diff --git a/tests/test_site_r2_gate.py b/tests/test_site_r2_gate.py
new file mode 100644
index 00000000..3a08cfa8
--- /dev/null
+++ b/tests/test_site_r2_gate.py
@@ -0,0 +1,273 @@
+"""Gate of the second site round, read from the built site (website/dist).
+
+Each package of the round was checked on its own. These checks hold the
+merged site to what the round set out to fix: no sidebar label a number, no
+"Next" tip, a sidebar label that is the page's own title, a card's link icon
+kept out of the heading a search indexes, crumbs that lead to a tab or a
+section, phone tables stacked where the reading level asks for it, and every
+page under 15,000 elements and 600 KB of HTML. The integration items of the
+round are held too: the platform logo beside the title, a section for every
+letter of the emulators index, system links to the section of the system's
+own name, dates in words and the moved getting-started page.
+
+Skips when the site is not built.
+"""
+
+from __future__ import annotations
+
+import json
+import re
+import sys
+import unittest
+from html import unescape
+from pathlib import Path
+
+import yaml
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(REPO_ROOT / "scripts"))
+
+DIST = REPO_ROOT / "website" / "dist"
+DOCS = REPO_ROOT / "website" / "src" / "content" / "docs"
+SIDEBAR = REPO_ROOT / "website" / "src" / "generated" / "sidebar.json"
+BASE = "/retrobios/"
+
+PAGE_BYTES = 600_000
+PAGE_ELEMENTS = 15_000
+PHONE_COLUMN = 358
+READER_LEVELS = {"start", "setup"}
+# The size from which a table is a list scanned by column and keeps its
+# scroll (website/src/plugins/rehype-stack-tables.mjs, RECORD_ROWS).
+RECORD_ROWS = 20
+
+_SCRIPT = re.compile(r"<(script|style)\b[^>]*>.*?</\1>", re.S | re.I)
+_ELEMENT = re.compile(r"<[a-zA-Z][a-zA-Z0-9-]*[\s/>]")
+_TAGS = re.compile(r"<[^>]+>")
+_TABLE = re.compile(r"<table\b(?P<attrs>[^>]*)>(?P<body>.*?)</table>", re.S)
+_CURRENT = re.compile(r'<a\b[^>]*aria-current="page"[^>]*>(?P<label>.*?)</a>', re.S)
+_H1 = re.compile(r"<h1\b[^>]*>(?P<title>.*?)</h1>", re.S)
+_HEADING = re.compile(r"<h(?P<level>[2-6])\b[^>]*>(?P<inner>.*?)</h(?P=level)>", re.S)
+_CRUMBS = re.compile(r'<nav\b[^>]*aria-label="Breadcrumb"[^>]*>(?P<inner>.*?)</nav>', re.S)
+_HREF = re.compile(r'href="(?P<href>[^"]*)"')
+_TIP = re.compile(r'<aside\b[^>]*starlight-aside--tip[^>]*>.*?</aside>', re.S)
+
+
+def _text(markup: str) -> str:
+    return " ".join(unescape(_TAGS.sub(" ", markup)).split())
+
+
+def _pages() -> list[Path]:
+    return sorted(DIST.rglob("*.html"))
+
+
+def _url(path: Path) -> str:
+    rel = path.relative_to(DIST).as_posix()
+    if rel.endswith("index.html"):
+        return BASE + rel[: -len("index.html")]
+    return BASE + rel
+
+
+def _levels() -> dict[str, str]:
+    """Page address -> the level its front matter declares."""
