/* ============================================================
   "From the first people to use it." — Wispr Flow section
   Standalone replica: design tokens, layout + card styles
   lifted 1:1 from the live Webflow stylesheet.
   ============================================================ */

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

:root {
  /* ---- Nestify palette (nestifyapp.org) ---- */
  --nes-ink:    #1a1a1a;  /* the logo's black (assets/nestify-logo.svg) */
  --nes-cream:  #f5f4f0;  /* review card surface       */
  --nes-sand:   #f4f2ea;  /* --secondary / --muted     */
  --nes-tan:    #dcd7c9;  /* --border                  */
  --nes-muted:  #787165;  /* --muted-foreground        */
  --nes-white:  #ffffff;
  --nes-blue:   #2497f0;  /* brand blue (buttons)      */
  --nes-blue-2: #1e88e5;  /* brand blue (headline)     */
  /* 🔴 The one blue WHITE TYPE CAN STAND ON, and it exists because none
     of the others can. Measured against #fff: --bead-blue 2.71:1,
     --nes-blue 3.11:1, --nes-blue-2 3.68:1 — all three fail AA's 4.5,
     and the first two fail even the 3:1 that large type gets. They are
     all correct blues; they are just blues for INK to sit on, which is
     what the site does everywhere else. This is the lightest step that
     clears 4.5 (it measures 4.85), so a filled button that has to carry
     a white label reaches for this and nothing else. */
  --nes-blue-3: #1573c9;  /* filled buttons with white labels · 4.85:1 */

  /* 🔴 THE CTA BLUE, AND IT IS ITS OWN TOKEN because three buttons on two
     pages wear it and they had been pointing at three different things.
     The home page's "Explore the Family Hub" and "Learn more" and the
     device page's "Join the waitlist" are one object with one job; when
     the pair of states was inverted and put back, it had to be found in
     three files. Now it is here.

     🔴 It is NOT --bead-blue. The bead is the logo's i-dot at #14a4fd,
     which is the right blue for a 12px dot and the wrong one under a white
     label: white on it measures 2.71:1, under the 3:1 a button label
     needs. #1A80EA takes the same white to 3.95:1 and still reads as the
     same blue at pill size. --nes-blue-3 (4.85:1) is darker still and is
     there if a label ever has to clear 4.5. */
  --cta-blue:   #1A80EA;  /* filled CTAs, white label · 3.95:1 */
  --nes-sky:    #dbeafc;  /* light blue tint           */
  /* 🔴 LIFTED FROM apple.com/ipad-air, and it is not a colour anyone would
     have typed. `body.ac-theme-light { background-color: #f3f6f6 }` in
     /v/ipad-air/ah/built/styles/overview.built.css — confirmed against the
     rendered page, which samples rgb(243,246,246) flat from the header to
     the footer with no gradient anywhere in it.

     🔴 IT IS CYAN, NOT BLUE, and that is the whole reason it works. R 243,
     G 246, B 246 — green and blue EQUAL, red one step under, which puts
     the hue at exactly 180°. The obvious near-white to reach for is
     Apple's other one, #f5f5f7 (R245 G245 B247, hue 240°, genuinely
     blue-violet), and it is what the same page uses for its ribbon. At a
     glance the two are the same grey. Side by side one is cool-green and
     one is cool-purple, and only the first reads as DAYLIGHT rather than
     as a screen that has not been colour-calibrated. */
  --nes-mist:   #f3f6f6;  /* the page's ground         */
  /* 🔴 THE BLUE THAT SITS ON THAT GROUND, off the same page — and the
     value that was here before, #d5ebf3, did NOT come off that page. Its
     note claimed "apple.com/ipad-air at 1440x900, straight down its left
     edge, rgb(213,235,243) at the top of the wash", and that reading does
     not reproduce: sampled at 1440x2000 the iPad Air page is FLAT
     rgb(243,246,246) from the header to y=2000 with no wash at all, and
     sampled at a viewport tall enough to lay the whole hero out it is
     rgb(187,226,240) — this value — held essentially solid from y=0 past
     y=2100, drifting one step over two thousand pixels. Whatever 213,235,243
     was measured from, it was not this.

     🔴 SAME HUE, MORE WEIGHT, which is why the old one looked close enough
     to keep for a while:

                    hue    saturation   lightness
       #d5ebf3      196        56%         89%
       #bbe2f0      196        64%         84%     <- theirs

     Five points of lightness and eight of saturation is the whole of the
     difference, and it is the whole of why the wash read as thin.

     🔴 It is NOT --nes-sky. --nes-sky is #dbeafc: blue 18 points over
     green, a tint of the BRAND blue, and mixed down over a cool ground it
     reads as a purple cast on a grey page. This one has blue 14 over green
     — it is daylight on white hardware, not a tinted panel. The brand blue
     is still the site's; this is the light the product is photographed in. */
  /* 🔴 PULLED OFF THE REFERENCE'S OWN VALUE, and the number that moved is
     the BLUE-OVER-GREEN GAP, not the lightness. #bbe2f0 is theirs exactly
     — sampled edge to edge, x=2 through x=1437, so it is a field and not
     a photograph — and on our page it read as sky. That gap is why:

                    hue     sat    lightness   B - G
       #bbe2f0     195.8    64%       84%       14     <- theirs
       #c1e4ec     191.2    53%       84%        8

     Blue standing 14 points over green is what "sky" IS; closing it to 8
     leaves the same weight and the same brightness and moves the colour
     from sky toward daylight on glass. Lightness is untouched at 84%,
     because the weight was not the complaint.

     🔴 TWO SEPARATE DIALS, if this needs moving again. B-G gap is HOW SKY
     it reads; saturation is HOW MUCH COLOUR there is. Cutting saturation
     alone at this lightness is the move that came out DIRTY on the footer
     band (see the --ft-tint note in footer.css) — on a ground with only
     14% saturation there is nothing to take away. Close the gap first. */
  --nes-haze:   #c1e4ec;  /* the hero wash, at its deepest */
  /* 🔴 THERE WAS A --nes-mist-2 HERE, #ddeaeb, and it is worth one note
     because the analysis behind it was right and the colour was still no
     good. Measured, each band against the ground it lies on:

                          hue gap   saturation   lightness drop
       home: tan on sand      4°       ×0.76          6.1
       --nes-haze on mist    16°       ×3.89          6.5
       the rejected one       4°       ×1.81          6.5

     The ×3.89 is real: --nes-mist has almost no colour in it (S 14%) and
     --nes-haze has a lot (S 56%), so the footer's band comes out nearly
     four times as saturated as the page it lies on. On paper that is the
     "different material, not a shadow" failure, and #ddeaeb fixed it by
     matching the ground's hue and holding the same depth.

