diff --git a/web/index.html b/web/index.html
index 966390c..f4d885b 100644
--- a/web/index.html
+++ b/web/index.html
@@ -241,5 +241,35 @@ <h2>Jobs</h2>
 }
 loadJobs();setInterval(loadJobs,4000);
 </script>
+<script>
+// Shared-backend contention hint: the L4 llama-server has 3 slots shared across
+// dataroom / KG / searchbox / llamaui. Poll the public slots endpoint (CORS-open)
+// and surface a banner when other work is in flight, so a queued run is expected.
+(function(){
+  var SLOTS_URL="https://hanxiao.io/llama-api/slots";
+  var el=document.createElement("div");
+  el.id="slotbar";
+  el.style.cssText="display:none;margin-top:14px;border:1px solid var(--line);padding:8px 12px;font-size:12px;background:var(--fill);color:var(--fg)";
+  var mh=document.querySelector(".masthead");
+  if(mh&&mh.parentNode) mh.parentNode.insertBefore(el, mh.nextSibling);
+  else document.body.insertBefore(el, document.body.firstChild);
+  async function poll(){
+    try{
+      var r=await fetch(SLOTS_URL,{cache:"no-store"});
+      var s=await r.json();
+      var busy=s.filter(function(x){return x.is_processing;}).length;
+      var total=s.length;
+      if(busy>=1){
+        el.style.display="block";
+        var warn=busy>=total;
+        el.innerHTML="● shared L4 backend: <b>"+busy+"/"+total+"</b> slots busy"+(warn?" — all slots in use, new runs will queue":" — a concurrent run may slow generation");
+        el.style.borderColor=warn?"#b00":"var(--line)";
+      }else{ el.style.display="none"; }
+    }catch(e){ el.style.display="none"; }
+  }
+  poll();setInterval(poll,4000);
+})();
+</script>
+
 </body>
 </html>