+    levels: dict[str, str] = {}
+    for source in DOCS.rglob("*.md*"):
+        text = source.read_text(encoding="utf-8")
+        if not text.startswith("---\n"):
+            continue
+        front = yaml.safe_load(text[4:text.find("\n---\n", 4)]) or {}
+        stem = str(front.get("slug") or source.relative_to(DOCS).with_suffix("").as_posix())
+        stem = re.sub(r"(^|/)index$", "", stem)
+        levels[BASE + (stem + "/" if stem else "")] = str(front.get("level") or "")
+    return levels
+
+
+@unittest.skipUnless((DIST / "index.html").is_file(), "site not built")
+class EveryBuiltPage(unittest.TestCase):
+
+    @classmethod
+    def setUpClass(cls):
+        cls.pages = {path: path.read_text(encoding="utf-8") for path in _pages()}
+        sidebar = json.loads(SIDEBAR.read_text(encoding="utf-8"))
+        cls.tab_roots = {BASE + topic["link"].strip("/") + ("/" if topic["link"].strip("/") else "")
+                         for topic in sidebar["topics"]}
+        cls.group_pages = {group["href"] for group in sidebar["groups"].values() if group["href"]}
+        cls.levels = _levels()
+
+    def test_under_the_page_budgets(self):
+        heavy, crowded = [], []
+        for path, html in self.pages.items():
+            size = len(html.encode("utf-8"))
+            if size >= PAGE_BYTES:
+                heavy.append((path.relative_to(DIST).as_posix(), size))
+            elements = len(_ELEMENT.findall(_SCRIPT.sub("", html)))
+            if elements >= PAGE_ELEMENTS:
+                crowded.append((path.relative_to(DIST).as_posix(), elements))
+        self.assertFalse(heavy, f"pages of {PAGE_BYTES:,} bytes or more")
+        self.assertFalse(crowded, f"pages of {PAGE_ELEMENTS:,} elements or more")
+
+    def test_no_sidebar_label_is_a_number(self):
+        for path, html in self.pages.items():
+            sidebar = html.split('id="starlight__sidebar"', 1)[-1].split("</nav>", 1)[0]
+            labels = [_text(m) for m in re.findall(r"<a\b[^>]*>(.*?)</a>", sidebar, re.S)]
+            numbers = [label for label in labels if re.fullmatch(r"\d+", label)]
+            with self.subTest(page=path.relative_to(DIST).as_posix()):
+                self.assertFalse(numbers)
+
+    def test_no_tip_titled_next(self):
+        for path, html in self.pages.items():
+            for aside in _TIP.findall(html):
+                title = re.search(r'class="starlight-aside__title"[^>]*>(.*?)</p>', aside, re.S)
+                with self.subTest(page=path.relative_to(DIST).as_posix()):
+                    self.assertNotEqual(_text(title[1]) if title else "", "Next")
+
+    def test_the_sidebar_names_a_page_by_its_title(self):
+        for path, html in self.pages.items():
+            url = _url(path)
+            current = _CURRENT.search(html.split('id="starlight__sidebar"', 1)[-1])
+            title = _H1.search(html)
+            # A tab, or a group of the sidebar, opens on its overview.
+            if url in self.tab_roots or url in self.group_pages or not current or not title:
+                continue
+            with self.subTest(page=url):
+                self.assertEqual(_text(current["label"]), _text(title["title"]))
+
+    def test_a_card_link_icon_says_nothing_in_its_heading(self):
+        for path, html in self.pages.items():
+            for heading in _HEADING.finditer(html):
+                for link in re.findall(r'<a\b[^>]*class="rb-permalink"[^>]*>(.*?)</a>',
+                                       heading["inner"], re.S):
+                    with self.subTest(page=path.relative_to(DIST).as_posix()):
+                        self.assertEqual(_text(link), "")
+
+    def test_every_crumb_leads_to_a_tab_or_a_section(self):
+        for path, html in self.pages.items():
+            crumbs = _CRUMBS.search(html)
+            if not crumbs:
+                continue
+            for href in _HREF.findall(crumbs["inner"]):
+                with self.subTest(page=_url(path), href=href):
+                    self.assertTrue(href in self.tab_roots or href in self.group_pages
+                                    or "#" in href, href)
+
+    def test_phone_tables_on_reader_pages_are_stacked(self):
+        """A table of three columns or more, under RECORD_ROWS rows, on a
+        level 0 or 1 page is read one row at a time on a phone: it carries
+        data-rb-stack. A two-column table is a list of label and value."""
+        unstacked = []
+        for path, html in self.pages.items():
+            if self.levels.get(_url(path)) not in READER_LEVELS:
+                continue
+            for table in _TABLE.finditer(html):
+                head = table["body"].split("</thead>", 1)[0]
+                columns = head.count("<th")
+                rows = table["body"].split("</thead>", 1)[-1].count("<tr")
+                # A table the page fills when it runs (the file picker) is
+                # not the build's.