     🔴 IT CAME OUT DIRTY, and the numbers do not show it. This palette
     has been here before, one file over: the note on --ft-tint in
     footer.css threw out an ink-mixed band for the same reason — a band
     that keeps its lightness but loses its saturation does not read as
     shade, it reads as grime. Sand had room to absorb that because it is
     S 31% to begin with. Mist is S 14%, so there is nothing to take away:
     any low-saturation step down from it lands in the grey-teal that
     #ddeaeb is, and grey-teal on a cool white is the cool version of
     exactly the grubbiness footer.css already rejected.

     So the band stays --nes-haze. It is more saturated than the rule
     wants and it is the one that looks like light rather than dirt, which
     on a ground with this little colour in it is the trade that has to be
     made. If it ever needs pulling back, take it toward --nes-mist a few
     percent rather than desaturating it in place — that keeps it on the
     light side of the choice. */
  --nes-sky-up: #1e88e5;  /* headline blue on cream    */
  --nes-amber:  #fbbf24;  /* review stars              */
  --bead-blue:  #14a4fd;  /* the dot on the i, exactly as the logo draws it */
  --mark-blue:  #cde9f5;  /* the highlight bar, and the dot it becomes */

  /* ---------- ONE GROUND FOR THE WHOLE PAGE ----------
     🔴 Every section's background comes from here and nowhere else. The
     page used to run four: white on the body, --nes-sand on the intro and
     the hero, --nes-cream on the hub, #efeadd on the testimonials and the
     footer. Three of those are within a few steps of each other, which is
     the worst possible spacing — too close to read as a deliberate change
     of ground, too far to read as one surface, so every seam showed as a
     faint band and the page looked like it had been assembled rather than
     designed.

     --nes-sand is the one they collapse to, and that argument was already
     made once: see the note on GROUND in hero.js, which moved the deck off
     #efeadd because it sits 14 steps of yellow away from both its
     neighbours. This just finishes the job everywhere else.

     A section with a ground of its OWN — one that is meant to be a
     different surface, not the page continuing — does not read this. The
     privacy section is the only one: it is --nes-ink, deliberately dark,
     and it is left alone. Same for anything INSIDE a section: cards, the
     device screens in the hub, the testimonial surfaces. Those are objects
     on the ground, not the ground. */
  /* 🔴 THE SITE'S GROUND IS STILL SAND, and --nes-mist above is NOT it.
     The cool grey was briefly this line — the home page and faq.html went
     cool with it, which is precisely what the one-ground rule promises and
     precisely what was not wanted. It is scoped to .dv-body in device.css
     instead. Read that note before moving this one: the device page is
     deliberately standing on a different surface from the rest of the
     site, and that is a thing this token is otherwise designed to prevent. */
  /* 🔴 2026-09-05：**全站白底试过一轮，已退回 sand**。用户先「帮我把网页底色
     改为白色」，看过实装后「算了背景改回去哦」。

     这是白色在这个 token 上的**第二次**往返——第一次记在 device.css 那段长注释
     里（那次是 device 页单独变白，结论是「跨页过来像换了个网站」）。两次的场景
     不同：那次只动一页，这次整个 ground 一起走，所以那条推理并不适用于这次；
     这次单纯是看过之后不要。都不构成「白色永远不行」，但连续两次都退回来了，
     下次再提之前先把两处都读一遍。

     实装那一版还留下两个可复用的结论：
       · 线稿（--sketch-ink，28% ink 混 sand）在白底上**更清楚**不是更淡：对
         sand 的差值 61，对白 72。「改白会让线稿看不见」是错的直觉。
       · 换这个 token 必须连 faq-page.css 的 .faqp 一起看——它当时写死
         --nes-sand，会变成上下白、中间一块米色。现在它已经改读 --page-bg，
         这个坑填掉了。 */
  --page-bg: var(--nes-sand);

  /* semantic — the two that are really "the page's ground" now defer to it,
     so the footer's colour-mix tints keep working untouched */
  --background-color--background-primary:   var(--page-bg);
  --background-color--background-secondary: var(--nes-sky);
  --background-color--background-tertiary:  var(--page-bg);
  --text-color--text-primary:   var(--nes-ink);
  --text-color--text-secondary: var(--nes-ink);   /* type that sits on the stage */

  /* type */
  /* 🔴 ONE FACE ON THE PAGE. THAT IS THE RULE.

       Google Sans — everything. Headlines, body, and the chrome.

     🔴 IT WAS TWO, and Nunito had the chrome — the header's pills, the
     footer's micro-labels, the calendar's small type. That split is gone
     and --font--ui now resolves to the same family as the other two, so
     every reference to it keeps working and nothing has to be chased
     through six stylesheets. The token stays because "this is chrome" is
     still worth saying at the point of use, even when the answer is the
     same face; if a second family ever comes back, it comes back here.

     Nunito also came out of the <link> in index.html, device.html and
     faq.html — a family nothing renders in is a font file the reader
     downloads for nothing. If it is ever restored, restore the link with
     it, and keep the ital axis: the calendar's "Nothing on" is set in
     italic and a browser-slanted roman looks broken.

     This is labs.google's own arrangement: one sans carrying the 120px
     headline, the 24px paragraph and the nav, with nothing else invited.
     The serif is gone — --font--primary and --font--body are now the same
     family, and the only thing separating display from text is size.

     ✅ GOOGLE SANS 已经在 Google Fonts 的公开字库里，正常引用即可。

     🔴 这段原本写的是相反的话——「GOOGLE SANS IS GOOGLE'S OWN BRAND FACE…
     not a Google Fonts release and not licensed for general use… has to be
     cleared or replaced before the page goes public」，并把「Fonts API 居然
     发得出来」当成一个侥幸。2026-09-04 核实：**那个前提已经不成立**。

     实测（fonts.google.com/metadata/fonts，官方字库清单）：
       公开字库共 1946 个字族
       Google Sans        在清单里 ✓   （Google Sans Code / Flex 也在）
       DM Sans / Inter    在清单里 ✓
     它跟 DM Sans、Inter 是并列关系，不是特例。上架公开字库即按开源许可发布，
     商用免费，用法就是 index.html 里那个普通的 <link>。

     留着这段是因为旧注释**会误导人**：它会让下一个人以为上线前必须换字，
     甚至真的去换成 DM Sans。判断过期了，但结论的痕迹要留着。
     （DM Sans 仍然是个合理的备选——同样的 Product-Sans 几何骨架，也在这个
     槽位待过——只是现在没有非换不可的理由。）

     Whatever sits in --font--primary must ship a REAL ITALIC. The page
     leans on it for every second line: the hero deck, the intro slogan,
     the hub headline and its caption, the testimonial h2. A face that
     leaves the browser to slant its own roman looks broken at 60px.
     The calendar's "Nothing on" needs one too, and now reads it off the
     same family as everything else.

