diff --git a/.claude/skills/release/SKILL.md b/.claude/skills/release/SKILL.md
index 75178fd..7799e76 100644
--- a/.claude/skills/release/SKILL.md
+++ b/.claude/skills/release/SKILL.md
@@ -87,6 +87,43 @@ quantization is what keeps that a non-issue. Only CSS `background-image` would a
 not worth trading `<img alt>` on the page's content images to save one
 language's worth of cached bytes.
 
+**The site is registered with Google Search Console, and the SEO is three
+files' worth of plumbing rather than copywriting.** The landing page went a
+month without appearing in a search, and the first suspicion — that something
+was blocking crawlers — was wrong: there is no `robots.txt` anywhere, no
+`noindex`, no `X-Robots-Tag`, and the repository is public. It had simply never
+been registered, and one thing about it was genuinely broken. **A crawler
+renders the page, so it negotiates like a browser** — Googlebot's
+`navigator.languages` is `en-US` — and the Japanese `<section>` therefore goes
+`display:none` on the one URL anyone links to. The Japanese page was not
+ranking badly; it did not exist. The fix keeps every decision above intact:
+hreflang alternates naming `?lang=en`, `?lang=ja` and the bare URL as
+`x-default`, real `<a href="?lang=…">` links in both footers because a
+`<select>` is not something a crawler can follow, and a `data-description` on
+each section beside the `data-title` that was already there — so adding a
+language is still a code, an `<option>` and a translated `<section>`.
+**There is deliberately no `rel=canonical`.** It would have to differ per URL,
+and one file cannot say three things: a static one pointing anywhere folds the
+other two into it and un-indexes the Japanese page, while a script-written one
+is the case Google says not to rely on. hreflang carries the relationship, and
+that is what it is for. The `?lang=` spelling is not free to change either —
+it is what `appstore/metadata*/*/marketing_url.txt` and `privacy_url.txt`
+already point at, so a move to `/ja/` paths would mean re-pushing both
+listings.
+`site/sitemap.xml` lists six URLs rather than two, because an hreflang set has
+to name every variant from every variant. It carries no `lastmod`: the site is
+deployed by hand at release time, so the date would go stale the first time
+someone forgot it, and Google ignores a lastmod it cannot trust. **It is
+submitted in Search Console, not announced from a `robots.txt`** — a
+`robots.txt` is only read at the *host* root, `temoki.github.io/robots.txt`,
+which belongs to no repository here; one placed in `site/` would be served,
+ignored, and mistaken for working. A sitemap has no such rule and may sit at
+any level at or above the URLs it lists.
+The JSON-LD `SoftwareApplication` block is the weakest of these and is kept
+honest on purpose: every value in it is checkable against the App Store
+listing, `LICENSE`, or the specs table on the page itself, and there is no
+`aggregateRating` — nor may one ever be added that nobody left.
+
 **The landing page is written for parents and teachers**, not for the kid and
 not for a developer — the technical account lives in the README. Three things
 about it are decisions rather than taste. The App Store badge is Apple's own
diff --git a/site/index.html b/site/index.html
index 1e9eb86..1beb6c1 100644
--- a/site/index.html
+++ b/site/index.html
@@ -7,7 +7,23 @@
 <meta name="description" content="Tortoise Blocks is a visual programming app for kids on iPad, Mac and Apple Vision Pro: snap blocks together, press play, and watch the tortoise draw. Tortoise Blocks は、ブロックをならべてカメに絵をかいてもらう、iPad と Mac と Apple Vision Pro のための子ども向けプログラミングアプリです。">
 <link rel="icon" href="icon-176.png">
 