+                if not rows:
+                    continue
+                if columns >= 2 and rows < RECORD_ROWS and "data-rb-stack" not in table["attrs"]:
+                    unstacked.append((_url(path), _text(head)[:60]))
+        self.assertFalse(unstacked)
+
+    def test_no_page_dates_itself_with_a_raw_stamp(self):
+        for path, html in self.pages.items():
+            with self.subTest(page=_url(path)):
+                self.assertNotIn("Generated on", html)
+
+
+@unittest.skipUnless((DIST / "index.html").is_file(), "site not built")
+class TheIntegrationItems(unittest.TestCase):
+
+    def test_a_platform_page_draws_its_logo_beside_the_title(self):
+        from common import load_platform_registry
+
+        registry = load_platform_registry(str(REPO_ROOT / "platforms"))
+        with_logo = [key for key, entry in registry.items() if entry.get("logo")]
+        self.assertTrue(with_logo)
+        for key in with_logo:
+            page = DIST / "platforms" / key / "index.html"
+            if not page.is_file():
+                continue
+            html = page.read_text(encoding="utf-8")
+            with self.subTest(platform=key):
+                block = re.search(r'<div class="rb-title-logo[^"]*">.*?</div>', html, re.S)
+                self.assertIsNotNone(block)
+                self.assertRegex(block[0], r'<img src="/retrobios/assets/logos/')
+                self.assertNotIn("align=", html.split("<main", 1)[-1])
+
+    def test_every_emulator_letter_lands_on_its_section(self):
+        sidebar = json.loads(SIDEBAR.read_text(encoding="utf-8"))
+        html = (DIST / "emulators" / "index.html").read_text(encoding="utf-8")
+        ids = set(re.findall(r'\bid="([^"]+)"', html))
+        letters = {key: group for key, group in sidebar["groups"].items()
+                   if key.startswith("Explore/Emulators/")}
+        self.assertGreater(len(letters), 20)
+        for key, group in letters.items():
+            with self.subTest(group=key):
+                self.assertRegex(group["href"] or "", r"^/retrobios/emulators/#")
+                self.assertIn(group["href"].split("#", 1)[1], ids)
+
+    def test_system_links_reach_the_section_of_the_name(self):
+        """Every link into a system page lands on an id the page holds."""
+        ids: dict[str, set[str]] = {}
+        for page in (DIST / "systems").glob("*/index.html"):
+            ids[page.parent.name] = set(re.findall(r'\bid="([^"]+)"', page.read_text(encoding="utf-8")))
+        target = re.compile(r'href="/retrobios/systems/(?P<page>[^/"#]+)/#(?P<anchor>[^"]+)"')
+        for page in ("emulators/index.html", "platforms/retroarch/index.html",
+                     "platforms/batocera/index.html", "systems/index.html"):
+            html = (DIST / page).read_text(encoding="utf-8")
+            for match in target.finditer(html):
+                with self.subTest(page=page, link=match[0]):
+                    self.assertIn(match["anchor"], ids.get(match["page"], set()))
+
+    def test_getting_started_sends_to_which_pack(self):
+        page = DIST / "wiki" / "getting-started" / "index.html"
+        self.assertTrue(page.is_file())
+        self.assertIn('url=/retrobios/which-pack/"', page.read_text(encoding="utf-8"))
+        self.assertFalse((REPO_ROOT / "wiki" / "getting-started.md").exists())
+
+    def test_existence_mode_pages_name_the_file_served(self):
+        for key in ("retroarch", "lakka", "retropie"):
+            page = DIST / "platforms" / key / "index.html"
+            if not page.is_file():
+                continue
+            with self.subTest(platform=key):
+                self.assertIn("Served by", page.read_text(encoding="utf-8"))
+
+
+class TheRules(unittest.TestCase):
+    """What the gate reads without a build."""
+
+    def test_the_logo_reaches_the_front_matter(self):
+        import sitestarlight
+
+        _target, text = sitestarlight.finalize_page(
+            "platforms/x.md", "# X\n\nBody.\n", level="setup", weight=None,
+            browser_title="X", description="d", structured={}, terms={},
+            logo="assets/logos/x.svg")
+        self.assertIn("logo: assets/logos/x.svg", text)
+
+    def test_a_large_data_directory_is_listed_on_pages_of_its_own(self):
+        """Dolphin's Sys tree put 16,900 elements on the Dolphin page. Its
+        files are found where they moved."""