     The display slot has been round the houses: EB Garamond → Newsreader
     (chasing claude.ai's Copernicus) → Schibsted Grotesk → Inter Tight
     (standing in for flim.ai's RL Swizzy) → EB Garamond → Google Sans.
     Body: Figtree → Nunito → DM Sans → Google Sans. */
  --font--body:    "Google Sans", Helvetica, Arial, sans-serif;
  --font--primary: "Google Sans", Helvetica, Arial, sans-serif;   /* display */
  --font--ui:      "Google Sans", Helvetica, Arial, sans-serif;   /* chrome */
  --heading--h2:     4rem;
  --heading--h2-big: 4.6875rem;
  /* 🔴 The SECTION-OPENER size, named once because two sections now have to
     agree at every width, not just at 1440: the privacy lockup and the
     testimonials headline. It is a clamp rather than the flat token so the
     line steps down with the viewport instead of at one breakpoint — and
     because the privacy section draws its padlock off this same number
     (--pv-title), a mark that did not scale with the words beside it would
     stop being part of the line the moment the clamp moved. */
  --heading--display: clamp(28px, 4.45vw, var(--heading--h2));
  --heading--h3:     3rem;
  --heading--h4:     2rem;
  --body--regular: 1rem;
  --body--small:   .875rem;
  --body--xsmall:  .8125rem;

  /* motion (used by the ported header + hero) */
  --ease-out-expo:  cubic-bezier(.16, 1, .3, 1);
  --ease-out-quint: cubic-bezier(.22, 1, .36, 1);
  --ease-in-out:    cubic-bezier(.65, 0, .35, 1);
  /* 1-(1-t)^3 — the curve labs.google turns its hero mask on */
  --ease-out-cubic: cubic-bezier(.33, 1, .68, 1);
  /* 1-(1-t)^4 — easeOutQuart。2026-09-04 建出来的：family-fit.css 里同一条
     曲线手写了三遍（.ffit__card-img 的 hover、以及滚动揭示的 opacity 与
     transform 各一次）。值一位不改地搬过来，所以这是纯收敛、没有观感变化。
     ⚠️ 它跟 --ease-out-quint (.22,1,.36,1) 只差一点，别互相替换——那会是
     行为改动，不是收敛。 */
  --ease-out-quart: cubic-bezier(.25, 1, .5, 1);
  /* 🔴 2026-09-04：原本是 cubic-bezier(.4, 0, .1, 1)，那条曲线**名不副实**——
     P1 = (.4, 0) 让 t=0 处切线斜率为 0，起步是静止的；P2 = (.1, 1) 让收尾也平。
     两头都平 = ease-in-out，跟本文件自己的 --ease-in-out 是同一种签名。它是随
     header/hero 一起 port 进来的，**没有任何注释为这个值做过论证**。

     实测三条曲线在同一时刻走过的距离：
                                  10%     25%     50%
       旧 --ease-out (.4,0,.1,1)   2.7%   30.0%   82.5%
       --ease-out-cubic           27.1%   57.4%   87.5%
       --ease-out-expo            49.4%   82.6%   97.2%
     动画走到十分之一时旧曲线只挪了 2.7% —— 元素入场、指针落上按钮的那一刻，
     屏幕上几乎什么都没发生。36 处消费方全部继承了这个死启动。

     指向 --ease-out-cubic 而不是另造一条：它就是 easeOutCubic，而 intro.js:251
     的 settle 用的 `1 - Math.pow(1 - t, 3)` 是同一条曲线的 JS 写法。这样 CSS 与
     JS 的 "ease-out" 从此是同一个东西，也救活了一个零引用的 token。 */
  --ease-out:       var(--ease-out-cubic);

  /* 🔴 整页左右 gutter。此前 `max(5vw, 32px)` 在 9 个文件、15 处各写一遍，
     没有共享变量——想调一次页面边距就得改 15 个地方。收成一个 token 之后
     改一行即可，这一层保留。

     🔴 值 2026-09-04 试过 **30px**（照 labs.google 的固定 30px，用户要「一起
     靠边、内容变宽一点」），当天又要求「完全和线上一样」，所以退回线上的
     max(5vw, 32px)（1440 上 = 72px）。header.css 现在直接写着同一个表达式，
     两边必须一致，否则顶栏和正文会差 42px——那正是这轮改动最初要修的问题。 */
  --gutter: max(5vw, 32px);

  /* spacing */
  --section-radius--large:   3rem;
  --section-radius--regular: 2rem;     /* card corner — 3rem read too soft */
  --section-radius--tiny:    1.25rem;  /* the photo/quote block inside a card */
  --spacers--medium: 1.5rem;
}

html { -webkit-font-smoothing: antialiased; }

body {
  margin: 0;
  background-color: var(--background-color--background-primary);
  color: var(--text-color--text-primary);
  font-family: var(--font--body);
  font-size: 1rem;
  line-height: 1.3;
}

img { max-width: 100%; display: inline-block; vertical-align: middle; border: 0; }
p   { margin: 0; }

h2 {
  font-family: var(--font--primary);
  font-size: var(--heading--h2);
  letter-spacing: -.03em;
  font-weight: 400;
  line-height: .95;
  margin: 0;
}

/* ---------- utilities ---------- */
.padding-global   { padding-left: 2.5rem; padding-right: 2.5rem; }
.container-large  { width: 100%; max-width: 77.5rem; margin-inline: auto; }
.text-align-center { text-align: center; }
.text-wrap-balance { text-wrap: balance; }
.z-index-2        { z-index: 2; position: relative; }
.opacity-0-5      { opacity: .5; }

.spacer-small  { width: 100%; padding-top: 1rem; }
.spacer-medium { width: 100%; padding-top: var(--spacers--medium); }
.spacer-xhuge  { width: 100%; padding-top: 8rem; }

.text-size-regular { font-size: var(--body--regular); }
.text-size-small   { font-size: var(--body--small); }
.text-size-xsmall  { font-size: var(--body--xsmall); }
.text-weight-medium { font-weight: 500; }
.text-color-secondary { color: var(--text-color--text-secondary); }
.text-color-muted     { color: var(--nes-muted); }
.text-brand           { color: var(--nes-sky-up); }
.opacity-0-6          { opacity: .6; }
.opacity-0-7          { opacity: .7; }

.tag {
  color: var(--text-color--text-primary);
  font-size: 1.25rem;
  letter-spacing: .03em;
  text-transform: uppercase;
  font-weight: 600;
  line-height: 1.5;
}
.tag.is-small {
  font-size: var(--body--small);
  letter-spacing: .08em;
  font-weight: 500;
  line-height: 1.3;
}
.tag.is-small.text-color-secondary { color: var(--text-color--text-secondary); }

/* 🔴 --heading--h2, not --heading--h2-big. This was 75px against 64 for
   every other h2 on the page, and it is the only heading that ever wore
   the "bigger" variant — one rule, one user, sized a step above the ladder
   it belongs to. The privacy block is its opposite number and reads the
   same token now, so the two section titles are equal by construction
   rather than by two numbers kept in step by hand.