-<!-- One card serves both languages, because both are served from one URL. -->
+<!-- Each language is a URL, so each language is a page a search engine can
+     index on its own. The set is identical on every variant, self-reference
+     included, which is exactly what lets it be static in one file that serves
+     all three; `x-default` is the bare URL because that is the one that
+     negotiates. There is deliberately no `rel=canonical`: it would have to
+     differ per URL, and one file cannot say three things — a static one
+     pointing anywhere folds the other two into it and un-indexes the Japanese
+     page, while a script-written one is the case Google says not to rely on.
+     hreflang carries the relationship instead, which is what it is for. -->
+<link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/?lang=en">
+<link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/?lang=ja">
+<link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/">
+
+<!-- One card for both languages, and English by default. The alternates
+     above give each language a URL, but a social scraper is not a browser: it
+     does not run the script that picks a language, so a per-language card
+     would have to be written by JavaScript nothing would execute. -->
 <meta property="og:type" content="website">
 <meta property="og:site_name" content="Tortoise Blocks">
 <meta property="og:title" content="Tortoise Blocks">
@@ -19,6 +35,49 @@
 <meta property="og:image:alt" content="The Tortoise Blocks tortoise, in a purple beret with a paintbrush for a tail.">
 <meta name="twitter:card" content="summary_large_image">
 
+<!-- What the page says about the app, in a form a search engine reads without
+     guessing. Every value here is checkable against something: the price, the
+     age rating and the seller come from the App Store listing, the license
+     from LICENSE, the requirements from the specs table below. There is no
+     aggregateRating and there must never be one that nobody left.
+
+     It is written once and in English rather than per language, because these
+     are facts about the app rather than prose — and because a script-written
+     one would arrive after the crawl for the same reason the OG card is
+     single. -->
+<script type="application/ld+json">
+{
+  "@context": "https://schema.org",
+  "@type": "SoftwareApplication",
+  "name": "Tortoise Blocks",
+  "description": "A visual programming app for kids on iPad, Mac and Apple Vision Pro. Snap blocks into a program, press play, and watch a tortoise draw your picture.",
+  "url": "https://temoki.github.io/TortoiseBlocks/",
+  "applicationCategory": "EducationalApplication",
+  "operatingSystem": "iPadOS 26 or later, macOS 26 or later, visionOS 26 or later",
+  "downloadUrl": "https://apps.apple.com/app/id6798677334",
+  "installUrl": "https://apps.apple.com/app/id6798677334",
+  "screenshot": "https://temoki.github.io/TortoiseBlocks/shots/ipad-en-1.png",
+  "inLanguage": [
+    "en",
+    "ja"
+  ],
+  "isFamilyFriendly": true,
+  "author": {
+    "@type": "Person",
+    "name": "Tomoki Kobayashi"
+  },
+  "license": "https://opensource.org/license/mit",
+  "isAccessibleForFree": true,
+  "offers": {
+    "@type": "Offer",
+    "price": "0",
+    "priceCurrency": "USD"
+  },
+  "privacyPolicy": "https://temoki.github.io/TortoiseBlocks/privacy.html",
+  "codeRepository": "https://github.com/temoki/TortoiseBlocks"
+}
+</script>
+
 <style>
   /* The palette is the app's own: the accent is the mascot's purple, the same
      literal both appearances use in the app. Kept in step with privacy.html. */
@@ -239,7 +298,9 @@
 <main>
 
 <!-- ============================ English ============================= -->
-<section lang="en" data-title="Tortoise Blocks — Snap blocks together, watch the tortoise draw">
+<section lang="en"
+         data-title="Tortoise Blocks — Snap blocks together, watch the tortoise draw"
+         data-description="A visual programming app for kids on iPad, Mac and Apple Vision Pro. Snap blocks into a program, press play, and watch a tortoise draw your picture.">
 
   <div class="hero">
     <img class="icon" src="icon-256.png" alt="" width="112" height="112">
@@ -451,7 +512,9 @@ <h2>Quiet about the child using it</h2>
 </section>
 
 <!-- ============================ 日本語 ============================== -->
-<section lang="ja" data-title="Tortoise Blocks — ブロックをならべて、カメに絵をかいてもらおう">
+<section lang="ja"
+         data-title="Tortoise Blocks — ブロックをならべて、カメに絵をかいてもらおう"
+         data-description="iPad と Mac と Apple Vision Pro のための子ども向けビジュアルプログラミングアプリ。ブロックをならべて再生ボタンを押すと、カメが1本ずつ線をかいていきます。">
 
