diff --git a/.github/workflows/docs-integrity.yml b/.github/workflows/docs-integrity.yml
index cc228ac..9c0d80a 100644
--- a/.github/workflows/docs-integrity.yml
+++ b/.github/workflows/docs-integrity.yml
@@ -66,6 +66,18 @@ jobs:
       - name: Discoverability surfaces are published
         run: uv run python scripts/seo_check.py
 
+      - name: Mike-style versioned site_url is slash-safe
+        # mike rewrites site_url to a version path without a trailing slash. Reproduce
+        # that exact input so concatenation bugs cannot hide behind the normal config.
+        run: |
+          cp mkdocs.yml .mkdocs-versioned.yml
+          trap 'rm -f .mkdocs-versioned.yml; rm -rf site-versioned' EXIT
+          sed -i 's|^site_url:.*|site_url: https://mskazemi.com/aobench/latest|' .mkdocs-versioned.yml
+          uv run mkdocs build --strict -f .mkdocs-versioned.yml --site-dir site-versioned
+          uv run python scripts/seo_check.py \
+            --site-dir site-versioned \
+            --base-url https://mskazemi.com/aobench/latest/
+
       - name: Examples still run
         run: uv run python -m pytest tests/test_examples.py -q
 
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 608c47f..e5dd607 100755
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,17 @@
 
 ## Unreleased
 
+### Fixed — versioned documentation URLs no longer lose the path separator
+
+- Normalize mike's slashless versioned `site_url` before appending paths in the theme
+  override. The `llms.txt` alternate, Twitter preview image, changelog banner link, and
+  breadcrumb ancestor URLs now resolve under `/aobench/latest/` instead of malformed
+  paths such as `/aobench/latestllms.txt`.
+- Exclude `docs/overrides/**` from published documentation so raw Jinja source is not
+  exposed as a public page.
+- Extend the SEO gate with a no-trailing-slash versioned build and site-wide assertions
+  for these URLs, including breadcrumb resolution and the override-source leak.
+
 ### Fixed — structured data now matches Google's breadcrumb and Dataset requirements
 
 - Breadcrumb JSON-LD now emits only real navigable ancestors, gives every emitted
diff --git a/docs/overrides/main.html b/docs/overrides/main.html
index 77baf01..e77d624 100644
--- a/docs/overrides/main.html
+++ b/docs/overrides/main.html
@@ -17,6 +17,7 @@
 {% block extrahead %}
   {%- set page_title = (page.meta and page.meta.title) or page.title or config.site_name -%}
   {%- set page_desc = (page.meta and page.meta.description) or config.site_description -%}
+  {%- set site_base = config.site_url.rstrip('/') ~ '/' -%}
 
   <meta name="author" content="Mohsen Seyedkazemi Ardebili">
 
@@ -26,7 +27,7 @@
   <meta name="twitter:card" content="summary_large_image">
   <meta name="twitter:title" content="{{ page_title }}">
   <meta name="twitter:description" content="{{ page_desc }}">
-  <meta name="twitter:image" content="{{ config.site_url }}assets/social-preview.png">
+  <meta name="twitter:image" content="{{ site_base }}assets/social-preview.png">
 
   {%- if page.meta and page.meta.keywords %}
   <meta name="keywords" content="{{ page.meta.keywords | join(', ') }}">
@@ -44,7 +45,7 @@
   <meta name="citation_public_url" content="https://mskazemi.com/aobench/">
   <meta name="citation_abstract_html_url" content="https://mskazemi.com/aobench/">
 
-  <link rel="alternate" type="text/plain" href="{{ config.site_url }}llms.txt" title="llms.txt — factual project map for AI answer engines">
+  <link rel="alternate" type="text/plain" href="{{ site_base }}llms.txt" title="llms.txt — factual project map for AI answer engines">
 
   <script type="application/ld+json">
   {
@@ -141,7 +142,7 @@
         "author": { "@id": "https://mskazemi.com/#person" },
         "license": "https://www.apache.org/licenses/LICENSE-2.0"
       }
-      {%- if page and page.canonical_url != config.site_url %}
+      {%- if page and page.canonical_url != site_base %}
       ,{
         "@type": "BreadcrumbList",
         "@id": "{{ page.canonical_url }}#breadcrumb",
@@ -150,7 +151,7 @@
             "@type": "ListItem",
             "position": 1,
             "name": "AOBench",
-            "item": "{{ config.site_url }}"
+            "item": "{{ site_base }}"
           }
           {%- set crumb = namespace(position=1) %}
           {%- if page.ancestors %}
@@ -161,7 +162,7 @@
             "@type": "ListItem",
             "position": {{ crumb.position }},
             "name": {{ ancestor.title | string | tojson }},
-            "item": "{{ config.site_url }}{{ ancestor.url }}"
+            "item": "{{ site_base }}{{ ancestor.url }}"
           }
           {%- endif %}
           {%- endfor %}
@@ -182,6 +183,7 @@
 {% endblock %}
 
 {% block announce %}
+  {%- set site_base = config.site_url.rstrip('/') ~ '/' -%}
   <strong>AOBench v0.4.1</strong> — 90 tasks · 29 environments · 6 grounded in real Marconi100 data.
-  &nbsp;<a href="{{ config.site_url }}about/changelog/">Read the changelog →</a>
+  &nbsp;<a href="{{ site_base }}about/changelog/">Read the changelog →</a>
 {% endblock %}
diff --git a/mkdocs.yml b/mkdocs.yml
index c93aad2..dcf623a 100755
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -5,6 +5,11 @@ repo_url: https://github.com/MSKazemi/aobench
 repo_name: MSKazemi/aobench
 edit_uri: edit/main/docs/
 