   (The 991px breakpoint below already stepped it back to --heading--h2 on
   the way down, which was the same conclusion reached for a narrow screen
   and not carried up to a wide one.) */
/* 🔴 Set to the SAME four numbers as .pv__title (privacy.css), because the
   two are the same thing on the page: a section opener over a block of
   panels. It used to differ on every one of them — 400 against 600, -.03em
   against -.02em, line-height 1 against .92, and the flat --heading--h2
   against the clamp — which is small enough to look like drift rather than
   a decision when you scroll from one to the other. Change one, change
   both; the size itself now comes from --heading--display. */
.h2-bigger {
  font-size: var(--heading--display);
  font-weight: 600;
  letter-spacing: -.02em;
  line-height: .92;
}
/* 🔴 手机上把 .92 放回 1.06。行距是**字号的比例**，但「两行之间该留多少
   空气」不是——.92 的负余量在 68px 的显示字号上读作紧凑，在手机的 28px 上
   两行就贴到一起了（用户 2026-08-30 在 What families feel 那句上说「行距
   有点太小」）。同一个比例在两个字号上不是同一个效果，所以断点在这儿。

   ⚠️ 上面那条注释要求这四个数与 .pv__title 同步——所以这条覆盖同时点了
   两个选择器，别只改一个。 */
@media screen and (max-width: 767px) {
  .h2-bigger, .pv__title { line-height: 1.06; }
}

/* 🔴 跟「This is Nestify, here to light up family life.」同尺寸
   （用户 2026-09-05：「这些标题都要和 this is nestify 一样大哦」）。
   --heading--display 的 4.45vw 是桌面列宽的斜率，到手机整段压在 28px 下限
   附近：666 宽下 29.6px 对那句的 40px，差了一整档。

   🔴 断点是 **900，不是 767**，尽管上面那条行距用的是 767。--hub-copy 定义在
   hub.css 的 900 断点里，消费者跟着它走，边界才不会各说各话：写成 767 的话
   768–900 之间 .h2-bigger 拿回退值（实测 34.18px）而同一屏的 .pv__title 和那句
   This is Nestify 都是 40px，三个「段落开场白」两个尺寸。

   line-height 一起带过来：字号在这一带涨到 40 之后，原来的 .92 会把两行压到
   一起——上面那条 1.06 只覆盖到 767，够不着这一带。

   ⚠️ 必须排在上面那个 767 块**之后**：两处特异度都是 0,1,0，媒体查询不加
   特异度，靠顺序定胜负。

   ⚠️ 这里只点 .h2-bigger。.pv__title 的 font-size 在本文件里写了也不生效：
   index.html 中 privacy.css（80 行）排在 styles.css（70 行）之后，同特异度
   后来者赢。它那一档改在 privacy.css 的 900 断点里，而且走 --pv-title 中介
   ——那边的锁图标是 `calc(--pv-title * 1.15)`，必须跟着一起长。
   （上面那条 1.06 其实也没点动 .pv__title，实测量到的一直是 .92。那句
   「别只改一个」描述的是意图，不是当时的效果。）

   回退同 privacy.css：device.html / faq.html 不加载 hub.css。 */
@media screen and (max-width: 900px) {
  .h2-bigger { font-size: var(--hub-copy, var(--heading--display)); line-height: 1.06; }
}
/* 🔴 NOTHING ON THIS PAGE LEANS. The second line of a headline used to
   turn in the italic, on the idea that the voice change carried the break;
   at display sizes it read as a different typeface dropped into the middle
   of a sentence instead (the same conclusion the hero reached on its own —
   see the note at .hero__title em).

   It is set on `em` itself rather than on each block that holds one,
   because the browser's own stylesheet is what makes an <em> lean: drop a
   block's rule and the emphasis comes back italic on the next <em> anyone
   writes. This is the one place that has to say normal. */
em { font-style: normal; font-weight: inherit; }

/* arrow link */
.text-link {
  display: flex;
  gap: .5rem;
  justify-content: flex-start;
  align-items: center;
  font-size: 1rem;
  font-weight: 600;
  text-decoration: none;
  overflow: visible;
}
.text-link-arrow-wrap {
  width: 1.25rem; min-width: 1.25rem;
  height: 1.25rem; min-height: 1.25rem;
  position: relative;
  overflow: clip;
}
.text-link-arrow {
  display: flex;
  flex-flow: column;
  justify-content: center;
  align-items: center;
  width: 100%; height: 100%;
  position: absolute; inset: 0%;
  transition: transform .35s;
}
.text-link-arrow.before { left: -100%; }
.testi_card:hover .text-link-arrow { transform: translateX(100%); }

/* ---------- section shell ---------- */
.section_new-testi {
  position: relative;
  border-radius: 0;   /* square edge against the hero */
  background-color: var(--background-color--background-tertiary);
  color: var(--text-color--text-secondary);
  overflow: clip;
}

/* deck: tall scroll track + sticky pinned stage. slider.js overwrites the
   height in px and absolutely positions every .testiv2_card. */
.testiv2_height {
  min-height: 350vh;
  position: relative;
}
.testiv2_wrap {
  display: flex;
  flex-flow: column;
  gap: 2rem;
  justify-content: center;
  align-items: center;
  min-height: 100vh;
  position: sticky;
  top: 0;
}
.testiv2_card { position: absolute; }

/* ---------- cards ---------- */
/* 🔴 The corners are cut by clip-path, NOT by overflow. This is the Safari
   black-corner fix, and it is a rendering-path change, not a taste one.

   Each card is rotated on rotateX inside a perspective, which makes it a
   composited layer. A `border-radius` + `overflow` on a DESCENDANT of that
   layer then has to be re-expressed as something the compositor can apply
   to a finished texture — and on macOS WebKit composites through CALayer,
   whose `cornerRadius` only works on a layer's OWN bounds. It cannot round
   someone else's texture that is being projected in 3D, so it falls back,
   and the region the radius removes is left as the backing store's initial
   contents: zeroed memory, which reads as opaque black.

   The tell is exact: only the card at the centre of the deck was clean, and
   that is the one whose rotAt() angle is ~0°. Every card either side of it
   — every card actually rotated — had black corners. Chrome does not do
   this because its compositor supports rounded-rect clips on a texture as a
   first-class operation.

   clip-path sidesteps the whole question: it is applied in the element's
   own local coordinate space while it PAINTS, so the rounding is baked into
   the texture before any 3D transform touches it. Nothing is asked of the
   compositor and the corners stay genuinely transparent — not filled with
   the page colour, which would only hold for as long as nothing ever passes
   behind a card.

   border-radius stays: it is what the element's own painting and any
   `border-radius: inherit` still read. */
.testi_card {
  position: relative;
  border-radius: var(--section-radius--regular);
  clip-path: inset(0 round var(--section-radius--regular));
  /* 🔴 BACK TO THE SOURCE'S OWN NUMBERS, ON REQUEST. These were 0.8× for a
     while — height, both widths, the quote and the author logo all divided
     by the same factor, on the argument that the deck read as three big
     slabs rather than as cards you could take in at a glance. The reference
     this component is lifted from (wisprflow.ai, "From the first people to
     use it" — same Webflow classes, same slider, same interaction) runs
     them full size, and full size is what was asked for.

