Reading Map — <mv-reading-map>

node scripts/add.mjs reading-map --out ./src/marvelous

<div id="rm-demo" style="width:min(100%,56rem);margin-inline:auto">
  <style>
    #rm-demo .rm-shell { display:flex; gap:1rem; align-items:stretch; height:34rem }
    #rm-demo .rm-scroll { flex:1; min-width:0; overflow:auto; border:1px solid var(--mv-border); border-radius:var(--mv-radius-xl); background:var(--mv-surface); scroll-behavior:auto; overscroll-behavior:contain }
    #rm-demo article { max-width:38rem; margin:0 auto; padding:2rem 1.75rem 3rem; font-size:.9375rem; line-height:1.7; color:var(--mv-fg) }
    #rm-demo .rm-kicker { margin:0 0 .5rem; color:var(--mv-accent-fg); font-size:.75rem; font-weight:600; letter-spacing:.06em; text-transform:uppercase }
    #rm-demo h1 { margin:0 0 .75rem; font-size:1.75rem; line-height:1.2; letter-spacing:-.02em }
    #rm-demo .rm-lead { margin:0 0 1.5rem; color:var(--mv-fg-muted); font-size:1.0625rem }
    #rm-demo .rm-byline { display:flex; gap:.5rem; align-items:center; margin:0 0 2rem; color:var(--mv-fg-muted); font-size:.8125rem }
    #rm-demo h2 { margin:2.25rem 0 .75rem; font-size:1.25rem; line-height:1.3; letter-spacing:-.01em; scroll-margin-top:1rem }
    #rm-demo h3 { margin:1.5rem 0 .5rem; font-size:1rem }
    #rm-demo p, #rm-demo ul, #rm-demo ol { margin:0 0 1rem }
    #rm-demo li { margin:.25rem 0 }
    #rm-demo pre { margin:0 0 1.25rem; padding:1rem 1.125rem; overflow:auto; border-radius:var(--mv-radius-lg); background:light-dark(var(--mv-gray-950), var(--mv-gray-900)); color:var(--mv-gray-100); font:.8125rem/1.6 var(--mv-font-mono) }
    #rm-demo code { font-family:var(--mv-font-mono); font-size:.85em }
    #rm-demo blockquote { margin:1.25rem 0; padding:.25rem 0 .25rem 1rem; border-left:3px solid var(--mv-accent); color:var(--mv-fg-muted); font-style:italic }
    #rm-demo figure { margin:1.5rem 0 }
    #rm-demo .rm-art { aspect-ratio:16/7; border-radius:var(--mv-radius-lg); background:
        linear-gradient(180deg, transparent 60%, color-mix(in oklab, var(--mv-accent) 30%, transparent)),
        radial-gradient(120% 90% at 20% 10%, color-mix(in oklab, var(--mv-info) 45%, transparent), transparent 60%),
        linear-gradient(135deg, color-mix(in oklab, var(--mv-accent) 55%, var(--mv-bg)), color-mix(in oklab, var(--mv-warning) 40%, var(--mv-bg))) }
    #rm-demo .rm-chart { aspect-ratio:16/6; display:flex; align-items:flex-end; gap:6%; padding:1rem 8% 0; border-radius:var(--mv-radius-lg); background:var(--mv-bg-muted) }
    #rm-demo .rm-chart span { flex:1; border-radius:6px 6px 0 0; background:var(--mv-accent); opacity:.85 }
    #rm-demo figcaption { margin-top:.5rem; color:var(--mv-fg-muted); font-size:.8125rem }
    #rm-demo table { width:100%; margin:0 0 1.25rem; border-collapse:collapse; font-size:.8125rem }
    #rm-demo th, #rm-demo td { padding:.45rem .5rem; border-bottom:1px solid var(--mv-border); text-align:left }
    #rm-demo td:last-child, #rm-demo th:last-child { text-align:right; font-variant-numeric:tabular-nums }
    #rm-demo mv-reading-map { --mv-reading-map-height: 100% }
    @media (max-width:40rem) { #rm-demo mv-reading-map { --mv-reading-map-width: 4.25rem } #rm-demo article { padding:1.5rem 1.25rem 2.5rem } }
  </style>

  <div class="rm-shell">
    <div class="rm-scroll" id="rm-scroll" tabindex="0" aria-label="Article">
      <article id="rm-article">
        <p class="rm-kicker">Engineering blog · Performance</p>
        <h1>How we made our store load three times faster</h1>
        <p class="rm-lead">Six weeks, no redesign, and a mobile Largest Contentful Paint down from 4.1 s to 1.3 s. Here’s what actually mattered — and what didn’t help at all.</p>
        <p class="rm-byline"><span>By Sofia Martinez, Front-end team</span><span aria-hidden="true">·</span><span>12 min read</span></p>

        <h2 id="rm-constat">Where we started</h2>
        <p>In March, our mobile conversion rate dropped for the third month in a row. User tests all told the same story: the product page “takes a while to make up its mind.” The hero image showed up last, buttons shifted right as you tapped them, and the cart responded half a second late.</p>
        <p>Our dashboards, meanwhile, were all green. They were measuring a MacBook on the office fiber line. Our customers were shopping on the subway, on mid-range phones.</p>
        <figure>
          <div class="rm-chart" role="img" aria-label="Load time per week, down from 4.1 to 1.3 seconds"><span style="height:92%"></span><span style="height:80%"></span><span style="height:61%"></span><span style="height:44%"></span><span style="height:33%"></span><span style="height:29%"></span></div>
          <figcaption>Median mobile LCP, week over week (field data, 75th percentile).</figcaption>
        </figure>

