@jant/core 0.6.15 → 0.7.0

package/src/lib/__tests__/feed.test.ts

@@ -51,6 +51,65 @@ function makeFeedData(post: FeedPostView): FeedData {
 }
 
 describe("feed renderers", () => {
+  // Stamping the render time on <updated> tells every reader the feed changed
+  // on every poll, which is untrue and useless for change detection.
+  it("dates the feed by its newest entry, not the render time", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      posts: [
+        makePostView({
+          id: "a",
+          permalink: "/a",
+          feedUpdatedAt: "2026-01-02T00:00:00.000Z",
+        }),
+        makePostView({
+          id: "b",
+          permalink: "/b",
+          feedUpdatedAt: "2026-05-09T00:00:00.000Z",
+        }),
+        makePostView({
+          id: "c",
+          permalink: "/c",
+          feedUpdatedAt: "2026-03-04T00:00:00.000Z",
+        }),
+      ],
+    });
+
+    const feedUpdated = /<id>[^<]*<\/id>\s*<updated>([^<]+)<\/updated>/.exec(
+      xml,
+    )?.[1];
+    expect(feedUpdated).toBe("2026-05-09T00:00:00.000Z");
+  });
+
+  // A directory listing this blog has no other machine-readable place to read
+  // an avatar from, so the feed is where it has to be.
+  it("emits the site avatar as an absolute atom:icon", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      siteIconUrl: "/media/avatar.png",
+    });
+
+    expect(xml).toContain("<icon>https://example.com/media/avatar.png</icon>");
+  });
+
+  it("leaves atom:icon out when the site has no avatar", () => {
+    const xml = defaultFeedRenderer(makeFeedData(makePostView()));
+
+    expect(xml).not.toContain("<icon>");
+  });
+
+  it("falls back to the render time for an empty feed", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      posts: [],
+    });
+
+    const feedUpdated = /<id>[^<]*<\/id>\s*<updated>([^<]+)<\/updated>/.exec(
+      xml,
+    )?.[1];
+    expect(feedUpdated).toMatch(/^\d{4}-\d{2}-\d{2}T/);
+  });
+
   it("keeps Atom entry titles empty for untitled posts and strips script tags from content", () => {
     const xml = defaultFeedRenderer(
       makeFeedData(
@@ -65,7 +124,10 @@ describe("feed renderers", () => {
     );
 
     expect(xml).toContain("<title></title>");
-    expect(xml).toContain('<summary type="text">哈哈哈😍</summary>');
+    // The summary carries the entry's text whether or not it was truncated.
+    expect(xml).toContain(
+      '<summary type="html"><![CDATA[<p>哈哈哈😍</p>]]></summary>',
+    );
     expect(xml).toContain("<![CDATA[<p>哈哈哈😍</p>]]>");
     expect(xml).not.toContain("data-jant-meta");
     expect(xml).not.toContain('{"kind":"text"}');
@@ -110,6 +172,24 @@ describe("feed renderers", () => {
     expect(xml).not.toContain("<iframe");
   });
 
+  it("resolves relative links and media URLs inside post HTML", () => {
+    const post = makePostView({
+      bodyHtml:
+        '<p><a href="/related">Related</a> <a href="#footnote">Footnote</a></p>' +
+        '<img src="/media/inline.jpg" alt="Inline">' +
+        '<video poster="/media/poster.jpg"><source src="/media/clip.mp4"></video>' +
+        '<span data-src="/leave-this-relative"></span>',
+    });
+    const xml = defaultFeedRenderer(makeFeedData(post));
+
+    expect(xml).toContain('href="https://example.com/related"');
+    expect(xml).toContain('href="#footnote"');
+    expect(xml).toContain('src="https://example.com/media/inline.jpg"');
+    expect(xml).toContain('poster="https://example.com/media/poster.jpg"');
+    expect(xml).toContain('src="https://example.com/media/clip.mp4"');
+    expect(xml).toContain('data-src="/leave-this-relative"');
+  });
+
   it("does not expose quote attribution as feed title", () => {
     const xml = defaultFeedRenderer(
       makeFeedData(
@@ -126,10 +206,89 @@ describe("feed renderers", () => {
 
     expect(xml).toContain("<title></title>");
     expect(xml).toContain(
-      '<summary type="text">What stands in the way becomes the way.</summary>',
+      '<blockquote cite="https://example.com/meditations">' +
+        "<p>What stands in the way becomes the way.</p></blockquote>",
     );
     expect(xml).toContain("Marcus Aurelius");
     expect(xml).toContain("https://example.com/meditations");
+    // The quoted text is the entry's text, so it rides in the summary too.
+    expect(xml).toContain(
+      '<summary type="html"><![CDATA[<figure><blockquote cite=',
+    );
+  });
+
+  // The site keeps quote line breaks with `white-space: pre-line`; feed
+  // readers strip CSS, so the breaks have to be markup or the quote collapses
+  // into one run-on paragraph.
+  it("keeps quote line breaks as markup", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "quote",
+          quoteText: "Roses are red\nViolets are blue\n\nSo it goes.",
+          summary: undefined,
+          excerpt: undefined,
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      "<blockquote><p>Roses are red<br/>Violets are blue</p>\n<p>So it goes.</p></blockquote>",
+    );
+  });
+
+  it("normalizes CRLF line breaks in quotes", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "quote",
+          quoteText: "First line\r\nSecond line",
+          summary: undefined,
+          excerpt: undefined,
+        }),
+      ),
+    );
+
+    const content =
+      /<content type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/content>/.exec(
+        xml,
+      )?.[1];
+    expect(content).toBe(
+      "<figure><blockquote><p>First line<br/>Second line</p></blockquote></figure>",
+    );
+  });
+
+  it("escapes quote text before inserting line-break markup", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "quote",
+          quoteText: "<script>alert(1)</script>\nsecond & last",
+          summary: undefined,
+          excerpt: undefined,
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      "<blockquote><p>&lt;script&gt;alert(1)&lt;/script&gt;<br/>second &amp; last</p></blockquote>",
+    );
+  });
+
+  it("falls back to the summary when a quote holds only whitespace", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "quote",
+          quoteText: "  \n  ",
+          summary: "A quote worth keeping",
+          excerpt: undefined,
+        }),
+      ),
+    );
+
+    expect(xml).not.toContain("<blockquote");
+    expect(xml).toContain("<p>A quote worth keeping</p>");
   });
 
   it("link posts point <link> to original URL with ★ permalink back to blog", () => {
@@ -158,6 +317,64 @@ describe("feed renderers", () => {
     );
   });
 
+  it("renders YouTube Link previews as a linked thumbnail with a provider-aware action", () => {
+    const post = makePostView({
+      format: "link",
+      title: "A useful video",
+      url: "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
+      bodyHtml: "<p>My notes on the video.</p>",
+      previewKind: "video",
+      previewProvider: "youtube",
+      previewImageUrl: "/media/previews/youtube.jpg",
+    });
+    const xml = defaultFeedRenderer(makeFeedData(post));
+
+    expect(xml).toContain(
+      '<figure><a href="https://www.youtube.com/watch?v=dQw4w9WgXcQ"><img src="https://example.com/media/previews/youtube.jpg" alt="A useful video"/></a><figcaption><a href="https://www.youtube.com/watch?v=dQw4w9WgXcQ">▶ Watch on YouTube</a></figcaption></figure>',
+    );
+    // The reference comes before the commentary. Scoped to the content: the
+    // summary carries the commentary without the preview, so a whole-document
+    // search would find the note there first.
+    const content =
+      /<content type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/content>/.exec(
+        xml,
+      )?.[1] ?? "";
+    expect(content.indexOf("<figure>")).toBeLessThan(
+      content.indexOf("My notes on the video."),
+    );
+    expect(xml).not.toContain("<iframe");
+    expect(xml).not.toContain('rel="enclosure"');
+  });
+
+  it("renders non-video Link previews without a video action", () => {
+    const post = makePostView({
+      format: "link",
+      title: "Illustrated article",
+      url: "https://external.com/illustrated",
+      previewKind: "image",
+      previewImageUrl: "https://example.com/previews/article.jpg",
+    });
+    const xml = defaultFeedRenderer(makeFeedData(post));
+
+    expect(xml).toContain(
+      '<figure><a href="https://external.com/illustrated"><img src="https://example.com/previews/article.jpg" alt="Illustrated article"/></a></figure>',
+    );
+    expect(xml).not.toContain("Watch video");
+    expect(xml).not.toContain("Watch on");
+  });
+
+  it("keeps the text fallback when a Link post has no preview image", () => {
+    const post = makePostView({
+      format: "link",
+      title: "Text-only link",
+      url: "https://external.com/text-only",
+    });
+    const xml = defaultFeedRenderer(makeFeedData(post));
+
+    expect(xml).toContain("<![CDATA[<p>Text-only link</p>");
+    expect(xml).not.toContain("<figure>");
+  });
+
   it("note posts still link to blog permalink without ★", () => {
     const post = makePostView({
       format: "note",
@@ -185,7 +402,7 @@ describe("feed renderers", () => {
     expect(xml).toContain("<updated>2026-03-20T09:45:00.000Z</updated>");
   });
 
-  it("renders thread replies with hr separator and time element", () => {
+  it("renders thread replies with hr separator and a dated permalink", () => {
     const reply = makePostView({
       id: "reply-1",
       permalink: "/reply-1",
@@ -210,7 +427,9 @@ describe("feed renderers", () => {
 
     expect(xml).toContain("<p>Root content</p>");
     expect(xml).toContain("<hr/>");
-    expect(xml).toContain('<time datetime="2026-03-19T12:00:00.000Z">');
+    expect(xml).toContain(
+      '<time class="dt-published" datetime="2026-03-19T12:00:00.000Z">',
+    );
     expect(xml).toContain("<p>This is a reply</p>");
   });
 
@@ -272,6 +491,38 @@ describe("feed renderers", () => {
     );
   });
 
+  it("renders a YouTube Link preview inside a thread reply", () => {
+    const reply = makePostView({
+      id: "reply-video",
+      permalink: "/reply-video",
+      slug: "reply-video",
+      format: "link",
+      title: "Thread video",
+      url: "https://youtu.be/dQw4w9WgXcQ",
+      previewKind: "video",
+      previewProvider: "youtube",
+      previewImageUrl: "/media/previews/thread-video.jpg",
+    });
+
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          title: "Thread Root",
+          bodyHtml: "<p>Root content</p>",
+          threadReplies: [reply],
+        }),
+      ),
+    );
+
+    const replyTitle =
+      '<h2><a href="https://youtu.be/dQw4w9WgXcQ">Thread video</a></h2>';
+    const preview =
+      '<figure><a href="https://youtu.be/dQw4w9WgXcQ"><img src="https://example.com/media/previews/thread-video.jpg" alt="Thread video"/></a><figcaption><a href="https://youtu.be/dQw4w9WgXcQ">▶ Watch on YouTube</a></figcaption></figure>';
+    expect(xml).toContain(replyTitle);
+    expect(xml).toContain(preview);
+    expect(xml.indexOf(replyTitle)).toBeLessThan(xml.indexOf(preview));
+  });
+
   it("includes thread reply media as Atom enclosures on the combined entry", () => {
     const reply = makePostView({
       id: "reply-1",
@@ -331,12 +582,12 @@ describe("feed renderers", () => {
       '<img src="https://example.com/media/photo.jpg" alt="A red bicycle" width="1200" height="800"/>',
     );
     expect(xml).toContain("<figcaption>A red bicycle</figcaption>");
-    expect(xml).toContain(
-      '<link rel="enclosure" type="image/jpeg" href="https://example.com/media/photo.jpg" length="245000"',
-    );
+    // The content already shows the picture inside a link to the original, so
+    // an enclosure would only ask an attachment shelf to list it again.
+    expect(xml).not.toContain('rel="enclosure"');
   });
 
-  it("renders video attachments as poster + caption (never inline <video>)", () => {
+  it("inlines a video player with the poster as an attribute", () => {
     const post = makePostView({
       media: [
         makeMediaView({
@@ -354,12 +605,19 @@ describe("feed renderers", () => {
     });
     const xml = defaultFeedRenderer(makeFeedData(post));
 
-    expect(xml).not.toContain("<video");
     expect(xml).toContain(
-      '<img src="https://example.com/media/clip-poster.jpg"',
+      '<video controls preload="none" ' +
+        'poster="https://example.com/media/clip-poster.jpg" ' +
+        'width="1920" height="1080">' +
+        '<source src="https://example.com/media/clip.mp4" type="video/mp4"/>' +
+        "</video>",
     );
-    // The action label is a link (not just the thumbnail); metadata sits
-    // outside the link in parens, matching the audio/text attachment style.
+    // The poster is an attribute now, not an <img> the reader might show
+    // beside the player.
+    expect(xml).not.toContain("<img");
+    // The action label stays a link outside the <video>, so it survives a
+    // sanitizer that drops the player; metadata sits outside the link in
+    // parens, matching the audio/text attachment style.
     expect(xml).toContain(
       '<figcaption><a href="https://example.com/media/clip.mp4">▶ Watch video</a> (0:42 · 1.1 MB)</figcaption>',
     );
@@ -368,6 +626,102 @@ describe("feed renderers", () => {
     );
   });
 
+  // The media pipeline only transforms image MIME types, so a video's
+  // `thumbnailUrl` is the file itself. Feeding that to `<img src>` produced a
+  // broken image pointing at an MP4.
+  it("omits the poster for a video with no poster frame", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              id: "med_clip",
+              url: "https://example.com/media/clip.mp4",
+              thumbnailUrl: "https://example.com/media/clip.mp4",
+              mimeType: "video/mp4",
+              size: 28_000_000,
+              width: 960,
+              height: 448,
+              originalName: "clip.mp4",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    // No <img>, and crucially no poster pointing at the video file itself.
+    expect(xml).not.toContain("<img");
+    expect(xml).not.toContain("poster=");
+    expect(xml).toContain(
+      '<video controls preload="none" width="960" height="448">' +
+        '<source src="https://example.com/media/clip.mp4" type="video/mp4"/>' +
+        "</video>",
+    );
+    // The play cue, the link, and the size all survive without the still.
+    expect(xml).toContain(
+      '<figcaption><a href="https://example.com/media/clip.mp4">▶ Watch video</a> (26.7 MB)</figcaption>',
+    );
+    // Media RSS must not invent a thumbnail out of the video either.
+    expect(xml).not.toContain("<media:thumbnail");
+    // The enclosure is what lets a reader offer its own player.
+    expect(xml).toContain(
+      '<link rel="enclosure" type="video/mp4" href="https://example.com/media/clip.mp4"',
+    );
+  });
+
+  // `<video>` has no `alt` attribute, so a described clip would lose its
+  // description entirely if the caption did not carry it.
+  it("moves a video's alt text into its caption", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              id: "med_described",
+              url: "https://example.com/media/described.mp4",
+              thumbnailUrl: "https://example.com/media/described.mp4",
+              mimeType: "video/mp4",
+              altText: "A heron takes off",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      '<figcaption><a href="https://example.com/media/described.mp4">' +
+        "▶ Watch video</a>: A heron takes off</figcaption>",
+    );
+  });
+
+  it("resolves relative attachment, poster, and enclosure URLs", () => {
+    const post = makePostView({
+      media: [
+        makeMediaView({
+          id: "med_local_vid",
+          url: "/media/local-clip.mp4",
+          thumbnailUrl: "/media/local-clip-thumb.jpg",
+          posterUrl: "/media/local-clip-poster.jpg",
+          mimeType: "video/mp4",
+        }),
+      ],
+    });
+    const xml = defaultFeedRenderer(makeFeedData(post));
+
+    expect(xml).toContain(
+      'poster="https://example.com/media/local-clip-poster.jpg"',
+    );
+    expect(xml).toContain(
+      '<source src="https://example.com/media/local-clip.mp4" type="video/mp4"/>',
+    );
+    expect(xml).toContain(
+      '<a href="https://example.com/media/local-clip.mp4">▶ Watch video</a>',
+    );
+    expect(xml).toContain(
+      '<link rel="enclosure" type="video/mp4" href="https://example.com/media/local-clip.mp4"',
+    );
+  });
+
   it("renders audio attachments as a labeled link with duration and size", () => {
     const post = makePostView({
       media: [
@@ -538,3 +892,1328 @@ describe("feed renderers", () => {
     expect(xml).not.toContain("<figure>");
   });
 });
+
+describe("feed Discover declaration", () => {
+  it("declares the mode and the feeds to poll", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      discover: "latest",
+      discoverFeedUrl: "https://example.com/latest/feed",
+      discoverFeaturedFeedUrl: "https://example.com/featured/feed",
+    });
+
+    expect(xml).toContain('xmlns:jant="https://jant.me/ns"');
+    expect(xml).toContain(
+      '<jant:discover feed="https://example.com/latest/feed" featured="https://example.com/featured/feed">latest</jant:discover>',
+    );
+  });
+
+  it("names no featured feed when none was given", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      discover: "latest",
+      discoverFeedUrl: "https://example.com/latest/feed",
+    });
+
+    expect(xml).toContain(
+      '<jant:discover feed="https://example.com/latest/feed">latest</jant:discover>',
+    );
+  });
+
+  it("points featured sites at the featured feed", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      discover: "featured",
+      discoverFeedUrl: "https://example.com/featured/feed",
+    });
+
+    expect(xml).toContain(
+      '<jant:discover feed="https://example.com/featured/feed">featured</jant:discover>',
+    );
+  });
+
+  // `none` is an answer, so it is still declared — a crawler that already
+  // knows the site has to be told to stop, and silence would read as "this
+  // site predates Discover" instead.
+  it("declares none without a feed attribute", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      discover: "none",
+      discoverFeedUrl: null,
+    });
+
+    expect(xml).toContain("<jant:discover>none</jant:discover>");
+    expect(xml).not.toContain("feed=");
+  });
+
+  it("declares nothing when the field is absent", () => {
+    const xml = defaultFeedRenderer(makeFeedData(makePostView()));
+
+    expect(xml).not.toContain("jant:discover");
+  });
+
+  // The namespace is declared for whatever is emitted in it, and an entry's
+  // format is emitted whatever the site answered about Discover. So the only
+  // feed that leaves the declaration out is one with nothing in it at all.
+  it("declares no namespace when nothing in it is emitted", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      posts: [],
+    });
+
+    expect(xml).not.toContain("xmlns:jant");
+  });
+
+  it("keeps a sitePathPrefix in the polled feed URL", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      siteUrl: "https://example.com/blog",
+      discover: "latest",
+      discoverFeedUrl: "https://example.com/blog/latest/feed",
+    });
+
+    expect(xml).toContain('feed="https://example.com/blog/latest/feed"');
+  });
+});
+
+describe("feed language alternates", () => {
+  it("links each language's copy of the same feed", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      languageAlternates: [
+        { hreflang: "zh-Hans", href: "https://example.com/latest/feed" },
+        { hreflang: "en", href: "https://example.com/en/latest/feed" },
+      ],
+    });
+
+    expect(xml).toContain(
+      '<link href="https://example.com/latest/feed" rel="alternate" type="application/atom+xml" hreflang="zh-Hans"/>',
+    );
+    expect(xml).toContain(
+      '<link href="https://example.com/en/latest/feed" rel="alternate" type="application/atom+xml" hreflang="en"/>',
+    );
+  });
+
+  it("emits nothing extra for a single-language site", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      languageAlternates: [],
+    });
+
+    expect(xml).not.toContain("hreflang");
+  });
+});
+
+describe("feed author", () => {
+  // The feed title is composed, so it is not a name. A consumer that wants to
+  // label the blog needs the site's own name somewhere, and this is the place
+  // Atom already has for it.
+  it("names the blog, separately from the composed feed title", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      siteName: "A blog",
+      title: "A blog - Latest posts",
+    });
+
+    expect(xml).toContain("<author><name>A blog</name></author>");
+    expect(xml).toContain("<title>A blog - Latest posts</title>");
+  });
+
+  it("escapes a name that contains markup characters", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      siteName: "Q&A <notes>",
+    });
+
+    expect(xml).toContain(
+      "<author><name>Q&amp;A &lt;notes&gt;</name></author>",
+    );
+  });
+
+  it("emits nothing when the site has no name", () => {
+    const xml = defaultFeedRenderer({
+      ...makeFeedData(makePostView()),
+      siteName: "",
+    });
+
+    expect(xml).not.toContain("<author>");
+  });
+});
+
+describe("feed entry format", () => {
+  // A quote and an untitled note are identical from the feed alone — both have
+  // an empty <title> and a body — so this element is the only way to tell them
+  // apart without fetching the post's page.
+  it.each([
+    ["note", "note"],
+    ["link", "link"],
+    ["quote", "quote"],
+  ] as const)("declares a %s post as %s", (format, declared) => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format,
+          url: format === "link" ? "https://external.com/article" : undefined,
+          quoteText: format === "quote" ? "A quote worth keeping" : undefined,
+        }),
+      ),
+    );
+
+    expect(xml).toContain(`<jant:format>${declared}</jant:format>`);
+    expect(xml).toContain('xmlns:jant="https://jant.me/ns"');
+  });
+
+  // Core's own three formats. A titled note stays a note; drawing it as an
+  // article is a decision the consumer makes from this and <title>.
+  it("calls a titled note a note", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(makePostView({ format: "note", title: "A titled note" })),
+    );
+
+    expect(xml).toContain("<jant:format>note</jant:format>");
+  });
+
+  // Threads arrive as one entry keyed to the root, so the format is the root's.
+  it("declares a thread by its root post", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "quote",
+          quoteText: "A quote worth keeping",
+          threadReplies: [
+            makePostView({
+              id: "post-2",
+              permalink: "/post-2",
+              format: "note",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain("<jant:format>quote</jant:format>");
+    expect(xml).not.toContain("<jant:format>note</jant:format>");
+  });
+});
+
+/**
+ * `<summary>` is the entry's text as the timeline renders it; `<content>` is
+ * the post's full page. Media lives in the content and in the Media RSS
+ * elements, never here.
+ */
+describe("feed entry summary", () => {
+  function makeBody(paragraphs: string[]): string {
+    return JSON.stringify({
+      type: "doc",
+      content: paragraphs.map((text) => ({
+        type: "paragraph",
+        content: [{ type: "text", text }],
+      })),
+    });
+  }
+
+  const longBody = makeBody([
+    "Alpha ".repeat(40).trim(),
+    "Bravo ".repeat(40).trim(),
+    "Charlie ".repeat(40).trim(),
+    "Delta ".repeat(40).trim(),
+  ]);
+
+  function getSummary(xml: string): string | undefined {
+    return /<summary type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/summary>/.exec(
+      xml,
+    )?.[1];
+  }
+
+  function getContent(xml: string): string | undefined {
+    return /<content type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/content>/.exec(
+      xml,
+    )?.[1];
+  }
+
+  it("sends a truncated summary alongside the full content", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          title: "A long read",
+          body: longBody,
+          bodyHtml: "<p>Alpha</p><p>Bravo</p><p>Charlie</p><p>Delta</p>",
+        }),
+      ),
+    );
+
+    const summary = getSummary(xml);
+    expect(summary).toBeDefined();
+    expect(summary).toContain("Alpha");
+    expect(summary).not.toContain("Delta");
+    expect(getContent(xml)).toContain("Delta");
+  });
+
+  // The field means one thing on its own — "this entry's text" — rather than
+  // one defined by what the content happens to hold, so a short note repeats
+  // itself here. That costs the words and buys `summary ?? ""` as the whole of
+  // a consumer's rule.
+  it("sends the summary even when it repeats the content", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          body: makeBody(["Just the one line."]),
+          bodyHtml: "<p>Just the one line.</p>",
+        }),
+      ),
+    );
+
+    expect(getSummary(xml)).toBe("<p>Just the one line.</p>");
+  });
+
+  // A feed cannot express the timeline's justified row, so a consumer computes
+  // it from the dimensions on `<media:content>`. Repeating the markup here
+  // would only give it something to strip back out.
+  it("leaves media out of the summary and keeps the rating", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          body: makeBody(["嘿嘿嘿"]),
+          bodyHtml: "<p>嘿嘿嘿</p>",
+          rating: 4,
+          media: [
+            makeMediaView({
+              url: "https://example.com/media/one.webp",
+              thumbnailUrl: "https://example.com/media/one.webp",
+              mimeType: "image/webp",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(getSummary(xml)).toBe("<p>嘿嘿嘿</p>\n<p>★★★★☆ 4/5</p>");
+    expect(getContent(xml)).toContain("one.webp");
+  });
+
+  it("leaves a link post's preview and ★ permalink out of the summary", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "link",
+          title: "A useful video",
+          url: "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
+          body: makeBody(["My notes."]),
+          bodyHtml: "<p>My notes.</p>",
+          previewImageUrl: "/media/previews/youtube.jpg",
+        }),
+      ),
+    );
+
+    expect(getSummary(xml)).toBe("<p>My notes.</p>");
+    const content = getContent(xml);
+    expect(content).toContain("youtube.jpg");
+    expect(content).toContain("★");
+  });
+
+  // Missing says exactly one thing: this post has no text.
+  it("omits the summary for a post with no text at all", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              url: "https://example.com/media/1031.pdf",
+              thumbnailUrl: "https://example.com/media/1031.pdf",
+              mimeType: "application/pdf",
+              size: 915872,
+              originalName: "1031.pdf",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).not.toContain("<summary");
+    expect(getContent(xml)).toContain("1031.pdf");
+  });
+
+  // An entry's content may not be empty, so a post with no body, no media and
+  // no title falls back to a bare `Post #<id>`. A summary may be empty, and
+  // must be — that placeholder is not text the author wrote.
+  it("omits the summary rather than falling back to a placeholder", () => {
+    const xml = defaultFeedRenderer(makeFeedData(makePostView()));
+
+    expect(xml).not.toContain("<summary");
+    expect(getContent(xml)).toContain("Post #post-1");
+  });
+
+  // The site draws this separator with `.feed-quote-commentary::before`, which
+  // feed readers strip along with the rest of the CSS.
+  it("separates a quote from its commentary with a rule", () => {
+    const content = getContent(
+      defaultFeedRenderer(
+        makeFeedData(
+          makePostView({
+            format: "quote",
+            title: "Marcus Aurelius",
+            url: "https://example.com/meditations",
+            quoteText: "What stands in the way becomes the way.",
+            bodyHtml: "<p>Still true.</p>",
+          }),
+        ),
+      ),
+    );
+
+    expect(content).toContain(
+      "<figcaption>— " +
+        '<a href="https://example.com/meditations">Marcus Aurelius</a>' +
+        "</figcaption></figure>\n<hr/>\n<p>Still true.</p>",
+    );
+  });
+
+  // The same fold the site's timeline applies: the first two replies as
+  // context, the last three with the newest as the hero, the run between them
+  // collapsed into a gap link.
+  it("folds a long thread the way the timeline does", () => {
+    const reply = (n: number) =>
+      makePostView({
+        id: `reply-${n}`,
+        permalink: `/reply-${n}`,
+        body: makeBody([`Reply ${n} body.`]),