     Taken from their stylesheet rather than eyeballed:

                            theirs      was (0.8×)
       .testi_card height     28.75rem     23rem
       .landscape width       46.25rem     37rem
       .square width          26rem        21rem
       quote font-size        --h4         --h4 × 0.8
       author logo            4.375rem     3.5rem

     🔴 AND .testi_card-img's max-height WAS ALREADY THEIRS. It sits at
     26.75rem, which is 28.75 minus this rule's 1rem padding top and bottom
     — their number, exact, and inert at the smaller height. It only starts
     doing its job again now. That is the tell that the 0.8× pass missed a
     value, and the reason to re-check anything else keyed to the height. */
  height: 28.75rem;
  min-height: 28.75rem;
  color: var(--text-color--text-primary);
  padding: 1rem;
  text-decoration: none;
}
.testi_card.landscape {
  display: grid;
  gap: 2rem;
  grid-template-rows: auto;
  grid-template-columns: 1.1fr 1fr;
  grid-auto-columns: 1fr;
  width: 46.25rem; min-width: 46.25rem; max-width: 46.25rem;
}
.testi_card.square {
  width: 26rem; min-width: 26rem; max-width: 26rem;
}

.testi_card-left {
  z-index: 2;
  display: flex;
  flex-flow: column;
  gap: 2rem;
  justify-content: space-between;
  height: 100%;
  padding: 1rem;
  position: relative;
}

.testi_quote-v2 {
  font-family: var(--font--primary);
  /* their --_text-collection---heading--h4, unscaled — the 0.8 factor came
     off with the card's (see .testi_card). Their own line-height of .95 is
     NOT copied; see the note below it. */
  /* 🔴 -2px 是用户 2026-08-30 直接指定的量（「稍微有点大 可以变小2px」），
     不是从 h4 那一档重新推的 —— 所以写成 calc 挂在 h4 上，而不是换一个 token：
     h4 本身再变的时候，这条差值跟着走。手机 24→22，桌面 32→30。 */
  font-size: calc(var(--heading--h4) - 2px);
  letter-spacing: -.03em;
  text-wrap: balance;
  font-weight: 400;
  /* Wispr sets .95 — too tight for these longer, descender-heavy quotes */
  line-height: 1.18;
}

.testi_logo { object-fit: contain; height: 1.875rem; }
.testi_logo.has-margin { margin-bottom: 1.5rem; }

/* same rounding path as .testi_card, and for the same reason — these sit
   INSIDE the rotated card, so they are descendants of the same composited
   layer and would blacken their own corners exactly the same way */
.testi_card-img-wrap {
  z-index: 2;
  border-radius: var(--section-radius--tiny);
  clip-path: inset(0 round var(--section-radius--tiny));
  width: 100%; height: 100%;
  position: relative;
}
.testi_card-img-wrap.is-first { order: -9999; }
.testi_card-img {
  object-fit: cover;
  width: 100%; height: 100%;
  max-height: 26.75rem;
}

.testi_card-author {
  display: flex;
  gap: .75rem;
  justify-content: flex-start;
  align-items: center;
}
.testi_card-author-logo {
  object-fit: cover;
  border-radius: 50%;
  /* theirs, back off the 0.8 factor with the card — see .testi_card */
  width: 4.375rem; min-width: 4.375rem;
  height: 4.375rem; min-height: 4.375rem;
  background-color: #0000000d;
}

/* five-star row, same amber as the Nestify review cards */
.testi_stars {
  display: flex;
  gap: .25rem;
  margin-bottom: 1.25rem;
}
.testi_stars svg {
  width: 1.25rem;
  height: 1.25rem;
  fill: var(--nes-amber);
}

.testi_card-bg-image {
  z-index: 0;
  object-fit: cover;
  width: 100%; height: 100%;
  position: absolute; inset: 0%;
}
.testi_card-bottom-wrap {
  display: flex;
  flex-flow: column;
  justify-content: flex-end;
  width: 100%; height: 100%;
  position: relative;
}
.testi_card-bottom-card {
  display: flex;
  flex-flow: column;
  gap: 2rem;
  justify-content: space-between;
  border-radius: var(--section-radius--tiny);
  clip-path: inset(0 round var(--section-radius--tiny));   /* see .testi_card */
  padding: 1.5rem;
  position: relative;
}
.testi_card-bottom-card-bg {
  z-index: 0;
  background-color: var(--nes-cream);
  position: absolute; inset: 0%;
}
.testi_card-bottom-card-bg.is-amber { background-color: var(--nes-amber); }
.testi_card-bottom-card-bg.is-sky   { background-color: var(--nes-sky); }

/* ---------- card backdrops — the card's OWN background, not a div ----------
   🔴 These used to be a `<div class="testi_card-bg-color is-x">` per card,
   absolutely positioned at inset: 0. That element is what blackened the
   corners in Safari, and the proof was sitting in the deck the whole time:
   of the six cards, exactly one had no such div — Priya K., whose backdrop
   is a full-bleed <img> — and that was exactly the one card whose corners
   came out clean. Five for five.

   The reason follows from how the corners go black at all (see the note on
   .testi_card). "Are these contents opaque?" is decided PER COMPOSITED
   LAYER, and the layer is .testiv2_card. An opaque div pinned to inset: 0
   covers that layer's box exactly, so the answer is yes, so the backing
   store is allocated with no alpha channel, so anything the rounded clip
   removes keeps the buffer's initial contents — black. Priya's <img> could
   not be proven opaque, her layer kept its alpha, and every rounded clip
   inside it then behaved — including the little amber card nested in it,
   which is built the square-opaque-div way and is clean regardless.

   As the card's own background-color there is no full-box opaque rectangle
   in the layer at all: .testi_card is rounded and clip-path'd, so what it
   paints is a rounded shape, and the corners stay honestly transparent.