   <div class="hero">
     <img class="icon" src="icon-256.png" alt="" width="112" height="112">
@@ -608,6 +671,11 @@ <h2>使う子どものことを、何も集めません</h2>
   <a href="https://github.com/temoki/TortoiseBlocks">GitHub</a>
   <a href="privacy.html">Privacy / プライバシー</a>
   <a href="https://github.com/temoki/TortoiseBlocks/issues">Issues</a>
+  <!-- Real links, and not a second picker: a <select> is not something a
+       crawler can follow, so without these the Japanese URL is declared in the
+       head and reachable from nowhere. -->
+  <a href="?lang=en" hreflang="en" lang="en">English</a>
+  <a href="?lang=ja" hreflang="ja" lang="ja">日本語</a>
   <span>© hiraku.space</span>
 </footer>
 
@@ -624,6 +692,13 @@ <h2>使う子どものことを、何も集めません</h2>
       // so a new language needs no edit here.
       var section = document.querySelector('section[lang="' + lang + '"]');
       if (section && section.dataset.title) document.title = section.dataset.title;
+      // The static description names both languages, because that is what the
+      // bare URL and a script-less reader get. Once a language is resolved the
+      // page is only in that one, and the description should say so.
+      var desc = document.querySelector('meta[name="description"]');
+      if (desc && section && section.dataset.description) {
+        desc.setAttribute("content", section.dataset.description);
+      }
       if (remember) {
         // The choice rides in the URL, not in storage: this site promises the
         // app stores nothing, and the page has no business doing otherwise —
diff --git a/site/privacy.html b/site/privacy.html
index c87675f..a6b89b3 100644
--- a/site/privacy.html
+++ b/site/privacy.html
@@ -7,8 +7,22 @@
 <meta name="description" content="Tortoise Blocks collects no data and makes no network connections. Tortoise Blocks は利用者の情報を一切収集せず、ネットワーク通信も行いません。">
 <link rel="icon" href="icon-176.png">
 
-<!-- The same card as the landing page, and English for the same reason: both
-     languages of this page live at one URL, so a preview can only be one. -->
+<!-- Each language is a URL, so each language is a page a search engine can
+     index on its own. The set is identical on every variant, self-reference
+     included, which is exactly what lets it be static in one file that serves
+     all three; `x-default` is the bare URL because that is the one that
+     negotiates. There is deliberately no `rel=canonical`: it would have to
+     differ per URL, and one file cannot say three things — a static one
+     pointing anywhere folds the other two into it and un-indexes the Japanese
+     page, while a script-written one is the case Google says not to rely on.
+     hreflang carries the relationship instead, which is what it is for. -->
+<link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=en">
+<link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=ja">
+<link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/privacy.html">
+
+<!-- The same card as the landing page, and English for the same reason: a
+     social scraper does not run the script that picks a language, so the
+     per-language URLs declared above buy a preview nothing. -->
 <meta property="og:type" content="article">
 <meta property="og:site_name" content="Tortoise Blocks">
 <meta property="og:title" content="Privacy Policy — Tortoise Blocks">
@@ -150,7 +164,8 @@ <h1>Tortoise&nbsp;Blocks</h1>
   </div>
 </header>
 
-<section lang="en">
+<section lang="en"
+         data-description="Tortoise Blocks collects no information about the people who use it: no accounts, no analytics, no advertising, and no network connections at all.">
 
   <h2 style="border-top:0;padding-top:0;margin-top:2.5rem">Privacy Policy</h2>
 
@@ -248,7 +263,8 @@ <h2>8. Contact</h2>
 
 </section>
 
-<section lang="ja">
+<section lang="ja"
+         data-description="Tortoise Blocks は利用者の情報を一切収集しません。アカウントも、解析も、広告も、ネットワーク通信もありません。">
 
   <h2 style="border-top:0;padding-top:0;margin-top:2.5rem">プライバシーポリシー</h2>
 
@@ -327,7 +343,9 @@ <h2>8. お問い合わせ</h2>
 
 <footer>
   <a href="./">Tortoise Blocks</a> ·
-  <a href="https://github.com/temoki/TortoiseBlocks">GitHub</a>
+  <a href="https://github.com/temoki/TortoiseBlocks">GitHub</a> ·
+  <a href="?lang=en" hreflang="en" lang="en">English</a> ·
+  <a href="?lang=ja" hreflang="ja" lang="ja">日本語</a>
 </footer>
 
