diff --git a/docs/evidence/agent-desk-title-canonical-2026-08-12.md b/docs/evidence/agent-desk-title-canonical-2026-08-12.md
new file mode 100644
index 00000000..352abe58
--- /dev/null
+++ b/docs/evidence/agent-desk-title-canonical-2026-08-12.md
@@ -0,0 +1,111 @@
+# Retired Agent Desk title/snippet: stop claiming the apex root as the canonical
+
+Date: 2026-08-12
+Scope: `public/agent-desk.html` (head canonical/og:url), `scripts/check-site.mjs`
+(regression guard). This receipt records a deterministic, repository-side
+change to what the legacy surface declares about itself. It is not a live
+Google-index measurement, and it claims nothing about when or whether Google's
+search results change after this change.
+
+## Why this pass exists
+
+Google has presented the retired self-serve "TinyStudio Agent Desk"
+title/snippet for tinystudio.io. Two earlier repository-side passes already
+addressed the legacy surface's own head
+(`docs/evidence/agent-desk-retired-title-2026-08-09.md` added `noindex,
+nofollow` and retired framing; PR #100 pointed the retired `app.`/`api.` 410
+hosts at The Website Appraisal), but one index-facing wire remained: the
+legacy page still declared the homepage as its canonical.
+
+- `public/agent-desk.html`, served at `/agent-desk` and `/agent-desk.html`
+  (worker `PUBLIC_ASSET_PATHS`), still carried
+  `<link rel="canonical" href="https://tinystudio.io/" />` and
+  `<meta property="og:url" content="https://tinystudio.io/" />`.
+- The 2026-08-09 receipt itself diagnosed the original mechanism: the legacy
+  page's title "consolidated through its canonical" onto the homepage URL
+  (`evidence-fixtures/ai-search/evidence.json`, q5 / google, 2026-08-06:
+  "tinystudio.io - TinyStudio Agent Desk" attached to the homepage URL).
+  The de-indexing pass cut the title and description wires but left the
+  canonical wire in place.
+- Live-before state (2026-08-12): `curl -s https://tinystudio.io/agent-desk`
+  returned 200 with `noindex, nofollow`, the retired title/description — and
+  still declared `https://tinystudio.io/` as its canonical and og:url. A
+  noindex page whose canonical names a live page tells Google it is a
+  duplicate of that page: the same consolidation that handed the retired
+  title to tinystudio.io, re-declared on every recrawl until the removal is
+  processed. Google's own guidance is that a page you do not want indexed
+  should not canonicalize to another URL.
+
+Every other owned page follows one convention — the canonical and og:url name
+the page's own served `.html` twin (`/agents.html`, `/pricing.html`,
+`/specimen.html`, and the audit page's clean `/audit`) — and the legacy page
+was the only one claiming the apex root.
+
+## What changed
+
+1. `public/agent-desk.html` head: the canonical and og:url now name the
+   legacy page itself (`https://tinystudio.io/agent-desk.html`, the same
+   self-address form the desk, pricing and specimen pages use) instead of the
+   homepage. The consolidation wire is cut: no page declares the retired
+   surface as a duplicate of `tinystudio.io/` any more, so a recrawl can no
+   longer merge the retired page's signals or title onto the homepage URL.
+2. `scripts/check-site.mjs`: the retired-desk dogfood guard now fails `npm
+   run check` when the legacy page's canonical is missing, duplicated, or
+   points anywhere but `https://tinystudio.io/agent-desk.html`, and when its
+   og:url points anywhere else — so the apex-root claim cannot silently
+   return. (Regression verified by reverting the HTML change: both
+   assertions fire on the old apex-root values.)
+
+## What deliberately did not change
+
+- The visible legacy UI copy and the tool itself: `agent-desk.html` remains a
+  working self-serve surface, still served at `/agent-desk` and
+  `/agent-desk.html`, still `noindex, nofollow`, still retired-framed. Old
+  deep links and bookmarks keep working.
+- `robots.txt` (no Disallow was added: a disallowed page cannot be crawled to
+  see the noindex meta, so that would slow de-indexing, not speed it up),
+  `sitemap.xml`, the worker's routing, and the 410 hosts (already corrected
+  by PR #100, verified live 2026-08-12).
+- The homepage and the other public pages: their own titles, descriptions and
+  canonicals were already current; this pass only fixes the legacy surface.
+- No redirect was added for `/agent-desk`: the page stays reachable, it is
+  just no longer indexable or canonicalizable onto the homepage.
+
+## What is tested and what is not
+
+Tested: the extended guard fails loudly on regression (the canonical and the
+og:url are asserted directly), and the full suite passes on this commit —
+`npm run check` green, `npm test` green (92 subtests: 6 headings, 7 sitemap,
+55 worker, 16 UI, 8 product contract), exit code 0.
+
+Not tested: whether or when Google drops the old title/snippet from its
+results. De-indexing and re-consolidation are live processes that start only
+after Google recrawls the pages; nothing here implies a ranking, visibility,
+lead or revenue outcome. The honest measure is a future controlled re-run
+recorded through `evidence-fixtures/ai-search` (or Search Console), once
+enough time has passed for a recrawl.
+
+## Verification (reproduce)
+
+```
+npm run check
+npm test
+git diff --check
+```
+
+All three checks pass on this commit. Live-before state (2026-08-12):
+
+```
+curl -s https://tinystudio.io/agent-desk   # noindex + retired head, canonical/og:url still -> homepage
+curl -s https://app.tinystudio.io/         # 410, already names The Website Appraisal (PR #100)
+curl -s https://api.tinystudio.io/         # 410, already names The Website Appraisal (PR #100)
+```
+
+After deploy, the same fetch of `/agent-desk` should show the self-canonical
+`https://tinystudio.io/agent-desk.html`.