   It is also simply less markup — one element per card, gone. */
.testi_card.is-cream { background-color: var(--nes-cream); }
.testi_card.is-sand  { background-color: var(--nes-sand); }
.testi_card.is-sky   { background-color: var(--nes-sky); }
.testi_card.is-tan   { background-color: var(--nes-tan); }
.testi_card.is-white { background-color: var(--nes-white); }
.testi_card.is-blue  { background-color: var(--nes-blue); }
.testi_card.is-ink   { background-color: var(--nes-ink); }
.testi_card.is-amber { background-color: var(--nes-amber); }

/* cards on saturated blue / ink flip to light type */
.testi_card.on-blue,
.testi_card.on-ink { color: var(--nes-cream); }

/* stats overlay on the case-study photos */
.testi_card-stats-wrap {
  display: flex;
  flex-flow: column;
  justify-content: flex-end;
  /* stronger than Wispr's #000000b3 — these photos are brighter at the bottom */
  background-image: linear-gradient(#fff0 30%, #000000cc);
  padding: 0 2rem 1.3rem;
  position: absolute; inset: 0%;
}
.testi_card-stats-grid {
  display: grid;
  gap: 1rem;
  grid-template-rows: auto;
  grid-template-columns: 1fr 1fr;
  align-items: start;
  color: var(--nes-cream);   /* sits on the dark gradient, not on the stage */
}
.testi_stat-card {
  display: flex;
  flex-flow: column;
  gap: .5rem;
  justify-content: flex-start;
  align-items: flex-start;
}
.testi_card-stat {
  font-family: var(--font--primary);
  font-size: var(--heading--h3);
  letter-spacing: -.03em;
  line-height: .95;
}

/* ============================================================
   responsive — Webflow breakpoints
   ============================================================ */
@media screen and (max-width: 991px) {
  :root {
    --heading--h2: 4rem;
    --heading--h3: 3rem;
    --heading--h4: 2rem;
  }
  .spacer-xhuge { padding-top: 6rem; }
  .testiv2_wrap { padding-top: 3rem; }
}

@media screen and (max-width: 767px) {
  :root {
    --heading--h2: 2.5rem;
    --heading--h2-big: 2.5rem;
    --heading--h3: 2rem;
    --heading--h4: 1.5rem;
    --section-radius--large: 2rem;
    --section-radius--regular: 1.25rem;
    --section-radius--tiny: 1rem;
  }
  .padding-global { padding-left: var(--gutter); padding-right: var(--gutter); }
  .spacer-xhuge { padding-top: 4rem; }

  .testiv2_wrap   { padding-top: 4rem; }
  .testiv2_height { min-height: 250vh; margin-top: 2rem; }

  .testi_card.landscape {
    display: flex;
    flex-flow: column-reverse;
    gap: .5rem;
    grid-template-columns: 1fr;
    width: 20rem; min-width: 20rem; max-width: 20rem;
    height: auto; min-height: auto;
  }
  .testi_card.square {
    width: 20rem; min-width: 20rem; max-width: 20rem;
    height: auto; min-height: 20rem;
  }
  .testi_card-left { padding-left: 0; padding-right: 0; }
  .testi_card-left.has-min-height-mobile { min-height: 20rem; }
  .testi_card-img-wrap { border-radius: 0; order: 9999; }
  .testi_card-img-wrap.is-first { order: 0; }
  .testi_card-img {
    aspect-ratio: 1;
    object-position: 50% 25%;
    border-radius: 1rem;
    width: 100%;
    max-height: none;
  }
  .testi_card-author-logo {
    width: 3rem; min-width: 3rem;
    height: 3rem; min-height: 3rem;
  }
  .testi_card-bottom-wrap { min-height: 20rem; }
  .testi_card-stats-wrap {
    border-radius: 1rem;
    width: 100%;
    padding-left: 1rem; padding-right: 1rem;
  }
  .testi_card-stat { font-size: 2.5rem; }
}

@media screen and (max-width: 479px) {
  .testiv2_wrap { padding-top: 6rem; }
}


/* ============================================================
   ONE HOVER FOR EVERY PILL ON THE PAGE

   The header's: a disc the colour of the destination opens from the
   middle of the button and swallows it, .55s on --ease-out. Any button
   anywhere gets it by taking .btn-disc and setting --btn-disc to the
   colour it should fill with.

   🔴 The LABEL HAS TO BE IN AN ELEMENT. The disc is a real child in the
   paint order, so a bare text node inside the button paints underneath
   it and the word disappears as the disc arrives. It cannot be solved
   with z-index: -1 the way an outlined button can — that puts the disc
   behind the button's own background, and a filled pill then hides it
   completely. Hence .btn-label, which is what .hd__cta-inner already
   does for the header.

   The header keeps its own .hd__fill span rather than this ::after: it
   needs the disc as a sibling of two separate label spans (the Menu
   button swaps label for "Close"), and the numbers there are the same
   .55s and 120%. If either moves, move both.
   ============================================================ */
/* ============================================================
   THE BALL THAT OPENS INTO A PILL

   One shape, two users: the intro's "Meet Nestify" and the hub's "Explore the
   Family Hub". It rests as a DISC the colour of the logo's i-dot and opens
   into a full pill — the clip-path does the opening, `scale` does the
   arrival, and the blue drains to cream a little behind the opening edge so
   the colour change is something the stretch DOES rather than a second
   event after it.

   🔴 It lives HERE, not in either section, because the two buttons have to
   be the same object. They were separate rules and had already drifted —
   the hub's was ink/cream at clamp(14px,1vw,18px) in Nunito 700 while the
   intro's was blue/cream at clamp(16px,1.32vw,21px) in Google Sans 500 —
   which is two buttons wearing one name.

   🔴 Geometry in `em` off ONE clamped font-size, so the pill scales with
   the viewport rather than with the 16px root. rem looks relative and is
   not: pinned to the root, this pill stayed 14px/45px on a 2560 screen
   while the headline above it grew.

   Opening from the LEFT is the default (the intro's); .btn-ball--mid opens
   from the middle instead. The only difference is where the closed
   clip-path sits and what the scale pivots on — and those two have to
   agree, or the ball stands in one place while it grows from another. */
.btn-ball {
  --cta-ball: 2.35em;
  position: relative;
  display: inline-flex; align-items: center;
  font-size: clamp(16px, 1.32vw, 21px);
  gap: .96em;
  height: var(--cta-ball); padding: 0 1.3em;
  border-radius: 100px;
  border: 0;
  /* the resting state IS the ball: blue, and clipped to a disc */
  background: var(--bead-blue);
  color: var(--nes-ink);
  font-family: var(--font--primary);
  /* 🔴 600, up from 500, and it is a real instance rather than a browser's
     synthesised bold: the face loads as a variable font across 400..700
     (the <link> in each page's head), so 600 is a drawn weight.

     This is the ONLY thing this button took from a round of trying to make
     it match the header's pill. The rest — one shared size, one tracking,
     caps — is gone, because they are not the same kind of object: the
     header's is chrome, set in caps at 14px, where 700 is what keeps it
     from dissolving. This is the page's primary action at nearly 19px,
     where 700 would shout. 600 is what a 19px button wants. */
  font-weight: 600; letter-spacing: -.01em;
  text-decoration: none; white-space: nowrap;
  overflow: hidden;                     /* the fill is clipped to the pill */
  isolation: isolate;
  opacity: 0;
  /* the standalone `scale` property, not `transform` — .btn-disc's hover
     fill owns a transform of its own and the two must not share a slot */
  scale: .55;
  /* 🔴 The pivot is the BALL's centre, not the pill's. Left at the default
     50% the scale pivots on the middle of the full-width pill, which is
     nowhere near the ball — measured, the ball landed at x 105 and then
     slid 23px LEFT to x 72 as it grew, so the thing that is supposed to
     stand still and open was also travelling. Half the ball's own diameter
     in from the left edge is the one point that does not move. */
  transform-origin: calc(var(--cta-ball) / 2) center;
  clip-path: inset(0 calc(100% - var(--cta-ball)) 0 0 round 100px);
  /* three beats that overlap rather than queue */
  transition:
    opacity .26s ease,
    scale .44s var(--ease-out-expo),
    clip-path .64s var(--ease-out-expo) .2s,
    background-color .44s var(--ease-out) .26s,
    border-color .35s var(--ease-out),
    color .3s var(--ease-out),
    /* 按压：见文件末尾那段 */
    transform .14s var(--ease-out-quint);
}

/* opens from the MIDDLE: the closed shape is a ball on the centre line, so
   the pivot has to be the centre too */
.btn-ball--mid {
  transform-origin: center;
  clip-path: inset(0 calc(50% - var(--cta-ball) / 2) round 100px);
}

/* the open state, whatever the section calls its cue */
.btn-ball.is-open {
  opacity: 1;
  scale: 1;
  background: var(--nes-cream);
  clip-path: inset(0 0 0 0 round 100px);
}

.btn-disc { position: relative; overflow: hidden; }
.btn-disc::after {
  content: "";
  position: absolute; z-index: 0;
  /* 🔴 WHERE THE POINTER CAME IN, written by disc.js. The 50% fallback is
     the old behaviour — dead centre — so a page that never loads that file
     keeps a working hover instead of losing one. */
  left: var(--disc-x, 50%); top: var(--disc-y, 50%);
  /* 🔴 240%, AND THE DOUBLING IS EXACTLY WHAT MOVING THE ORIGIN COSTS.
     From the centre, a circle covers the button once its radius clears the
     half-diagonal: width% >= sqrt(1 + (h/w)^2), which is 120% for every
     pill on the site and is where the old number came from. From an
     ARBITRARY point the worst case is a corner, and the far corner is a
     full diagonal away rather than half — so the same coverage costs
     exactly twice the diameter.

     240% therefore holds the same rule the old note stated: it covers
     while h <= 0.66w. Measured, the widest and the tightest —
       JOIN THE WAITLIST  212x56   needs 219, has 254
       MENU                88x45   needs  99, has 106   <- the tight one
     A squarer button than 0.66 needs more; re-check these if one appears. */
  width: 240%; aspect-ratio: 1; height: auto;
  border-radius: 50%;
  background: var(--btn-disc, var(--nes-blue-2));
  transform: translate(-50%, -50%) scale(0);
  /* 🔴 .55s → .22s（2026-09-04）。这是全站唯一的共用按钮 hover，被触发次数最多，
     而半秒的墨盘扫过意味着指针落上去要等大半拍才有回应——叠加上面那条曲线原本
     的零初速，前 60ms 几乎完全静止。
     ⚠️ styles.css:761 记着「这里和 header 的数字必须一致」，header.css 的
     .hd__fill 已同步改。那条注释约束的是两处同步，不是时长本身。 */
  transition: transform .22s var(--ease-out);
  pointer-events: none;
}
.btn-disc:hover::after { transform: translate(-50%, -50%) scale(1); }
.btn-label { position: relative; z-index: 1; display: inline-flex; align-items: center; gap: .5em; }

/* The arrow that follows a label. It is the page's ONE arrow drawing — the
   same path the header's pills use — turned 45° so it points up and out of
   the corner rather than along the line.

   🔴 EVERY BUTTON THAT HAS IT POINTS THE SAME WAY, which is why the turn is
   a constant and not a variable. It was briefly both, so the intro's CTA
   could aim down at the page instead of out of it — a scroll cue rather
   than a link. The arrow went; the variables had nothing left to vary, and
   a knob nobody turns is just a place for the next person to introduce a
   second arrow direction. If one is ever genuinely needed, note that the
   turn and the hover nudge below are ONE decision: an arrow aimed down that
   leaps up on hover is two controls arguing.

   🔴 `rotate`, not `transform: rotate()`. They are separate properties, and
   the individual ones compose as translate → rotate → scale: so the hover
   nudge below still travels up-and-right in the button's own axes. Fold
   both into one `transform` and the translate inherits the 45°, which
   sends the arrow off along its own shaft instead.

   Sized in `em`, so it tracks whatever the button's clamped font-size
   resolves to at that viewport. */
.btn-arrow {
  display: inline-flex; flex: none;
  width: 1em; height: 1em;
  rotate: -45deg;
  transition: translate .35s var(--ease-out);
}
.btn-arrow svg { width: 100%; height: 100%; display: block; }
.btn-disc:hover .btn-arrow { translate: .16em -.16em; }

/* ============================================================
   the lock, and the shackle that swings shut

   🔴 IT IS HERE, NOT IN privacy.css, because two pages draw it now: the
   home page's privacy heading and the device page's privacy card. It came
   from privacy.css and this is the whole of what moved — the geometry, the
   hinge and the keyframes. What did NOT move is the TRIGGER, which is
   genuinely per-page: the home page gates on .pv--anim (privacy.js) plus
   its own .is-in, the device page on the card's [data-reveal] .is-in. Each
   owns one line; neither owns a copy of the animation.

   The alternative was to duplicate 30 lines into device.css, and the
   numbers below are exactly the kind that get tuned in one copy and not
   the other.

   🔴 The keyhole is CUT out of the case with a mask, not painted over it
   in the ground's colour — see the markup. A hole has to stay a hole
   whatever ends up behind the lock, and on the device page what is behind
   it is a white card rather than the home page's cream.

   🔴 The bar is drawn UNDER the body in the markup (source order), so as
   it drops, the last of it disappears behind the case rather than crossing
   it. Keep that order in any markup that reuses this.

   ---------- the shackle ----------
   🔴 A TRUE SWIVEL, THROUGH A FULL 180°. The hinge is the fixed leg's own
   vertical axis — rotateY — which is what a real padlock does, and the
   travel is the whole of it: the shackle starts swung right round to the
   far side and comes back.

   That is why 180 and not 90. The two ends of the swing are the SAME arch
   mirrored about the leg they share: an inverted U reaching left at 180°,
   an inverted U reaching right at 0°, with the fixed leg common to both.
   Halfway, at 90°, it is edge-on and projects to a line — the flat state.
   So one rotation reads as three things in order: out on the far side,
   flat, home. A 90° version only ever showed the second half.

