hero-wash-none) 74%
      ),
      /* Feathers the map's left edge into the page so the inset does not read as a
         pasted rectangle. */
        linear-gradient(to right, var(--lp-bg) 0%, var(--lp-bg-none) 26%),
      /* And the bottom, so the section below meets the map rather than colliding. */
        linear-gradient(to bottom, var(--lp-bg-none) 62%, var(--lp-bg) 100%);
  }
}

/*
 * The copy sits at the page's left margin, in line with the wordmark above it —
 * not centred, and above both map layers.
 *
 * The grid carries `.lp-wrap` too, so capping *it* would leave `.lp-wrap`'s
 * `margin-inline: auto` centring a narrow box in a wide viewport, which is what
 * it did on the first attempt. The cap belongs on the child; the wrap keeps its
 * full width and its padding.
 */
.lp-hero__grid {
  position: relative;
  z-index: 1;
}

.lp-hero__copy {
  max-width: var(--lp-measure);
}

.lp-headline {
  margin: 0 0 var(--lp-hero-gap);
  font-size: var(--lp-display-1);
  line-height: var(--lp-leading-display);
  letter-spacing: var(--lp-tracking-display);
  text-transform: uppercase;
  font-weight: var(--lp-weight-display);
  text-wrap: balance;
}

/* The full stop after MAKE IT YOURS. Punctuation, not a control. */
.lp-headline__stop {
  color: var(--lp-brand);
}

.lp-lede {
  margin: 0 0 var(--lp-hero-gap);
  color: var(--lp-text-secondary);
  max-width: var(--lp-measure-lede);
  font-size: var(--lp-lede-size);
}

.lp-pillars {
  display: flex;
  flex-wrap: wrap;
  gap: var(--lp-hero-gap-tight) var(--lp-space-6);
  margin: 0 0 var(--lp-hero-gap);
  padding: 0;
  list-style: none;
}

.lp-pillars__item {
  display: flex;
  align-items: center;
  gap: var(--lp-space-2);
  font-size: var(--lp-text-sm);
  line-height: var(--lp-leading-tight);
  color: var(--lp-text-secondary);
  max-width: 8.5rem;
}

/* Feature shorthand, beside the mark's own colour. Decorative, never a control. */
.lp-pillars__icon {
  flex: none;
  color: var(--lp-brand);
}

/* ---- sign-in ----------------------------------------------------------- */

.lp-signin {
  border: 1px solid var(--lp-border);
  border-radius: var(--lp-radius-lg);
  background: var(--lp-surface);
  padding: var(--lp-space-6);
  max-width: 30rem;
}

.lp-signin__title {
  margin: 0 0 var(--lp-space-4);
  font-size: var(--lp-text-md);
  line-height: var(--lp-leading-snug);
}

/*
 * A refused sign-in, explained inside the card (LANDING-04).
 *
 * Inside rather than above it: the card is what the visitor was interacting with,
 * and an error banner at the top of a long hero is an error they scroll away
 * from. It sits between the heading and the buttons, so the explanation is read
 * before the button that failed is pressed again.
 *
 * Rendered by the server — this page's funnel carries no JavaScript, and an
 * error only a scripting-enabled browser can see is one some visitors never get.
 */
.lp-alert {
  display: flex;
  align-items: flex-start;
  gap: var(--lp-space-3);
  margin: 0 0 var(--lp-space-4);
  padding: var(--lp-space-3);
  border: 1px solid var(--lp-status-refused);
  border-radius: var(--lp-radius-md);
  background: var(--lp-status-refused-quiet);
  color: var(--lp-text);
  font-size: var(--lp-text-sm);
  line-height: var(--lp-leading-snug);
}

/*
 * The shape half of the pair. H.4 is explicit that status is never carried by
 * hue alone, and this is the one place on the page that shows a status at all —
 * so the ring is not decoration, it is what survives greyscale, colour blindness
 * and a monochrome print.
 */
.lp-alert__glyph {
  flex: none;
  display: grid;
  place-items: center;
  width: 1.25rem;
  height: 1.25rem;
  border: 1px solid var(--lp-status-refused);
  border-radius: var(--lp-radius-round);
  color: var(--lp-status-refused);
  font-weight: var(--lp-weight-bold);
  font-size: var(--lp-text-xs);
  line-height: 1;
}

.lp-signin__buttons {
  display: grid;
  gap: var(--lp-space-3);
}

/*
 * Each provider wears its own colour, because a sign-in button that does not
 * look like the service it opens is the one people hesitate over. Both marks are
 * inlined as SVG paths rather than fetched, so the document keeps making zero
 * external requests.
 *
 * The size and weight are load-bearing, not styling — see `--lp-text-provider`.
 * White on Google's blue is 3.56:1, which conforms only as WCAG large text
 * (>=18.66px bold). Shrink this and the button stops being accessible.
 */
.lp-provider {
  display: flex;
  align-items: center;
  justify-content: center;
  gap: var(--lp-space-3);
  width: 100%;
  padding: var(--lp-space-3) var(--lp-space-4);
  min-height: 3rem;
  border-radius: var(--lp-radius-md);
  font-size: var(--lp-text-provider);
  font-weight: var(--lp-weight-bold);
  text-decoration: none;
  color: var(--lp-provider-ink);
  transition: filter var(--lp-motion-fast) var(--lp-ease);
}