+
+## Closeout
+
+This closes the repository-side item against current main: the legacy Agent
+Desk surface no longer declares itself a duplicate of the homepage through
+its canonical or og:url, and CI fails if either returns.
diff --git a/public/agent-desk.html b/public/agent-desk.html
index 4f68447e..12564494 100644
--- a/public/agent-desk.html
+++ b/public/agent-desk.html
@@ -15,8 +15,8 @@
       content="The self-serve Agent Desk is retired and is not the current offer. TinyStudio's current offer is the free leak audit of high-ticket service homepages."
     />
     <meta property="og:type" content="website" />
-    <meta property="og:url" content="https://tinystudio.io/" />
-    <link rel="canonical" href="https://tinystudio.io/" />
+    <meta property="og:url" content="https://tinystudio.io/agent-desk.html" />
+    <link rel="canonical" href="https://tinystudio.io/agent-desk.html" />
     <meta property="og:image" content="https://tinystudio.io/og-image.png" />
     <meta property="og:image:width" content="1200" />
     <meta property="og:image:height" content="630" />
diff --git a/scripts/check-site.mjs b/scripts/check-site.mjs
index 0e301463..ca219fda 100644
--- a/scripts/check-site.mjs
+++ b/scripts/check-site.mjs
@@ -1115,9 +1115,11 @@ for (const [pageName, pageHtml] of ownedPages) {
 // google (2026-08-06): title "tinystudio.io - TinyStudio Agent Desk" against
 // the homepage URL, the legacy page's title consolidated through its canonical.
 // The legacy page must therefore stay out of the index: its head keeps a
-// robots noindex, nofollow meta, and its title and description frame the
-// surface as retired, so neither the search index nor a scraper can re-present
-// the retired self-serve product as the current offer.
+// robots noindex, nofollow meta, its title and description frame the
+// surface as retired, and its canonical and og:url name the page's own
+// served .html address instead of the apex root — so neither the search
+// index nor a scraper can re-present the retired self-serve product as the
+// current offer or fold its identity into the homepage.
 const retiredDeskHead = retiredDesk.match(/<head\b[^>]*>([\s\S]*?)<\/head>/i)?.[1] ?? "";
 const robotsMeta = /<meta\b(?=[^>]*\bname="robots")(?=[^>]*\bcontent="noindex,\s*nofollow")[^>]*>/i;
 if (!robotsMeta.test(retiredDeskHead)) {
@@ -1131,6 +1133,42 @@ const retiredDeskDescription =
 if (!/\bretired\b/i.test(retiredDeskDescription)) {
   failures.push("Retired Agent Desk description must frame the surface as retired.");
 }
+// The legacy page must not claim the apex root as its canonical: while the
+// page was served at the root, its canonical and og:url pointed at
+// https://tinystudio.io/, which is exactly how the retired title consolidated
+// onto the homepage (the q5 evidence above). A noindex page canonicalizing to
+// a live page tells Google it is a duplicate of that page. The legacy page
+// now declares its own canonical .html twin — https://tinystudio.io/agent-desk.html,
+// the same self-address form every other owned page canonicals to (the desk
+// page names /agents.html, pricing names /pricing.html, specimen names
+// /specimen.html) — so neither a search engine nor a scraper can fold the
+// retired surface's identity into the homepage again.
+const retiredDeskLiveHead = retiredDeskHead.replace(/<!--[\s\S]*?-->/g, "");
+const retiredDeskCanonicalLinks = [
+  ...retiredDeskLiveHead.matchAll(/<link\b[^>]*\brel\s*=\s*["']canonical["'][^>]*>/gi)
+].map((match) => match[0]);
+const retiredDeskCanonicalHref =
+  retiredDeskCanonicalLinks[0]?.match(/\bhref\s*=\s*["']([^"']*)["']/i)?.[1] ?? "";
+if (retiredDeskCanonicalLinks.length !== 1) {
+  failures.push(
+    `Retired Agent Desk page must carry exactly one canonical link in its head (found ${retiredDeskCanonicalLinks.length}).`
+  );
+} else if (retiredDeskCanonicalHref.trim() !== "https://tinystudio.io/agent-desk.html") {
+  failures.push(
+    `Retired Agent Desk canonical must point at its own address https://tinystudio.io/agent-desk.html, not the apex root (found ${JSON.stringify(retiredDeskCanonicalHref)}).`
+  );
+}
+const retiredDeskOgUrl =
+  retiredDeskLiveHead.match(/<meta\b[^>]*\bproperty\s*=\s*["']og:url["'][^>]*>/i)?.[0] ?? "";
+const retiredDeskOgUrlContent =
+  retiredDeskOgUrl.match(/\bcontent\s*=\s*["']([^"']*)["']/i)?.[1] ?? "";
+if (!retiredDeskOgUrlContent) {
+  failures.push("Retired Agent Desk page must keep an og:url meta in its head.");
+} else if (retiredDeskOgUrlContent.trim() !== "https://tinystudio.io/agent-desk.html") {
+  failures.push(
+    `Retired Agent Desk og:url must be its own address https://tinystudio.io/agent-desk.html, not the apex root (found ${JSON.stringify(retiredDeskOgUrlContent)}).`
+  );
+}
 
 // ---- Meta descriptions (dogfood) -------------------------------------------
 // The leak audit this site sells flags a homepage whose served HTML carries no