   (An in-plane rotate was tried and dropped: it TILTS the whole U over
   like a bin lid, which no shackle does. The foreshortening is the tell
   that the thing is turning about its own leg rather than falling over.)

   perspective() has to sit INSIDE the transform and BEFORE the rotate, or
   the Y rotation projects flat and the U merely narrows — the near leg
   never grows, and without that there is no depth cue at all. 40 is in
   user units against a 24-unit box.

   Two beats: swing home, then press down. Composed in that order below —
   CSS applies transforms right to left, so the rotation runs first, about
   the hinge, and `translateY` then carries the squared-up shackle into the
   case.

   🔴 2.6, between the two extremes this went through. At 3.8 the shackle
   floated well clear and the last beat read as a separate part landing on
   a lock; at 1.3 it never left the rim at all and the press was 4px, which
   is a twitch rather than a movement.

   The number that divides them is 1.65, and it is arithmetic rather than
   taste: the legs end at y=11, the 2.1 stroke's round cap paints 1.05 past
   that to y=12.05, and the case's top edge is at 10.4 — so 1.65 is exactly
   the lift at which the painted end clears the rim. 2.6 sits just under a
   unit above it: a real gap you can read as "open", still close enough
   that the shackle is plainly sitting in its own lock rather than hovering
   over one, and an 8px press to close it.

   🔴 transform-origin: 0 50% with transform-box: fill-box IS that hinge.
   The path's own box runs x 8→16, so its left edge is the fixed leg — and
   for a Y rotation only the x of the origin matters. Without fill-box the
   origin resolves against the whole SVG viewport and the thing swings
   around a point off in the corner.
   ============================================================ */
.pv__lock {
  flex: none;
  display: block;
  /* 🔴 A VARIABLE WITH A DEFAULT, because the two callers size it two
     different ways and neither should have to override width and height
     separately. The home page ties it to its section title; the device
     page gives it a flat size in its own card. The fallback is the em so
     that a third caller can just drop it next to type and get something
     sensible. */
  width: var(--lock-size, 1.15em);
  height: var(--lock-size, 1.15em);
  /* it sits on the type's baseline-ish line, not hanging below it */
  align-self: center;
  color: var(--nes-ink);
}
.pv__lock svg { display: block; width: 100%; height: 100%; overflow: visible; }
.pv__lock-bar {
  transform-box: fill-box;
  transform-origin: 0 50%;                  /* the fixed leg's axis — the hinge */
}
/* 🔴 One keyframe track, because the two beats have to be SEQUENTIAL and a
   transition cannot express "finish turning, then start dropping" on a
   single property. 0→62% is the swivel; 62→100% is the press. The shackle
   is rigid: it cannot be square and swung at once, and it cannot drop until
   it is square, so the two must never overlap. */
@keyframes pv-latch {
  0%   { transform: translateY(-2.6px) perspective(40px) rotateY(-180deg); }
  62%  { transform: translateY(-2.6px) perspective(40px) rotateY(0deg); }
  100% { transform: translateY(0)      perspective(40px) rotateY(0deg); }
}


/* ---------- Lenis (see smooth.js) ----------
   The library's own required rules, inlined rather than pulled as a second
   CDN request. Nothing here is a style choice: `height: auto` stops a
   100% html/body from capping the scrollable length, and the
   scroll-behavior override is because a native `smooth` and Lenis animating
   the same scrollTop at the same time fight each other every frame. */
html.lenis, html.lenis body { height: auto; }
.lenis.lenis-smooth { scroll-behavior: auto !important; }
.lenis.lenis-smooth [data-lenis-prevent] { overscroll-behavior: contain; }
.lenis.lenis-stopped { overflow: hidden; }
.lenis.lenis-smooth iframe { pointer-events: none; }

/* ============================================================
   🔴 2026-09-04 —— 按压反馈（触屏上唯一的反馈）

   本站的主 CTA（.btn-ball + .btn-disc，三处：intro / hub / device）、header
   的 Device 药丸、footer 的商店徽章、hero 的翻页点，此前**全都只有 hover**。
   全站 `:active` 只在 device.css 出现过——而那个文件自己写着（device.css:277）：
     「每个可按的东西都要回应一次按压，不能只有 hover……在触摸屏上等于没有反馈」
   同一个仓库里已经有这条标准，只是没推广到别的页面。这里补齐。

   数值照抄 device.css 已论证过的那套：scale(.97) / 140ms / --ease-out-quint
   （device.css:275-287 论证了幅度与时长，并说明它**故意在 reduced-motion 下
   保留**——「这么小这么短的一次缩放没有前庭负担」）。

   ⚠️ .btn-ball 用的是独立的 `scale` 属性（见它自己那段注释：「transform 这个
   槽位要留给别人」），所以按压走 transform，两者互不占位，也不会打架。
   ============================================================ */
.btn-ball:active,
.hd__btn:active,
.ft__store:active,
.hero__dot:active { transform: scale(.97); }

/* ============================================================
   🔴 2026-09-04 —— prefers-reduced-motion（此前**整个文件一条都没有**）

   这个文件拥有全站共用的按钮原语，每一页都在用，而它们的动效全是大位移：
     · .btn-disc::after —— 一个 width: 240% 的圆盘从 scale(0) 扫过整个按钮
     · .btn-disc:hover .btn-arrow —— 箭头位移
     · .btn-ball —— 球开成药丸：scale .44s + clip-path .64s 的六轨形变
     · .testi_card:hover .text-link-arrow —— translateX(100%)
   证据表明这个缺口是已知的：intro.css:552 单独给 .intro__cta 打了补丁，
   说明有人在**调用点**发现过它，但没有回到定义处修。

   保留什么、去掉什么，照 device.css:3218 那套已经论证过的口径：
   不做 `* { transition: none }` 一刀切，只砍**位移**，留下 opacity/颜色。

   ⚠️ 按压反馈（scale .97）**故意留着**——device.css:275-286 论证过：
   「这么小这么短的一次缩放没有前庭负担」，而它在触屏上是唯一的反馈。
   ============================================================ */
@media (prefers-reduced-motion: reduce) {
  /* 墨盘不再扫过，直接以最终色出现 */
  .btn-disc::after { transition: opacity .2s ease; transform: translate(-50%, -50%) scale(1); opacity: 0; }
  .btn-disc:hover::after { opacity: 1; }
  /* 箭头不再窜动 */
  .btn-disc:hover .btn-arrow { translate: 0 0; }
  .btn-arrow { transition: none; }
  /* 球→药丸：形变去掉，只留淡入 */
  .btn-ball { transition: opacity .2s ease, background-color .2s ease, color .2s ease, transform .14s var(--ease-out-quint); }
  .testi_card:hover .text-link-arrow { transform: none; }
}

/* One mobile text gutter for the header, intro and section headings. */
@media (max-width: 767px) {
  :root { --gutter: 1.25rem; }
}