+        bodyHtml: `<p>Reply ${n} body.</p>`,
+      });
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          title: "Thread root",
+          body: longBody,
+          bodyHtml: "<p>Alpha</p><p>Delta</p>",
+          threadReplies: [1, 2, 3, 4, 5, 6, 7].map(reply),
+        }),
+      ),
+    );
+
+    const summary = getSummary(xml);
+    // Leading context.
+    expect(summary).toContain("Reply 1 body.");
+    expect(summary).toContain("Reply 2 body.");
+    // The gap stands for replies 3 and 4, and opens the first of them.
+    expect(summary).toContain(
+      '<p><small><a href="https://example.com/reply-3">2 more posts</a></small></p>',
+    );
+    expect(summary).not.toContain("Reply 3 body.");
+    expect(summary).not.toContain("Reply 4 body.");
+    // Trailing context, then the hero.
+    expect(summary).toContain("Reply 5 body.");
+    expect(summary).toContain("Reply 6 body.");
+    expect(summary).toContain("Reply 7 body.");
+
+    // The content still carries every reply.
+    const content = getContent(xml);
+    for (const n of [1, 2, 3, 4, 5, 6, 7]) {
+      expect(content).toContain(`Reply ${n} body.`);
+    }
+  });
+
+  // Up to six posts fit the fold, so a shorter thread hides nothing at all.
+  it("hides nothing in a thread the fold fits whole", () => {
+    const reply = (n: number) =>
+      makePostView({
+        id: `reply-${n}`,
+        permalink: `/reply-${n}`,
+        body: makeBody([`Reply ${n} body.`]),
+        bodyHtml: `<p>Reply ${n} body.</p>`,
+      });
+    const summary = getSummary(
+      defaultFeedRenderer(
+        makeFeedData(
+          makePostView({
+            body: makeBody(["Root."]),
+            bodyHtml: "<p>Root.</p>",
+            threadReplies: [1, 2, 3, 4, 5].map(reply),
+          }),
+        ),
+      ),
+    );
+
+    for (const n of [1, 2, 3, 4, 5]) {
+      expect(summary).toContain(`Reply ${n} body.`);
+    }
+    expect(summary).not.toContain("more post");
+  });
+
+  it("writes the gap count in the singular for a single hidden post", () => {
+    const reply = (n: number) =>
+      makePostView({
+        id: `reply-${n}`,
+        permalink: `/reply-${n}`,
+        body: makeBody([`Reply ${n}.`]),
+        bodyHtml: `<p>Reply ${n}.</p>`,
+      });
+    // Six replies: two lead, three trail, one falls in the gap.
+    const summary = getSummary(
+      defaultFeedRenderer(
+        makeFeedData(
+          makePostView({
+            body: makeBody(["Root."]),
+            bodyHtml: "<p>Root.</p>",
+            threadReplies: [1, 2, 3, 4, 5, 6].map(reply),
+          }),
+        ),
+      ),
+    );
+
+    expect(summary).toContain(">1 more post</a>");
+  });
+
+  it("keeps a thread's root and newest reply without a gap when none is hidden", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          body: makeBody(["Root."]),
+          bodyHtml: "<p>Root.</p>",
+          threadReplies: [
+            makePostView({
+              id: "reply-1",
+              permalink: "/reply-1",
+              body: makeBody(["Only reply."]),
+              bodyHtml: "<p>Only reply.</p>",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    const summary = getSummary(xml);
+    expect(summary).toContain("<p>Root.</p>");
+    expect(summary).toContain("<p>Only reply.</p>");
+    expect(summary).not.toContain("more post");
+  });
+
+  // The site renders a quote's commentary whole — `QuoteCard` passes `bodyHtml`
+  // through and nothing clamps `.feed-quote-commentary` — so a summary that cut
+  // it would not be the timeline's rendering.
+  it("keeps a quote's commentary whole, however long it runs", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "quote",
+          title: "Marcus Aurelius",
+          url: "https://example.com/meditations",
+          quoteText: "What stands in the way becomes the way.",
+          body: longBody,
+          bodyHtml: "<p>Alpha</p><p>Bravo</p><p>Charlie</p><p>Delta</p>",
+        }),
+      ),
+    );
+
+    const summary = getSummary(xml);
+    expect(summary).toContain("<p>Delta</p>");
+    expect(summary).toBe(getContent(xml));
+    expect(xml).not.toContain("<jant:truncated/>");
+  });
+});
+
+/**
+ * What an entry says about itself beyond its text: whether the timeline cuts
+ * it, and which collections the author filed it under.
+ */
+describe("feed entry metadata", () => {
+  function makeBody(paragraphs: string[]): string {
+    return JSON.stringify({
+      type: "doc",
+      content: paragraphs.map((text) => ({
+        type: "paragraph",
+        content: [{ type: "text", text }],
+      })),
+    });
+  }
+
+  // `<summary>` is sent whenever there is text, so only this tells a consumer
+  // whether the site would offer a "Read more".
+  it("marks an entry whose summary text was cut", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          title: "A long read",
+          body: makeBody([
+            "Alpha ".repeat(90).trim(),
+            "Omega ".repeat(90).trim(),
+          ]),
+          bodyHtml: "<p>Alpha</p><p>Omega</p>",
+        }),
+      ),
+    );
+
+    expect(xml).toContain("<jant:truncated/>");
+  });
+
+  it("leaves the mark off an entry the timeline shows in full", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          body: makeBody(["Short enough."]),
+          bodyHtml: "<p>Short enough.</p>",
+        }),
+      ),
+    );
+
+    expect(xml).not.toContain("<jant:truncated/>");
+  });
+
+  it("marks an entry whose newest reply was cut", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          body: makeBody(["Root."]),
+          bodyHtml: "<p>Root.</p>",
+          threadReplies: [
+            makePostView({
+              id: "reply-1",
+              permalink: "/reply-1",
+              title: "A long reply",
+              body: makeBody([
+                "Alpha ".repeat(90).trim(),
+                "Omega ".repeat(90).trim(),
+              ]),
+              bodyHtml: "<p>Alpha</p><p>Omega</p>",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain("<jant:truncated/>");
+  });
+
+  // A collection is a label the author chose, which is what `<category>` is
+  // for — unlike the format, which no author typed.
+  it("files an entry under its collections", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          collections: [
+            { slug: "reading", title: "Reading", url: "/reading" },
+            { slug: "notes", title: "Field Notes", url: "/notes" },
+          ],
+        }),
+      ),
+    );
+
+    // A single collection lives in the root URL namespace and a site path
+    // prefix makes it unguessable from the term, so the URL rides along.
+    expect(xml).toContain(
+      '<category term="reading" label="Reading" jant:page="https://example.com/reading"/>',
+    );
+    expect(xml).toContain(
+      '<category term="notes" label="Field Notes" jant:page="https://example.com/notes"/>',
+    );
+  });
+
+  it("emits no categories for a post in no collection", () => {
+    expect(defaultFeedRenderer(makeFeedData(makePostView()))).not.toContain(
+      "<category",
+    );
+  });
+});
+
+describe("feed attachment metadata", () => {
+  it("describes an image with Media RSS instead of an enclosure", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              url: "https://example.com/media/photo.jpg",
+              thumbnailUrl: "https://example.com/media/photo-thumb.jpg",
+              mimeType: "image/jpeg",
+              size: 245000,
+              width: 1600,
+              height: 1200,
+              altText: "A quiet street",
+              originalName: "photo.jpg",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain('xmlns:media="http://search.yahoo.com/mrss/"');