.lp-provider--google {
  background: var(--lp-google);
}

.lp-provider--discord {
  background: var(--lp-discord);
}

.lp-provider--twitch {
  background: var(--lp-twitch);
}

.lp-provider:hover {
  filter: brightness(1.08);
}

/* The Google mark is four fixed brand colours and must not inherit ours. */
.lp-provider--google .lp-provider__mark {
  background: var(--lp-google-field);
  border-radius: 2px;
  padding: 1px;
}

.lp-signin__note {
  margin: var(--lp-space-3) 0 0;
  font-size: var(--lp-text-sm);
  color: var(--lp-text-secondary);
}

/* ---- sections ---------------------------------------------------------- */

.lp-section {
  padding-block: clamp(2.5rem, 6vw, 4.5rem);
}

.lp-section__title {
  margin: 0 0 var(--lp-space-6);
  font-size: var(--lp-display-2);
  line-height: var(--lp-leading-tight);
  letter-spacing: var(--lp-tracking-title);
}

.lp-split {
  display: grid;
  gap: var(--lp-space-6);
  grid-template-columns: minmax(0, 1fr);
}

/*
 * Two columns from the laptop band up. This is the boundary the mock is most
 * likely to break at, and it is deliberately not the tablet one: a 2fr/1fr split
 * at 48rem leaves the aside too narrow to hold its own headline.
 */
@media (min-width: 64rem) {
  .lp-split {
    grid-template-columns: minmax(0, 2fr) minmax(0, 1fr);
    align-items: start;
  }
}

.lp-cards {
  display: grid;
  gap: var(--lp-space-4);
  grid-template-columns: repeat(auto-fit, minmax(15rem, 1fr));
}

.lp-card {
  border: 1px solid var(--lp-border);
  border-radius: var(--lp-radius-lg);
  background: var(--lp-surface);
  padding: var(--lp-space-4);
}

.lp-card__title {
  margin: 0 0 var(--lp-space-4);
  font-size: var(--lp-text-md);
  line-height: var(--lp-leading-snug);
}

.lp-card__body {
  margin: var(--lp-space-4) 0 0;
  color: var(--lp-text-secondary);
  font-size: var(--lp-text-sm);
}

.lp-figure {
  min-height: 6.5rem;
  border-radius: var(--lp-radius-md);
  background: var(--lp-surface-quiet);
  display: grid;
  place-items: center;
  color: var(--lp-text-secondary);
  font-size: var(--lp-text-xs);
  text-align: center;
  padding: var(--lp-space-3);
}

/* ---- the fleet carousel (LANDING-06) ----------------------------------- */

.lp-fleet {
  position: relative;
  margin-bottom: var(--lp-space-4);
}

.lp-fleet__viewport {
  overflow: clip;
  border-radius: var(--lp-radius-md);
  background: var(--lp-surface-quiet);
}

.lp-fleet__track {
  display: flex;
  margin: 0;
  padding: 0;
  list-style: none;
  /*
   * `transform` rather than `left` or `scroll-left` so the slide runs on the
   * compositor and never lays the page out again mid-move. Reduced motion zeroes
   * the token at the root — the carousel still works, it just arrives instead of
   * gliding, and this rule needs no media query of its own to know that.
   */
  transition: transform var(--lp-motion-slide) var(--lp-ease-slide);
}

.lp-fleet__slide {
  flex: 0 0 100%;
  min-width: 0;
  text-align: center;
  padding: var(--lp-space-1) var(--lp-space-1) var(--lp-space-3);
}

.lp-fleet__image {
  display: block;
  width: 100%;
  height: auto;
  border-radius: var(--lp-radius-sm);
}

.lp-fleet__type {
  margin: var(--lp-space-2) 0 0;
  font-size: var(--lp-text-sm);
  font-weight: var(--lp-weight-semibold);
  color: var(--lp-text);
}

/*
 * The arrows.
 *
 * Hidden until `landing.js` sets `data-ready`. A carousel button that does
 * nothing because the script was blocked, slow or broken is worse than no
 * button — it reads as a bug rather than as a page that degrades.
 */
.lp-fleet__nav {
  display: none;
  position: absolute;
  top: calc(50% - 0.9rem);
  transform: translateY(-50%);
  width: 1.85rem;
  height: 1.85rem;
  place-items: center;
  border: 1px solid var(--lp-border);
  border-radius: var(--lp-radius-round);
  background: var(--lp-scrim);
  color: var(--lp-text);
  font-size: var(--lp-text-lg);
  line-height: 1;
  cursor: pointer;
  padding: 0;
  transition: background var(--lp-motion-fast) var(--lp-ease);
}

.lp-fleet[data-ready] .lp-fleet__nav {
  display: grid;
}

.lp-fleet__nav:hover {
  background: var(--lp-scrim-strong);
  border-color: var(--lp-border-strong);
}