 </main>
@@ -343,8 +361,15 @@ <h2>8. お問い合わせ</h2>
       select.value = lang;
       // Taken from the page rather than a parallel list of titles, so a new
       // language needs no edit here.
-      var heading = document.querySelector('section[lang="' + lang + '"] h2');
+      var section = document.querySelector('section[lang="' + lang + '"]');
+      var heading = section && section.querySelector("h2");
       if (heading) document.title = heading.textContent.trim() + " — Tortoise Blocks";
+      // The static description names both languages, because that is what the
+      // bare URL and a script-less reader get.
+      var desc = document.querySelector('meta[name="description"]');
+      if (desc && section && section.dataset.description) {
+        desc.setAttribute("content", section.dataset.description);
+      }
       if (remember) {
         // The choice rides in the URL, not in storage: a page promising no
         // data collection has no business writing localStorage, and this
diff --git a/site/sitemap.xml b/site/sitemap.xml
new file mode 100644
index 0000000..6fec477
--- /dev/null
+++ b/site/sitemap.xml
@@ -0,0 +1,58 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!--
+  Six URLs, not two: an hreflang set has to name every variant from every
+  variant, so each of the three addresses each page answers to gets its own
+  entry carrying the whole set. They are the same alternates the pages
+  declare in their heads — a sitemap is where a crawler finds the Japanese
+  URL before it has rendered anything.
+
+  No <lastmod>, <changefreq> or <priority>. Google ignores the last two
+  outright, and lastmod only counts while it is honest: the site is
+  deployed by hand at release time, so a date in here would be one more
+  thing to remember and would go stale the first time it was forgotten.
+
+  This file is submitted in Search Console rather than announced from a
+  robots.txt, because a robots.txt is only read at the host root —
+  temoki.github.io/robots.txt — which belongs to no repository here. A
+  sitemap has no such rule: it may sit at any level at or above the URLs
+  it lists.
+-->
+<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"
+        xmlns:xhtml="http://www.w3.org/1999/xhtml">
+  <url>
+    <loc>https://temoki.github.io/TortoiseBlocks/?lang=en</loc>
+    <xhtml:link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/?lang=en"/>
+    <xhtml:link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/?lang=ja"/>
+    <xhtml:link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/"/>
+  </url>
+  <url>
+    <loc>https://temoki.github.io/TortoiseBlocks/?lang=ja</loc>
+    <xhtml:link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/?lang=en"/>
+    <xhtml:link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/?lang=ja"/>
+    <xhtml:link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/"/>
+  </url>
+  <url>
+    <loc>https://temoki.github.io/TortoiseBlocks/</loc>
+    <xhtml:link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/?lang=en"/>
+    <xhtml:link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/?lang=ja"/>
+    <xhtml:link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/"/>
+  </url>
+  <url>
+    <loc>https://temoki.github.io/TortoiseBlocks/privacy.html?lang=en</loc>
+    <xhtml:link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=en"/>
+    <xhtml:link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=ja"/>
+    <xhtml:link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/privacy.html"/>
+  </url>
+  <url>
+    <loc>https://temoki.github.io/TortoiseBlocks/privacy.html?lang=ja</loc>
+    <xhtml:link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=en"/>
+    <xhtml:link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=ja"/>
+    <xhtml:link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/privacy.html"/>
+  </url>
+  <url>
+    <loc>https://temoki.github.io/TortoiseBlocks/privacy.html</loc>
+    <xhtml:link rel="alternate" hreflang="en" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=en"/>
+    <xhtml:link rel="alternate" hreflang="ja" href="https://temoki.github.io/TortoiseBlocks/privacy.html?lang=ja"/>
+    <xhtml:link rel="alternate" hreflang="x-default" href="https://temoki.github.io/TortoiseBlocks/privacy.html"/>
+  </url>
+</urlset>