+    expect(xml).toContain(
+      '<media:content url="https://example.com/media/photo.jpg" ' +
+        'type="image/jpeg" medium="image" fileSize="245000" ' +
+        'width="1600" height="1200">',
+    );
+    expect(xml).toContain(
+      '<media:description type="plain">A quiet street</media:description>',
+    );
+    expect(xml).toContain(
+      '<media:thumbnail url="https://example.com/media/photo-thumb.jpg"/>',
+    );
+    // No enclosure for a picture the content already renders in full.
+    expect(xml).not.toContain('rel="enclosure"');
+  });
+
+  // An enclosure is for a file the content cannot inline. Audio is exactly
+  // that; a picture the reader is already looking at is not.
+  it("encloses audio but not the image beside it, and describes both", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              id: "med_img",
+              url: "https://example.com/media/photo.jpg",
+              thumbnailUrl: "https://example.com/media/photo.jpg",
+              mimeType: "image/jpeg",
+              size: 245000,
+            }),
+            makeMediaView({
+              id: "med_audio",
+              url: "https://example.com/media/song.mp3",
+              thumbnailUrl: "https://example.com/media/song.mp3",
+              mimeType: "audio/mpeg",
+              size: 5242880,
+              durationSeconds: 201,
+              originalName: "song.mp3",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      '<link rel="enclosure" type="audio/mpeg" href="https://example.com/media/song.mp3"',
+    );
+    expect(xml).not.toContain('<link rel="enclosure" type="image/jpeg"');
+
+    // Media RSS still describes both — it is a metadata layer, not a shelf.
+    expect(xml).toContain('medium="image"');
+    expect(xml).toContain('medium="audio"');
+    expect(xml).toContain('duration="201"');
+  });
+
+  it("carries duration for timed media and resolves relative URLs", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              url: "/media/clip.mp4",
+              thumbnailUrl: "/media/clip-poster.jpg",
+              mimeType: "video/mp4",
+              durationSeconds: 42.4,
+              width: 1920,
+              height: 1080,
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      '<media:content url="https://example.com/media/clip.mp4" ' +
+        'type="video/mp4" medium="video" width="1920" height="1080" ' +
+        'duration="42">',
+    );
+  });
+
+  it("calls non-visual attachments documents", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              url: "https://example.com/media/spec.pdf",
+              thumbnailUrl: "https://example.com/media/spec.pdf",
+              mimeType: "application/pdf",
+              originalName: "spec.pdf",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain('medium="document"');
+    expect(xml).toContain('<media:title type="plain">spec.pdf</media:title>');
+  });
+
+  // Card and grid views need a representative image. Without one declared, a
+  // reader has to scrape the first `<img>` out of the content HTML.
+  it("declares a link preview as the entry's representative image", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          format: "link",
+          url: "https://external.com/article",
+          title: "An article",
+          previewImageUrl: "/previews/article.jpg",
+        }),
+      ),
+    );
+
+    expect(xml).toContain('xmlns:media="http://search.yahoo.com/mrss/"');
+    expect(xml).toContain(
+      '<media:thumbnail url="https://example.com/previews/article.jpg"/>',
+    );
+    // A scraped thumbnail of someone else's page is not a published file, so
+    // it must not tell podcast and download clients to fetch it.
+    expect(xml).not.toContain('rel="enclosure"');
+  });
+
+  // The site lays a post's attachments out as one strip and marks the container
+  // `data-post-media`. Carrying it into the feed lets a consumer style the
+  // strip instead of reassembling it from the Media RSS elements.
+  it("groups a post's attachments in one container", () => {
+    const content =
+      /<content type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/content>/.exec(
+        defaultFeedRenderer(
+          makeFeedData(
+            makePostView({
+              bodyHtml: "<p>Look.</p>",
+              media: [
+                makeMediaView({
+                  id: "med_a",
+                  url: "https://example.com/media/a.webp",
+                  thumbnailUrl: "https://example.com/media/a.webp",
+                  mimeType: "image/webp",
+                }),
+                makeMediaView({
+                  id: "med_b",
+                  url: "https://example.com/media/b.webp",
+                  thumbnailUrl: "https://example.com/media/b.webp",
+                  mimeType: "image/webp",
+                }),
+              ],
+            }),
+          ),
+        ),
+      )?.[1];
+
+    expect(content).toContain("<div data-post-media>");
+    expect(content?.match(/<div data-post-media>/g)).toHaveLength(1);
+    expect(content).toContain("a.webp");
+    expect(content).toContain("b.webp");
+  });
+
+  // A flat list cannot say which post in a thread an attachment belongs to.
+  // The content can, because each post's strip sits beside its own text.
+  it("gives each post in a thread its own container", () => {
+    const content =
+      /<content type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/content>/.exec(
+        defaultFeedRenderer(
+          makeFeedData(
+            makePostView({
+              bodyHtml: "<p>Root.</p>",
+              media: [
+                makeMediaView({
+                  id: "med_root",
+                  url: "https://example.com/media/root.webp",
+                  thumbnailUrl: "https://example.com/media/root.webp",
+                  mimeType: "image/webp",
+                }),
+              ],
+              threadReplies: [
+                makePostView({
+                  id: "reply-1",
+                  permalink: "/reply-1",
+                  bodyHtml: "<p>Reply.</p>",
+                  media: [
+                    makeMediaView({
+                      id: "med_reply",
+                      url: "https://example.com/media/reply.webp",
+                      thumbnailUrl: "https://example.com/media/reply.webp",
+                      mimeType: "image/webp",
+                    }),
+                  ],
+                }),
+              ],
+            }),
+          ),
+        ),
+      )?.[1];
+
+    expect(content?.match(/<div data-post-media>/g)).toHaveLength(2);
+    // Each strip sits after its own post's text.
+    expect(content?.indexOf("root.webp")).toBeLessThan(
+      content?.indexOf("Reply.") ?? -1,
+    );
+    expect(content?.indexOf("Reply.")).toBeLessThan(
+      content?.indexOf("reply.webp") ?? -1,
+    );
+  });
+
+  // A markdown file is not worth opening raw — the browser downloads it or
+  // dumps it unstyled — so a text attachment points at the page that renders
+  // it while `url` stays the file.
+  it("points a text attachment at the page that renders it", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          permalink: "/hn2v7",
+          media: [
+            makeMediaView({
+              id: "med_txt",
+              url: "https://cdn.example.com/files/med_txt.md",
+              thumbnailUrl: "https://cdn.example.com/files/med_txt.md",
+              mimeType: "text/markdown; charset=utf-8",
+              size: 21,
+              summary: "松松哈哈哈",
+              originalName: "attached-text.md",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain('jant:page="https://example.com/hn2v7/text/med_txt"');
+    // `url` stays the file, for fetching and for the enclosure.
