diff --git a/README.md b/README.md
index a277d4c..b6dffba 100644
--- a/README.md
+++ b/README.md
@@ -25,7 +25,7 @@
 
 </div>
 
-![Terminal demo: Python fit() code and verified 2,366 to 1,376 token savings from the first frame, followed by Core-first processing and optional Tokenfold Select. Select illustration only.](https://raw.githubusercontent.com/snchimata/tokenfold/main/docs/assets/tokenfold-demo.gif)
+![Terminal demo: Python fit() code and verified 2,366 to 1,376 token savings from the first frame, followed by Core-first processing and optional Tokenfold Select. Select illustration only.](https://raw.githubusercontent.com/snchimata/tokenfold/a5aafaf9fb48047ed03e42e9cc379de49118b25d/docs/assets/tokenfold-demo.gif)
 
 <details>
 <summary><strong>Demo scope and regeneration</strong></summary>
diff --git a/python-tests/test_readme_layout.py b/python-tests/test_readme_layout.py
index cc34dba..de3d0d6 100644
--- a/python-tests/test_readme_layout.py
+++ b/python-tests/test_readme_layout.py
@@ -15,6 +15,11 @@ class ReadmeLayout(unittest.TestCase):
     def setUp(self):
         self.text = (ROOT / 'README.md').read_text(encoding='utf-8')
 
+    def test_demo_url_is_pinned_to_an_immutable_commit(self):
+        url = re.search(r'!\[Terminal demo:[^\]]+\]\(([^)]+)\)', self.text).group(1)
+        self.assertRegex(url, r'^https://raw\.githubusercontent\.com/snchimata/tokenfold/'
+                              r'[0-9a-f]{40}/docs/assets/tokenfold-demo\.gif$')
+
     def test_common_path_and_visible_limits(self):
         self.assertTrue(self.text.isascii())
         self.assertEqual(self.text.count('Exact decode proves data recovery'), 1)