        <h2 id="rm-mesurer">Measure before you optimize</h2>
        <p>First decision: no more optimizing without field data. We hooked the <code>web-vitals</code> library up to 10% of traffic and sent the results to our data warehouse, broken down by phone model, page type and network quality.</p>
        <pre><code>import { onLCP, onINP, onCLS } from "web-vitals";

const send = (metric) =&gt; {
  navigator.sendBeacon("/rum", JSON.stringify({
    name: metric.name,
    value: Math.round(metric.value),
    page: document.body.dataset.template,
  }));
};

onLCP(send);
onINP(send);
onCLS(send);</code></pre>
        <p>Within a week, the picture was crystal clear. Three items explained most of the delay:</p>
        <ul>
          <li>product images served at 2,400 pixels to 390-pixel screens;</li>
          <li>180 KB of JavaScript executed before first paint;</li>
          <li>a CDN cache that expired every five minutes, for no good reason.</li>
        </ul>

        <h2 id="rm-images">Images, the biggest win</h2>
        <p>Our photographers deliver gorgeous files, and we were serving them as is. Switching to responsive images, in AVIF with a WebP fallback, was the simplest and most profitable project of the quarter.</p>
        <figure>
          <div class="rm-art" role="img" aria-label="Product shot: a ceramic lamp on a pastel background"></div>
          <figcaption>The hero image now weighs 38 KB instead of 610 KB, with no visible loss.</figcaption>
        </figure>
        <p>We also reserved space for every image with <code>width</code> and <code>height</code> attributes. No more buttons jumping under your thumb: CLS went from 0.21 to 0.02.</p>
        <p>One last, often forgotten detail: the hero image is the only one loaded with high priority. All the others wait patiently until they get close to the viewport.</p>

        <h2 id="rm-js">Less JavaScript</h2>
        <p>Most of our JavaScript powered features almost nobody used on load: product zoom, the engraving configurator, the chat. We now load them on first interaction.</p>
        <pre><code>button.addEventListener("pointerenter", async () =&gt; {
  const { openZoom } = await import("./zoom.js");
  button.onclick = openZoom;
}, { once: true });</code></pre>
        <blockquote>“The fastest code is the code you don’t ship. The second fastest is the code you ship later.”</blockquote>
        <p>The result: 112 KB less on the critical path, and INP under 200 ms for 80% of sessions.</p>

        <h2 id="rm-cache">Caching, finally done right</h2>
        <p>Our product pages rarely change more than once a day. Yet the CDN treated them as stale after five minutes. We adopted <code>stale-while-revalidate</code> and a targeted purge on every catalog update.</p>
        <table>
          <thead><tr><th>Page</th><th>Cache hit rate before</th><th>After</th></tr></thead>
          <tbody>
            <tr><td>Product page</td><td>41%</td><td>96%</td></tr>
            <tr><td>Category</td><td>55%</td><td>94%</td></tr>
            <tr><td>Home</td><td>72%</td><td>99%</td></tr>
          </tbody>
        </table>

        <h2 id="rm-resultats">Results</h2>
        <p>Six weeks later, median mobile LCP was at 1.3 s, INP at 160 ms and CLS close to zero. Above all, mobile conversion rose 11%, and the bounce rate on product pages fell by a quarter.</p>
        <p>None of these projects was spectacular. None required rewriting the app. They were simply tackled in the order the data dictated.</p>

        <h2 id="rm-suite">What’s next?</h2>
        <p>Performance isn’t a project, it’s a habit. Every pull request now shows its impact on page weight, and a budget blocks anything over 10 KB added to the critical path. Next up: fonts, which still account for 90 KB at first paint.</p>
        <p>Questions, or a story of your own? Write to us — we’ll publish the best replies in an upcoming post.</p>
      </article>
    </div>
    <mv-reading-map id="rm-map" target="#rm-article" scroller="#rm-scroll" label="Article reading map"></mv-reading-map>
  </div>

  <script type="module">
    const map = document.getElementById("rm-map");
    const box = document.getElementById("rm-scroll");
    await customElements.whenDefined("mv-reading-map");
    // First visit: pretend a previous reading session (the map remembers real ones per URL).
    requestAnimationFrame(() => {
      if (Object.keys(map.dwell).length) return;
      const dwell = {};
      const give = (key, seconds, n = 24) => { for (let i = 0; i < n; i++) dwell[`${key}:${i}`] = seconds; };
      give("intro", 999);
      give("rm-constat", 999);
      give("rm-mesurer", 999, 2);
      for (let i = 3; i < 8; i++) dwell[`rm-mesurer:${i}`] = 1.4;
      give("rm-cache", 2.5, 3);
      map.dwell = dwell;
      box.scrollTop += document.getElementById("rm-mesurer").getBoundingClientRect().top - box.getBoundingClientRect().top - 24;
    });
  </script>
</div>