+    expect(xml).toContain('url="https://cdn.example.com/files/med_txt.md"');
+    // A text attachment has no alt text; its excerpt is what the card prints.
+    expect(xml).toContain(
+      '<media:description type="plain">松松哈哈哈</media:description>',
+    );
+  });
+
+  it("gives an attachment no page of its own when the file is the page", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makePostView({
+          media: [
+            makeMediaView({
+              url: "https://example.com/media/photo.jpg",
+              thumbnailUrl: "https://example.com/media/photo.jpg",
+              mimeType: "image/jpeg",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).not.toContain("jant:page");
+  });
+
+  it("leaves the Media RSS namespace out of a feed with no attachments", () => {
+    const xml = defaultFeedRenderer(makeFeedData(makePostView()));
+    expect(xml).not.toContain("xmlns:media");
+    expect(xml).not.toContain("<media:content");
+    expect(xml).not.toContain("<media:thumbnail");
+  });
+});
+
+/**
+ * A thread arrives as one entry, so both text constructs run several posts
+ * together. These are the three things that let a consumer take it apart:
+ * which entries are threads, where one post's words end, and which post owns
+ * each file.
+ */
+describe("feed entry thread segmentation", () => {
+  function makeBody(paragraphs: string[]): string {
+    return JSON.stringify({
+      type: "doc",
+      content: paragraphs.map((text) => ({
+        type: "paragraph",
+        content: [{ type: "text", text }],
+      })),
+    });
+  }
+
+  function getSummary(xml: string): string | undefined {
+    return /<summary type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/summary>/.exec(
+      xml,
+    )?.[1];
+  }
+
+  function getContent(xml: string): string | undefined {
+    return /<content type="html"><!\[CDATA\[([\s\S]*?)\]\]><\/content>/.exec(
+      xml,
+    )?.[1];
+  }
+
+  function makeReply(n: number, overrides: Partial<FeedPostView> = {}) {
+    return makePostView({
+      id: `reply-${n}`,
+      permalink: `/reply-${n}`,
+      slug: `reply-${n}`,
+      body: makeBody([`Reply ${n} body.`]),
+      bodyHtml: `<p>Reply ${n} body.</p>`,
+      publishedAt: `2026-03-2${n}T00:00:00.000Z`,
+      publishedAtFormatted: `Mar 2${n}, 2026`,
+      ...overrides,
+    });
+  }
+
+  function makeRoot(overrides: Partial<FeedPostView> = {}) {
+    return makePostView({
+      body: makeBody(["Root body."]),
+      bodyHtml: "<p>Root body.</p>",
+      ...overrides,
+    });
+  }
+
+  // The root's own marker is what makes the rule "every marker ends the block
+  // before it" hold. Without it the root and the first reply share a segment.
+  it("closes the root's own block in a thread, not just the replies'", () => {
+    const summary = getSummary(
+      defaultFeedRenderer(
+        makeFeedData(makeRoot({ threadReplies: [makeReply(1)] })),
+      ),
+    );
+
+    expect(summary).toContain(
+      '<p>Root body.</p>\n<p><small><a href="https://example.com/post-1" class="u-url"><time class="dt-published" datetime="2026-03-19T00:00:00.000Z">Mar 19, 2026</time></a></small></p>',
+    );
+    expect(summary).toContain(
+      '<p>Reply 1 body.</p>\n<p><small><a href="https://example.com/reply-1" class="u-url"><time class="dt-published" datetime="2026-03-21T00:00:00.000Z">Mar 21, 2026</time></a></small></p>',
+    );
+  });
+
+  // A reader prints <published> itself, so a lone post that also ended with
+  // its own date would show it twice.
+  it("leaves a standalone entry unmarked", () => {
+    const xml = defaultFeedRenderer(makeFeedData(makeRoot()));
+
+    expect(getSummary(xml)).not.toContain("dt-published");
+    expect(getContent(xml)).not.toContain("dt-published");
+    expect(xml).not.toContain("<jant:thread");
+  });
+
+  // The marker ends a block, so a post that contributed no block to the
+  // summary — a photo with no caption — must not leave one behind.
+  it("marks no block for a thread root with no text", () => {
+    const summary = getSummary(
+      defaultFeedRenderer(
+        makeFeedData(
+          makePostView({
+            media: [
+              makeMediaView({
+                url: "https://example.com/media/photo.jpg",
+                mimeType: "image/jpeg",
+              }),
+            ],
+            threadReplies: [makeReply(1)],
+          }),
+        ),
+      ),
+    );
+
+    expect(summary).not.toContain("https://example.com/post-1");
+    expect(summary).toContain('href="https://example.com/reply-1"');
+  });
+
+  it("closes every reply's block in the content, in order", () => {
+    const content = getContent(
+      defaultFeedRenderer(
+        makeFeedData(makeRoot({ threadReplies: [makeReply(1), makeReply(2)] })),
+      ),
+    );
+
+    const markers = [...(content ?? "").matchAll(/class="u-url"/g)];
+    expect(markers).toHaveLength(3);
+    expect(content?.indexOf("Reply 1 body.")).toBeLessThan(
+      content?.indexOf("Reply 2 body.") ?? -1,
+    );
+  });
+
+  it("declares a thread's shape so a consumer need not read the HTML", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [1, 2, 3, 4, 5, 6, 7].map((n) => makeReply(n)),
+        }),
+      ),
+    );
+
+    // Seven replies: 1 and 2 lead, 5-7 trail, 3 and 4 fall in the gap.
+    expect(xml).toContain(
+      '<jant:thread posts="8" hidden="2" gap="https://example.com/reply-3" latest="https://example.com/reply-7">',
+    );
+  });
+
+  // One reply is the hero, so nothing folds away and there is nowhere to send
+  // a reader who wants the middle.
+  it("declares no gap when a thread hides nothing", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(makeRoot({ threadReplies: [makeReply(1)] })),
+    );
+
+    expect(xml).toContain(
+      '<jant:thread posts="2" hidden="0" latest="https://example.com/reply-1">',
+    );
+  });
+
+  // `<jant:format>` describes the entry, which is the root. A reply that is a
+  // Quote says so nowhere else — the `<blockquote>` in the content is not a
+  // declaration a consumer laying the thread out itself can read.
+  it("lists every post in the thread with its own format and date", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [
+            makeReply(1, {
+              format: "quote",
+              title: "Marcus Aurelius",
+              quoteText: "What stands in the way becomes the way.",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/post-1" format="note" published="2026-03-19T00:00:00.000Z"/>',
+    );
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-1" format="quote" published="2026-03-21T00:00:00.000Z"/>',
+    );
+    // The entry still declares the root's format, unchanged.
+    expect(xml).toContain("<jant:format>note</jant:format>");
+  });
+
+  // A row carries what Atom would put on this post's entry. `<title>` and
+  // `link[rel="alternate"]` only ever describe the root, so a titled reply and
+  // a Link reply had nowhere to say either.
+  it("gives a row the title, target and preview its own entry would carry", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [
+            makeReply(1, {
+              format: "link",
+              title: "The AeroPress guide",
+              url: "https://other.example/aeropress",
+              previewImageUrl: "/media/preview.jpg",
+            }),
+            makeReply(2, { title: "A titled reply" }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-1" format="link" published="2026-03-21T00:00:00.000Z"' +
+        ' title="The AeroPress guide" url="https://other.example/aeropress"' +
+        ' thumbnail="https://example.com/media/preview.jpg"/>',
+    );
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-2" format="note" published="2026-03-22T00:00:00.000Z" title="A titled reply"/>',
