From a8bc40037163d9f996de5ec6a57d0454a972d1ea Mon Sep 17 00:00:00 2001
From: Jorijn van der Graaf <jorijnvdgraaf@catcrafts.net>
Date: Wed, 19 Aug 2026 00:02:32 +0200
Subject: [PATCH] video seo

---
 deploy/Caddyfile.example                      | 21 +++++--
 shared/interfaces/Catcrafts.Shared-Views.cppm | 56 +++++++++++++++++-
 tests/ShouldEmitStructuredData/main.cpp       | 58 +++++++++++++++++++
 3 files changed, 128 insertions(+), 7 deletions(-)

diff --git a/deploy/Caddyfile.example b/deploy/Caddyfile.example
index 7e63b71..273d05e 100644
--- a/deploy/Caddyfile.example
+++ b/deploy/Caddyfile.example
@@ -208,14 +208,25 @@ catcrafts.net {
 
 	# ── access log ────────────────────────────────────────────────────────
 	#
-	# Feeds the GoAccess analytics (see deploy/catcrafts-analytics). Retention
-	# sized for that: ~15 MB/day means a roll roughly weekly; 52 compressed
-	# rolls ~ a year of raw logs (~6 MB each gzipped). roll_keep_for must be
-	# explicit — lumberjack's default silently deletes after 90 days.
+	# Feeds the GoAccess analytics (see deploy/catcrafts-analytics), and is the
+	# only record of everything those aggregates discard: Sec-Fetch headers,
+	# Range behaviour, query strings, per-IP request sequences. Every bot fleet
+	# identified so far was caught in the raw log, not the report.
+	#
+	# roll_keep is a COUNT, so retention is really min(roll_keep × roll_size,
+	# roll_keep_for) — a byte budget, not the time window it reads as. 52 rolls
+	# is 5.2 GB, which is under a year for anything busier than 14 MB/day, and
+	# nothing here has ever been that quiet: 19-24 MB/day on a slow week, and
+	# 617 MB/day at the Hacker News peak, where 52 rolls is EIGHT DAYS. 365
+	# rolls (~2.6 GB gzipped, at the ~14:1 these compress to) makes
+	# roll_keep_for the constraint that actually decides, up to ~100 MB/day.
+	#
+	# roll_keep_for must be explicit — lumberjack's default silently deletes
+	# after 90 days.
 	log {
 		output file /var/log/caddy/catcrafts.net.log {
 			roll_size 100MiB
-			roll_keep 52
+			roll_keep 365
 			roll_keep_for 8760h
 		}
 	}
diff --git a/shared/interfaces/Catcrafts.Shared-Views.cppm b/shared/interfaces/Catcrafts.Shared-Views.cppm
index 2af75dd..c27278a 100644
--- a/shared/interfaces/Catcrafts.Shared-Views.cppm
+++ b/shared/interfaces/Catcrafts.Shared-Views.cppm
@@ -446,6 +446,57 @@ export RenderedPage RenderPosts(std::span<const Post> posts) {
     return page;
 }
 
+// The post's video, as a schema.org VideoObject hung off the BlogPosting.
+//
+// Google already indexed and carouselled these videos on the strength of the
+// bare <video> element; this makes the placement deliberate rather than lucky,
+// and a typed thumbnailUrl is what earns the still on the text result too.
+//
+// thumbnailUrl and uploadDate are what the video rich result requires, so a
+// video whose mirror produced no poster is skipped rather than emitted
+// incomplete — the same both-or-neither rule Dimensions() applies to width and
+// height. Only our own mirrored paths qualify, for the reason ogImage below
+// gives: an absolute src or a third-party poster is the off-site fetch the
+// mirror exists to avoid, just performed by a crawler.
+//
+// Headline media only. That IS the post — these posts are their recording —
+// and it is the one video a search result should be pointing at.
+//
+// No encodingFormat and no duration: neither is required, the mirror records no
+// duration, and naming a container guessed from a file extension would assert
+// something we do not know. A wrong type is worse than a missing one, for the
+// same reason MimeFor() refuses to guess.
+std::string PostVideoLd(const Post& p) {
+    std::vector<std::string> videos;
+    for (const PostMedia& m : p.media) {
+        if (m.kind != "video") continue;
+        if (!m.src.starts_with("/") || !m.poster.starts_with("/")) continue;
+
+        std::string node = std::format(
+            R"({{"@type":"VideoObject","name":{},"uploadDate":{},)"
+            R"("thumbnailUrl":{},"contentUrl":{})",
+            JsonStr(p.title), JsonStr(p.published),
+            JsonStr("https://catcrafts.net" + m.poster),
+            JsonStr("https://catcrafts.net" + m.src));
+        if (!p.excerpt.empty()) node += ",\"description\":" + JsonStr(p.excerpt);
+        if (m.width > 0 && m.height > 0) {
+            node += std::format(R"(,"width":{},"height":{})", m.width, m.height);
+        }
+        node += "}";
+        videos.push_back(std::move(node));
+    }
+    if (videos.empty()) return {};
+    if (videos.size() == 1) return ",\"video\":" + videos.front();
+
+    std::string out = ",\"video\":[";
+    for (std::size_t i = 0; i < videos.size(); ++i) {
+        if (i) out += ',';
+        out += videos[i];
+    }
+    out += ']';
+    return out;
+}
+
 // ── one post, in full ─────────────────────────────────────────────────
 