+        import generate_site
+
+        members = {f"Sys/{i:05d}.ini": {"sha1": f"{i:040x}", "size": 1}
+                   for i in range(generate_site.DATA_TREE_ROWS_PER_PAGE + 1)}
+        pages = generate_site.data_tree_pages("dolphin-sys", members)
+        self.assertEqual([stem for stem, _names in pages], ["data-dolphin-sys", "data-dolphin-sys-2"])
+        self.assertEqual(generate_site.data_tree_pages("small", dict(list(members.items())[:3])), [])
+        items = generate_site.file_location_items(
+            {"files": {}}, {"dolphin": {"data_directories": [{"ref": "dolphin-sys"}]}},
+            {"dolphin-sys": members})
+        last = f"{generate_site.DATA_TREE_ROWS_PER_PAGE:040x}"
+        self.assertIn({"sha1": last, "url": f"emulators/data-dolphin-sys-2/#sha1-{last[:12]}"}, items)
+
+    def test_no_float_rule_for_right_aligned_images(self):
+        css = (REPO_ROOT / "website" / "src" / "styles" / "retrobios.css").read_text(encoding="utf-8")
+        self.assertNotIn("img[align='right']", css)
+        self.assertIn(":is(h2, h3) small", css)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/wiki/testing-guide.md b/wiki/testing-guide.md
index abdbeb0d..158e1673 100644
--- a/wiki/testing-guide.md
+++ b/wiki/testing-guide.md
@@ -55,7 +55,7 @@ Every module in `tests/`, by the area it guards.
 | Handheld port | stubbed PortMaster dialogs, fake installer | `test_port` |
 | Truth and native export | synthetic + cached originals | `test_export_counts`, `test_export_native_layer`, `test_export_report`, `test_exporters`, `test_truth_data_dir_readers`, `test_truth_diff`, `test_truth_layout` |
 | Scrapers, upstream and provenance | inline sources, synthetic | `test_check_freshness`, `test_dedup`, `test_elec_scrapers`, `test_fbneo_parser`, `test_find_pinned_revision`, `test_hash_merge`, `test_launchers`, `test_launchers_emudeck`, `test_launchers_recalbox`, `test_launchers_retrobat`, `test_launchers_retrodeck`, `test_launchers_retropie`, `test_launchers_rocknix`, `test_mame_devices`, `test_mame_parser`, `test_mame_ref_audit`, `test_profile_sync`, `test_provenance`, `test_refresh_data_dirs`, `test_refresh_failures`, `test_refresh_stale`, `test_retrodeck_scraper_errors`, `test_romset_recipes`, `test_scraper_contract`, `test_upstream`, `test_upstream_index` |
-| Site, docs and repository contracts | rendered pages, sources | `test_contributing_text`, `test_coverage_cache`, `test_data_dir_refs`, `test_dataclass_defaults`, `test_doc_pages`, `test_doc_surfaces`, `test_optional_dependencies`, `test_prose_style`, `test_schema_semantic_gate`, `test_site_availability`, `test_site_content_data`, `test_site_copy`, `test_site_cores`, `test_site_emulator_page`, `test_site_exports`, `test_site_funnel`, `test_site_images`, `test_site_picker`, `test_site_platform_hashes`, `test_site_r2_components`, `test_site_r2_css`, `test_site_r2_emulators`, `test_site_r2_entry`, `test_site_r2_nav`, `test_site_r2_p0`, `test_site_r2_reference`, `test_site_r2_systems`, `test_site_r2_wiki`, `test_site_render`, `test_site_system_links`, `test_site_validation`, `test_site_weight`, `test_site_write`, `test_validate_pr`, `test_validate_pr_flow`, `test_validate_pr_hashes`, `test_workflow_paths` |
+| Site, docs and repository contracts | rendered pages, sources | `test_contributing_text`, `test_coverage_cache`, `test_data_dir_refs`, `test_dataclass_defaults`, `test_doc_pages`, `test_doc_surfaces`, `test_optional_dependencies`, `test_prose_style`, `test_schema_semantic_gate`, `test_site_availability`, `test_site_content_data`, `test_site_copy`, `test_site_cores`, `test_site_emulator_page`, `test_site_exports`, `test_site_funnel`, `test_site_images`, `test_site_picker`, `test_site_platform_hashes`, `test_site_r2_components`, `test_site_r2_css`, `test_site_r2_emulators`, `test_site_r2_entry`, `test_site_r2_gate`, `test_site_r2_nav`, `test_site_r2_p0`, `test_site_r2_reference`, `test_site_r2_systems`, `test_site_r2_wiki`, `test_site_render`, `test_site_system_links`, `test_site_validation`, `test_site_weight`, `test_site_write`, `test_validate_pr`, `test_validate_pr_flow`, `test_validate_pr_hashes`, `test_workflow_paths` |
 
 ## Test architecture