+    );
+  });
+
+  // The entry's own `<title>` is empty for a quote because the attribution is
+  // not a title. A row follows the same rule.
+  it("keeps a quote reply's attribution out of its row title", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [
+            makeReply(1, {
+              format: "quote",
+              title: "Marcus Aurelius",
+              quoteText: "What stands in the way becomes the way.",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-1" format="quote" published="2026-03-21T00:00:00.000Z"/>',
+    );
+    expect(xml).not.toContain('title="Marcus Aurelius"');
+  });
+
+  // Truncation lives in `<summary>`, which renders the root and the newest
+  // reply. A folded post has no block to cut, so its row must stay silent
+  // rather than claim it arrived whole.
+  it("marks every cut post's row, and no folded post's", () => {
+    const longBody = makeBody([
+      "Alpha ".repeat(80).trim(),
+      "Bravo ".repeat(80).trim(),
+      "Charlie ".repeat(80).trim(),
+      "Delta ".repeat(80).trim(),
+    ]);
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [1, 2, 3, 4, 5, 6, 7].map((n) =>
+            makeReply(n, { body: longBody }),
+          ),
+        }),
+      ),
+    );
+
+    // Reply 1 leads and reply 7 is the hero: both render, so both can be cut.
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-1" format="note" published="2026-03-21T00:00:00.000Z" truncated="true"/>',
+    );
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-7" format="note" published="2026-03-27T00:00:00.000Z" truncated="true"/>',
+    );
+    // Reply 3 is behind the gap. Its body is just as long, and its row says it
+    // was folded rather than that it arrived whole.
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-3" format="note" published="2026-03-23T00:00:00.000Z" folded="true"/>',
+    );
+    // The entry still answers the single-card reader with one boolean.
+    expect(xml).toContain("<jant:truncated/>");
+  });
+
+  // Every other decision on a row is declared. This one used to be left to a
+  // consumer reading the summary's shape, which misreads a photo with no
+  // caption: the site renders it, it contributes no text, and it comes out
+  // looking exactly like a post the fold hid.
+  it("declares which posts the fold hid", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [1, 2, 3, 4, 5, 6, 7].map((n) => makeReply(n)),
+        }),
+      ),
+    );
+
+    const folded = [...xml.matchAll(/<jant:post [^>]*folded="true"[^>]*\/>/g)];
+    expect(folded).toHaveLength(2);
+    // The same two the gap stands for, and the same count `@hidden` reports.
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-3" format="note" published="2026-03-23T00:00:00.000Z" folded="true"/>',
+    );
+    expect(xml).toContain(
+      '<jant:post href="https://example.com/reply-4" format="note" published="2026-03-24T00:00:00.000Z" folded="true"/>',
+    );
+    expect(xml).toContain('hidden="2"');
+  });
+
+  // A caption-less photo is a normal Jant post. It renders on the site, so its
+  // row must not claim the fold hid it — a consumer that believed otherwise
+  // would drop its attachments and the post would vanish.
+  it("leaves a rendered post with no text unfolded", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [
+            makePostView({
+              id: "reply-1",
+              permalink: "/reply-1",
+              slug: "reply-1",
+              media: [
+                makeMediaView({
+                  url: "https://example.com/media/photo.jpg",
+                  thumbnailUrl: "https://example.com/media/photo.jpg",
+                  mimeType: "image/jpeg",
+                }),
+              ],
+            }),
+            makeReply(2),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain('hidden="0"');
+    expect(xml).not.toContain('folded="true"');
+    // It contributes nothing to the summary, which is exactly why the row has
+    // to speak for it.
+    expect(getSummary(xml)).not.toContain("https://example.com/reply-1");
+    expect(xml).toContain('jant:post="https://example.com/reply-1"');
+  });
+
+  // A reply carries chrome the entry's own fields carry for the root, so a
+  // consumer drawing native cards has to drop it before redrawing from the
+  // row. Matching it by shape is fragile — a body can open with a link.
+  it("wraps a reply's title and source line in a header element", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          threadReplies: [
+            makeReply(1, {
+              format: "link",
+              title: "The AeroPress guide",
+              url: "https://other.example/aeropress",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(getSummary(xml)).toContain(
+      '<header><p><a href="https://other.example/aeropress">other.example</a></p>' +
+        '<h2><a href="https://other.example/aeropress">The AeroPress guide</a></h2></header>',
+    );
+    // The root's chrome stays in the entry's own fields, so its block has no
+    // header to drop and a consumer needs no special case for it.
+    expect(getSummary(xml)?.indexOf("<header>")).toBeGreaterThan(
+      getSummary(xml)?.indexOf("Root body.") ?? -1,
+    );
+  });
+
+  it("gives a reply with no title and no source no header at all", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(makeRoot({ threadReplies: [makeReply(1)] })),
+    );
+
+    expect(getSummary(xml)).not.toContain("<header>");
+  });
+
+  it("lists no thread rows on a lone post", () => {
+    const xml = defaultFeedRenderer(makeFeedData(makeRoot()));
+
+    expect(xml).not.toContain("<jant:post");
+    expect(xml).not.toContain("jant:post=");
+  });
+
+  it("names the post carrying every attachment in a thread, the root's included", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          media: [
+            makeMediaView({
+              id: "med_root",
+              url: "https://example.com/media/beans.jpg",
+              thumbnailUrl: "https://example.com/media/beans.jpg",
+              mimeType: "image/jpeg",
+            }),
+          ],
+          threadReplies: [
+            makeReply(1, {
+              media: [
+                makeMediaView({
+                  id: "med_reply",
+                  url: "https://example.com/media/cup.jpg",
+                  thumbnailUrl: "https://example.com/media/cup.jpg",
+                  mimeType: "image/jpeg",
+                }),
+              ],
+            }),
+          ],
+        }),
+      ),
+    );
+
+    // The root's file names its post too. Guessing wrong here hangs a photo
+    // under the wrong post, which is not what a missing attribute should mean.
+    expect(xml).toContain(
+      '<media:content url="https://example.com/media/beans.jpg" type="image/jpeg" medium="image" jant:post="https://example.com/post-1"/>',
+    );
+    expect(xml).toContain(
+      '<media:content url="https://example.com/media/cup.jpg" type="image/jpeg" medium="image" jant:post="https://example.com/reply-1"/>',
+    );
+  });
+
+  // A lone post's entry has exactly one post, so the attribute would only
+  // repeat `<id>` on every file.
+  it("leaves a lone post's attachment unattributed", () => {
+    const xml = defaultFeedRenderer(
+      makeFeedData(
+        makeRoot({
+          media: [
+            makeMediaView({
+              url: "https://example.com/media/beans.jpg",
+              thumbnailUrl: "https://example.com/media/beans.jpg",
+              mimeType: "image/jpeg",
+            }),
+          ],
+        }),
+      ),
+    );
+
+    expect(xml).toContain("<media:content");
+    expect(xml).not.toContain("jant:post=");
+  });
+});