.lp-fleet__nav--prev {
  left: var(--lp-space-1);
}

.lp-fleet__nav--next {
  right: var(--lp-space-1);
}

/* ---- the persistent-world aside (LANDING-07) --------------------------- */

.lp-aside {
  border: 1px solid var(--lp-border);
  border-radius: var(--lp-radius-lg);
  background: var(--lp-surface);
  padding: var(--lp-space-5);
}

.lp-aside__title {
  margin: 0 0 var(--lp-space-3);
  font-size: var(--lp-display-3);
  line-height: var(--lp-leading-tight);
  letter-spacing: var(--lp-tracking-title);
}

.lp-aside__body {
  margin: 0 0 var(--lp-space-4);
  color: var(--lp-text-secondary);
  font-size: var(--lp-text-sm);
}

/* ---- stats (LANDING-09) ------------------------------------------------ */

.lp-stats {
  display: grid;
  gap: 0;
  grid-template-columns: repeat(auto-fit, minmax(12rem, 1fr));
  border: 1px solid var(--lp-border);
  border-radius: var(--lp-radius-lg);
  background: var(--lp-surface);
  overflow: clip;
}

.lp-stat {
  padding: var(--lp-space-5);
  border-right: 1px solid var(--lp-border);
}

.lp-stat:last-child {
  border-right: 0;
}

.lp-stat__value {
  font-size: var(--lp-display-4);
  font-weight: var(--lp-weight-bold);
  line-height: var(--lp-leading-tight);
  letter-spacing: var(--lp-tracking-title);
}

/*
 * An unknown number reads as unknown, not as zero — and not in a dimmer colour
 * either, because the dim register on this page is size and weight rather than a
 * third ink that does not clear AA on this surface.
 */
.lp-stat__value--pending {
  font-weight: var(--lp-weight-regular);
}

/*
 * ...and a figure the server actually measured takes the full weight back.
 *
 * Keyed on `:has([data-count])` rather than on a second modifier class, because
 * the thing that decides is whether a *number* arrived, and that is decided
 * inside the element by a server-filled slot. A modifier would mean the server
 * writing a class onto a parent it does not otherwise touch, and the two could
 * then disagree — a bold em-dash, or a measured figure wearing the unknown
 * treatment.
 *
 * `data-count` is the same attribute the count-up animation keys on, so the
 * styling and the behaviour cannot drift apart: one attribute, one meaning.
 */
.lp-stat__value:has([data-count]) {
  font-weight: var(--lp-weight-bold);
}

.lp-stat__label {
  color: var(--lp-text-secondary);
  font-size: var(--lp-text-sm);
}

/* ---- footer ------------------------------------------------------------ */

.lp-footer {
  border-top: 1px solid var(--lp-border);
  padding-block: var(--lp-space-8) var(--lp-space-10);
  text-align: center;
}

.lp-footer__line {
  display: flex;
  align-items: center;
  justify-content: center;
  gap: var(--lp-space-2);
  font-weight: var(--lp-weight-semibold);
}

.lp-footer__mark {
  color: var(--lp-brand);
}

.lp-footer__sub {
  margin: var(--lp-space-1) 0 0;
  color: var(--lp-text-secondary);
  font-size: var(--lp-text-sm);
}

/*
 * Scaffolding. Every one of these names an issue that owes this page something
 * real, and LANDING-04…09 delete them as they land.
 */
.lp-todo {
  display: inline-block;
  margin-top: var(--lp-space-2);
  padding: 0 var(--lp-space-2);
  border: 1px dashed var(--lp-border-strong);
  border-radius: var(--lp-radius-sm);
  color: var(--lp-text-secondary);
  font-size: var(--lp-text-xs);
  letter-spacing: var(--lp-tracking-label);
  text-transform: uppercase;
}

/*
 * Short viewports: the hero, compressed to fit its own CTA.
 *
 * The criterion this serves is that the sign-in button is above the fold, and it
 * was failing at 1280x720 — first by 58px, and then, after the padding and gaps
 * were made height-aware, by 45px more on CI than on the machine it was tuned
 * on. That gap is font metrics: the system stack resolves to Segoe UI on
 * Windows and to something wider on a Linux runner, so the headline wraps
 * differently and every measurement taken on one platform is a guess about the
 * other.
 *
 * Hence a real reduction rather than another few pixels. The headline is the
 * largest single consumer of vertical space on the page, and 70px of uppercase
 * display type is a desktop treatment being shown on a screen that cannot hold
 * it. Shrinking it here buys enough room that the fold assertion passes with
 * margin on both platforms instead of passing on one and failing on the other.
 *
 * A height query, not a width one, and it is the first on this page. The four
 * width bands are the right model for layout; they cannot express "this screen
 * is short", which is the only thing wrong at 1280x720.
 */
@media (max-height: 50rem) {
  .lp-hero {
    padding-block: clamp(1rem, 3svh, 4rem);
  }

  .lp-headline {
    font-size: clamp(2rem, 6svh, 3.75rem);
  }

  .lp-lede {
    font-size: var(--lp-text-base);
  }
}