[grade=B] fix: stop aggressive API polling and SSE reconnect when auth lost

When unauthenticated (TOTP gate active), the dashboard was polling
/api/v1/services/status every few seconds and reconnecting SSE every
2-8s indefinitely. Now: 401/403 sets _dcAuthLost flag to skip the
polling interval entirely (no fallback to direct probes which
misleadingly mark services as healthy), SSE exposes _sseClose()
coordinated with _dcAuthLost, and queued refreshes are guarded.

status/js/core/grid.js

@@ -349,6 +349,9 @@
   }
 
   async function refreshAll() {
+    // Skip if auth has been lost (e.g. TOTP gate activated externally).
+    // The polling interval in init.js also checks this flag.
+    if (window._dcAuthLost) return;
     if (refreshInFlight) {
       refreshQueued = true;
       return refreshInFlight;
@@ -402,9 +405,15 @@
       try {
         const response = await fetch('/api/v1/services/status', { cache: 'no-store' });
         if (response.status === 401 || response.status === 403) {
-          // Auth lost — stop the polling loop from hammering every few seconds
+          // Auth lost — stop the polling loop and close SSE; do NOT fall
+          // through to direct probes (those would misleadingly mark
+          // services as healthy since /probe/ treats 401/403 as "up").
           window._dcAuthLost = true;
-          throw new Error(`Authentication required (${response.status})`);
+          if (window._sseReconnect && window._sseClose) {
+            window._sseClose(); // tell SSE to stop reconnecting
+          }
+          updateStamp('auth required');
+          return; // skip the fallback entirely
         }
         if (!response.ok) {
           throw new Error(`Status refresh failed (${response.status})`);
@@ -424,9 +433,11 @@
         }
       } finally {
         refreshInFlight = null;
-        if (refreshQueued) {
+        if (refreshQueued && !window._dcAuthLost) {
           refreshQueued = false;
           setTimeout(() => { window.refreshAll(); }, 0);
+        } else {
+          refreshQueued = false;
         }
       }
     })();

status/js/live-events.js

@@ -3,9 +3,12 @@
   let es = null;
   let reconnectDelay = 1000;
   const MAX_RECONNECT = 30000;
+  let _sseFailCount = 0;
+  let _sseManuallyClosed = false;
 
   function connect() {
     if (es) { try { es.close(); } catch (_) {} }
+    if (_sseManuallyClosed) return; // auth-lost: don't reconnect
 
     es = new EventSource('/api/v1/events/stream');
 
@@ -102,11 +105,16 @@
     // Reconnect on error
     es.onerror = () => {
       es.close();
-      // Stop reconnecting entirely after multiple consecutive failures
+      // If auth was explicitly lost (401/403 from the polling loop),
-      // (likely auth-gate / session expired — no point hammering forever)
+      // don't attempt reconnection at all.
-      _sseFailCount = (_sseFailCount || 0) + 1;
+      if (window._dcAuthLost || _sseManuallyClosed) {
+        console.warn('[SSE] Auth lost — stopping reconnection');
+        return;
+      }
+      // Transient failures: retry with exponential backoff, stop after 5
+      _sseFailCount++;
       if (_sseFailCount > 5) {
-        console.warn('[SSE] Max reconnect attempts reached — stopping (auth gate active or server unreachable)');
+        console.warn('[SSE] Max reconnect attempts reached — stopping (server unreachable)');
         return;
       }
       console.warn(`[SSE] Disconnected, reconnecting in ${reconnectDelay / 1000}s...`);
@@ -115,11 +123,16 @@
     };
   }
 
-  let _sseFailCount = 0;
+  // Called by grid.js when the polling loop detects auth loss (401/403)
+  function closeAndStop() {
+    _sseManuallyClosed = true;
+    if (es) { try { es.close(); } catch (_) {} }
+  }
 
   // Start on page load
   connect();
 
-  // Expose for debugging
+  // Expose for debugging and cross-module coordination
   window._sseReconnect = connect;
+  window._sseClose = closeAndStop;
 })();