diff --git a/publish_episode.py b/publish_episode.py
index 4308795..c9a8c7c 100755
--- a/publish_episode.py
+++ b/publish_episode.py
@@ -1871,10 +1871,6 @@ def main():
     shutil.copy2(str(transcript_path), str(website_transcript_path))
     print(f"    Transcript copied to website/transcripts/")
 
-    # Build this episode's static page and regenerate the sitemap wholesale.
-    # generate_episode_pages.py is the only writer of sitemap.xml.
-    regenerate_website_pages()
-
     # Sync any remaining episode media to BunnyCDN (cover art, etc.)
     print("    Syncing remaining episode media to CDN...")
     sync_episode_media_to_bunny(episode["id"], uploaded_keys)
@@ -1891,6 +1887,15 @@ def main():
         else:
             raise
 
+    # Build this episode's static page and regenerate the sitemap wholesale.
+    # generate_episode_pages.py is the only writer of sitemap.xml.
+    #
+    # MUST run after the publish above: the generator is driven entirely by the
+    # RSS feed, so running it earlier silently omits the episode being published
+    # and its page only appears on the NEXT publish. Social posts link to
+    # /episode/<slug>/, so that left every launch-day link 404ing.
+    regenerate_website_pages()
+
     # Step 5: Deploy website (transcript + sitemap must be live before social links go out)
     print("[5/5] Deploying website...")
     project_dir = Path(__file__).parent
diff --git a/tests/test_publish_regenerates_pages.py b/tests/test_publish_regenerates_pages.py
index 41da4b3..73dc6d3 100644
--- a/tests/test_publish_regenerates_pages.py
+++ b/tests/test_publish_regenerates_pages.py
@@ -78,3 +78,39 @@ def test_publish_flow_calls_it_after_copying_the_transcript():
     copy_at = source.index("Transcript copied to website/transcripts/")
     call_at = source.index("regenerate_website_pages()", copy_at)
     assert call_at > copy_at
+
+
+def test_generator_runs_after_the_publish_call():
+    """Ordering regression guard.
+
+    generate_episode_pages.py is driven entirely by the RSS feed, so it only
+    sees an episode once Castopod has published it. For 58 episodes the
+    regenerate call sat at step 3.7 — before the step-4 publish — so the
+    episode being published was silently absent from its own run and its page
+    only appeared on the NEXT publish. Social posts link to /episode/<slug>/,
+    so every launch-day link 404'd.
+
+    Asserted on source order because the surrounding publish flow does network
+    and filesystem work that isn't practical to drive end to end here.
+    """
+    src = Path(publish_episode.__file__).read_text()
+    body = src[src.index("def main("):]
+
+    publish_call = body.index("published = publish_episode(")
+    regen_call = body.index("regenerate_website_pages()")
+    deploy_call = body.index("wrangler")
+
+    assert publish_call < regen_call, (
+        "regenerate_website_pages() must run AFTER publish_episode(), or the new "
+        "episode is missing from the feed the generator reads"
+    )
+    assert regen_call < deploy_call, (
+        "regenerate_website_pages() must run BEFORE the wrangler deploy, or the "
+        "freshly built page never ships"
+    )
+
+
+def test_regenerate_is_called_exactly_once_in_main():
+    src = Path(publish_episode.__file__).read_text()
+    body = src[src.index("def main("):]
+    assert body.count("regenerate_website_pages()") == 1