+# The theme override lives under docs/ so Material can load it, but it is source code,
+# not public documentation. Without this exclusion MkDocs copies raw Jinja into the site.
+exclude_docs: |
+  overrides/**
+
 theme:
   name: material
   custom_dir: docs/overrides
diff --git a/scripts/seo_check.py b/scripts/seo_check.py
index 901ccfa..2b16d77 100644
--- a/scripts/seo_check.py
+++ b/scripts/seo_check.py
@@ -10,6 +10,8 @@
 
     python scripts/seo_check.py            # check ./site
     python scripts/seo_check.py --site-dir some/other/site
+    python scripts/seo_check.py --site-dir site-versioned \\
+        --base-url https://mskazemi.com/aobench/latest/
 
 Exits 1 on any failure.
 """
@@ -63,8 +65,15 @@
 
 
 class Checker:
-    def __init__(self, site: Path) -> None:
+    def __init__(self, site: Path, base_url: str) -> None:
         self.site = site
+        self.base_url = base_url.rstrip("/") + "/"
+        parsed_base = urlparse(self.base_url)
+        if parsed_base.scheme != "https" or not parsed_base.netloc:
+            raise ValueError(f"base URL must be absolute https: {self.base_url}")
+        self.base_scheme = parsed_base.scheme
+        self.base_netloc = parsed_base.netloc
+        self.base_path = parsed_base.path
         self.failures: list[str] = []
         self.passes = 0
 
@@ -107,6 +116,11 @@ def check_sitemap(self) -> None:
             all(u.startswith("https://") for u in locs),
             "every sitemap URL is absolute https",
         )
+        self.check(
+            all(u.startswith(self.base_url) for u in locs),
+            "every sitemap URL uses the expected site base",
+            self.base_url,
+        )
 
     def check_llms_txt(self) -> None:
         path = self.site / "llms.txt"
@@ -141,6 +155,14 @@ def check_llms_txt(self) -> None:
                 f"only {n_bytes:,} bytes — did the concatenation break?",
             )
 
+    def check_template_leaks(self) -> None:
+        leaked = self.site / "overrides" / "main.html"
+        self.check(
+            not leaked.exists(),
+            "raw theme override is not published as documentation",
+            str(leaked),
+        )
+
     def check_page(self, rel: str) -> None:
         path = self.site / rel
         if not path.is_file():
@@ -181,12 +203,11 @@ def check_page(self, rel: str) -> None:
 
     def _breadcrumb_target_exists(self, url: str) -> bool:
         parsed = urlparse(url)
-        if parsed.scheme != "https" or parsed.netloc != "mskazemi.com":
+        if parsed.scheme != self.base_scheme or parsed.netloc != self.base_netloc:
             return False
-        prefix = "/aobench/"
-        if not parsed.path.startswith(prefix):
+        if not parsed.path.startswith(self.base_path):
             return False
-        suffix = parsed.path[len(prefix) :].lstrip("/")
+        suffix = parsed.path[len(self.base_path) :].lstrip("/")
         if not suffix:
             return (self.site / "index.html").is_file()
         target = self.site / suffix
@@ -286,6 +307,21 @@ def check_structured_data(self, rel: str, html: str) -> None:
                 f"{rel}: Dataset distributions declare URL and encodingFormat",
             )
 
+    def check_sitewide_base_urls(self) -> None:
+        pages = sorted(self.site.rglob("index.html"))
+        self.check(bool(pages), "slash-safe URL audit found built pages")
+        expected_refs = (
+            ("llms.txt", "llms.txt alternate uses slash-safe base"),
+            ("assets/social-preview.png", "Twitter image uses slash-safe base"),
+            ("about/changelog/", "announcement link uses slash-safe base"),
+        )
+        for path in pages:
+            rel = path.relative_to(self.site).as_posix()
+            html = path.read_text(encoding="utf-8")
+            for suffix, label in expected_refs:
+                expected = self.base_url + suffix
+                self.check(expected in html, f"{rel}: {label}", expected)
+
     def check_sitewide_structured_data(self) -> None:
         pages = sorted(self.site.rglob("index.html"))
         self.check(bool(pages), "structured-data audit found built pages")
@@ -297,6 +333,11 @@ def check_sitewide_structured_data(self) -> None:
 def main() -> int:
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("--site-dir", type=Path, default=ROOT / "site")
+    parser.add_argument(
+        "--base-url",
+        default="https://mskazemi.com/aobench/",
+        help="Expected public base URL for this build; normalized to one trailing slash.",
+    )
     args = parser.parse_args()
 
     if not args.site_dir.is_dir():
@@ -306,12 +347,14 @@ def main() -> int:
         )
         return 1
 
-    checker = Checker(args.site_dir)
+    checker = Checker(args.site_dir, args.base_url)
     checker.check_robots()
     checker.check_sitemap()
     checker.check_llms_txt()
+    checker.check_template_leaks()
     for page in KEY_PAGES:
         checker.check_page(page)
+    checker.check_sitewide_base_urls()
     checker.check_sitewide_structured_data()
 
     if checker.failures: