diff --git a/surfaces/shell/src/display/sw.ts b/surfaces/shell/src/display/sw.ts
index 364ee27..6385a9a 100644
--- a/surfaces/shell/src/display/sw.ts
+++ b/surfaces/shell/src/display/sw.ts
@@ -1,5 +1,6 @@
 /// <reference lib="webworker" />
 import {
+  classifyDisplayRequest,
   DISPLAY_CACHE_NAMES,
   type DisplayOfflineEnvironment,
   type DisplayOfflineStatus,
@@ -56,14 +57,23 @@ self.addEventListener("activate", (event) => {
 });
 
 self.addEventListener("fetch", (event) => {
-  const handled = handleDisplayFetch(event.request, self.location.origin, environment);
+  // Classified synchronously, and `respondWith` is called **only** for what this worker actually
+  // handles. Answering every request — even by re-issuing it — would route the whole page through
+  // the worker, re-send request bodies from a different context, and make the worker's presence
+  // observable on requests it has no interest in. Not responding leaves the browser doing exactly
+  // what it would have done without a worker at all.
+  if (!classifyDisplayRequest(event.request, self.location.origin)) {
+    return;
+  }
   event.respondWith(
-    handled.then(
-      (response) =>
-        // `undefined` means this worker recognized nothing and must not change what the browser would
-        // have done. Re-issuing the request is the only way to say that from inside `respondWith`.
-        response ?? fetch(event.request),
-    ),
+    handleDisplayFetch(event.request, self.location.origin, environment).then((response) => {
+      if (!response) {
+        // Unreachable: the classification above is the same one `handleDisplayFetch` performs. A
+        // network error is the honest answer if that ever stops being true.
+        throw new TypeError("Ekklesia display request was not handled.");
+      }
+      return response;
+    }),
   );
 });