 // The whole point of hosting the body: a page a search engine can index, a
@@ -494,14 +545,15 @@ export RenderedPage RenderPost(const Post& p) {
         R"("@type":"Person","name":"Jorijn van der Graaf"}},)"
         R"("publisher":{{"@id":"https://catcrafts.net/#organization",)"
         R"("@type":"Organization","name":"Catcrafts"}},)"
-        R"("discussionUrl":{}{}{}}})",
+        R"("discussionUrl":{}{}{}{}}})",
         JsonStr(p.title), JsonStr(p.published),
         JsonStr("https://catcrafts.net" + canonical),
         JsonStr("https://catcrafts.net" + canonical),
         JsonStr(p.permalink),
         p.excerpt.empty() ? std::string{} : ",\"description\":" + JsonStr(p.excerpt),
         ogImage.empty() ? std::string{}
-                        : ",\"image\":" + JsonStr("https://catcrafts.net" + ogImage));
+                        : ",\"image\":" + JsonStr("https://catcrafts.net" + ogImage),
+        PostVideoLd(p));
 
     page.main = Format(
         R"(<article class="post">)"
diff --git a/tests/ShouldEmitStructuredData/main.cpp b/tests/ShouldEmitStructuredData/main.cpp
index 8cf1065..048494d 100644
--- a/tests/ShouldEmitStructuredData/main.cpp
+++ b/tests/ShouldEmitStructuredData/main.cpp
@@ -181,5 +181,63 @@ int main(int argc, char** argv) {
         Check(personOk, "about Person schema parses and names the founder");
     }
 
+    // ── the post's video ──────────────────────────────────────────────
+    // Google carouselled these recordings off the bare <video> element; the
+    // VideoObject is what makes that deliberate rather than lucky.
+    //
+    // PostVideoLd only marks up our OWN mirrored copies, so this can only be
+    // asserted when the fixture was actually mirrored — an unmirrored checkout
+    // still carries the instance's absolute URLs, and skipping is the correct
+    // behaviour there, not a failure. Same reasoning (and same note) as the
+    // format-ladder checks in ShouldServePostPages.
+    {
+        constexpr std::string_view kVideoPost =
+            "/posts/fairphone-6-postmarketos-working-main-camera";
+        const std::string postBody = srv.Body(std::string(kVideoPost));
+        if (postBody.find("poster=\"/media/") == std::string::npos) {
+            std::println("note: no locally-mirrored poster on {} — the "
+                         "VideoObject checks did not run", kVideoPost);
+        } else {
+            bool found = false;
+            bool complete = false;
+            bool ownMirror = false;
+            for (const Json::Value& doc : ExtractLd(srv, std::string(kVideoPost))) {
+                if (doc.Str("@type") != "BlogPosting") continue;
+                const Json::Value* v = doc.Find("video");
+                if (!v) continue;
+                if (v->IsArray() && v->array.empty()) continue;
+                // One video: the object itself, not a single-element array.
+                const Json::Value& vid = v->IsArray() ? v->array.front() : *v;
+                found = true;
+                const std::string_view thumb = vid.Str("thumbnailUrl");
+                const std::string_view content = vid.Str("contentUrl");
+                complete = vid.Str("@type") == "VideoObject"
+                        && !vid.Str("name").empty()
+                        && !vid.Str("uploadDate").empty()
+                        && !thumb.empty() && !content.empty();
+                ownMirror = thumb.starts_with("https://catcrafts.net/media/")
+                         && content.starts_with("https://catcrafts.net/media/");
+            }
+            Check(found, "a video post's BlogPosting carries a video node");
+            Check(complete, "VideoObject has the fields the rich result requires");
+            Check(ownMirror, "video thumbnail and content URLs point at our own mirror");
+        }
+    }
+
+    // A post that leads with no video must not grow an empty video property:
+    // an Article claiming a video it does not have is a structured-data error,
+    // not a harmless extra key.
+    {
+        bool sawPosting = false;
+        bool bare = true;
+        for (const Json::Value& doc :
+             ExtractLd(srv, "/posts/the-linux-phone-travel-experience")) {
+            if (doc.Str("@type") != "BlogPosting") continue;
+            sawPosting = true;
+            if (doc.Find("video")) bare = false;
+        }
+        Check(sawPosting && bare, "a post with no video carries no video node");
+    }
+
     return Finish();
 }