Deploy: Stack A (NIM brain + Maverick judge + Sarvam voice). D-019.

40-data/llm_health.json

@@ -1,22 +1,16 @@
 {
-  "updated_at": "2026-05-15T01:41:07Z",
+  "updated_at": "2026-05-15T01:46:37Z",
   "models": {
     "qwen/qwen3-next-80b-a3b-instruct": {
       "model": "qwen/qwen3-next-80b-a3b-instruct",
       "status": "healthy",
-      "last_success_at": "2026-05-15T01:41:07Z",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": null,
       "last_error": null,
-      "latency_ms": 289,
+      "latency_ms": 2573,
       "consecutive_failures": 0,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": true,
-          "latency_ms": 278,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": true,
           "latency_ms": 282,
@@ -40,6 +34,12 @@
           "latency_ms": 289,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 2573,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -53,18 +53,12 @@
       "model": "qwen/qwen3.5-122b-a10b",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "timeout",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 8002,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 8001,
@@ -88,6 +82,12 @@
           "latency_ms": 8002,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 8001,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -101,18 +101,12 @@
       "model": "openai/gpt-oss-120b",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "empty_content",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 411,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 269,
@@ -136,6 +130,12 @@
           "latency_ms": 406,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 473,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -148,19 +148,13 @@
     "mistralai/mistral-large-3-675b-instruct-2512": {
       "model": "mistralai/mistral-large-3-675b-instruct-2512",
       "status": "healthy",
-      "last_success_at": "2026-05-15T01:41:07Z",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": "2026-05-15T01:27:51Z",
       "last_error": null,
-      "latency_ms": 615,
+      "latency_ms": 449,
       "consecutive_failures": 0,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 8003,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": true,
           "latency_ms": 378,
@@ -184,6 +178,12 @@
           "latency_ms": 615,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 449,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -196,19 +196,13 @@
     "nvidia/llama-3.3-nemotron-super-49b-v1.5": {
       "model": "nvidia/llama-3.3-nemotron-super-49b-v1.5",
       "status": "healthy",
-      "last_success_at": "2026-05-15T01:41:07Z",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": "2026-05-15T01:24:45Z",
       "last_error": null,
-      "latency_ms": 304,
+      "latency_ms": 307,
       "consecutive_failures": 0,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": true,
-          "latency_ms": 391,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": true,
           "latency_ms": 359,
@@ -232,6 +226,12 @@
           "latency_ms": 304,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 307,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -243,20 +243,14 @@
     },
     "meta/llama-3.3-70b-instruct": {
       "model": "meta/llama-3.3-70b-instruct",
-      "status": "degraded",
-      "last_success_at": "2026-05-15T01:38:54Z",
+      "status": "healthy",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": "2026-05-15T01:41:07Z",
-      "last_error": "http_429",
-      "latency_ms": 511,
-      "consecutive_failures": 2,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "last_error": null,
+      "latency_ms": 210,
+      "consecutive_failures": 0,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": true,
-          "latency_ms": 411,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": true,
           "latency_ms": 1223,
@@ -280,6 +274,12 @@
           "latency_ms": 34,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 210,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -293,18 +293,12 @@
       "model": "deepseek-ai/deepseek-v4-pro",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "timeout",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 8002,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 8001,
@@ -328,6 +322,12 @@
           "latency_ms": 8002,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 8001,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -341,18 +341,12 @@
       "model": "nvidia/nemotron-3-nano-30b-a3b",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "empty_content",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 264,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 1730,
@@ -376,6 +370,12 @@
           "latency_ms": 275,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 268,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -389,18 +389,12 @@
       "model": "deepseek-ai/deepseek-v4-flash",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "timeout",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 8001,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 8002,
@@ -424,6 +418,12 @@
           "latency_ms": 8002,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 8002,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -437,18 +437,12 @@
       "model": "moonshotai/kimi-k2-instruct-0905",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "http_404",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 84,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 92,
@@ -472,6 +466,12 @@
           "latency_ms": 94,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 84,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -485,18 +485,12 @@
       "model": "minimaxai/minimax-m2.5",
       "status": "down",
       "last_success_at": null,
-      "last_failure_at": "2026-05-15T01:41:07Z",
+      "last_failure_at": "2026-05-15T01:46:37Z",
       "last_error": "http_410",
       "latency_ms": null,
-      "consecutive_failures": 15,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "consecutive_failures": 16,
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 31,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 32,
@@ -520,6 +514,12 @@
           "latency_ms": 33,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": false,
+          "latency_ms": 31,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -532,19 +532,13 @@
     "meta/llama-4-maverick-17b-128e-instruct": {
       "model": "meta/llama-4-maverick-17b-128e-instruct",
       "status": "healthy",
-      "last_success_at": "2026-05-15T01:41:07Z",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": "2026-05-14T01:03:07Z",
       "last_error": null,
-      "latency_ms": 314,
+      "latency_ms": 309,
       "consecutive_failures": 0,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": true,
-          "latency_ms": 202,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": true,
           "latency_ms": 1087,
@@ -568,6 +562,12 @@
           "latency_ms": 314,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 309,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -579,20 +579,14 @@
     },
     "openrouter:openai/gpt-oss-120b": {
       "model": "openrouter:openai/gpt-oss-120b",
-      "status": "degraded",
-      "last_success_at": "2026-05-15T01:41:07Z",
+      "status": "healthy",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": "2026-05-15T01:38:54Z",
       "last_error": null,
-      "latency_ms": 5821,
+      "latency_ms": 1349,
       "consecutive_failures": 0,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": false,
-          "latency_ms": 599,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": false,
           "latency_ms": 912,
@@ -616,6 +610,12 @@
           "latency_ms": 5821,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 1349,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,
@@ -628,19 +628,13 @@
     "groq:llama-3.3-70b-versatile": {
       "model": "groq:llama-3.3-70b-versatile",
       "status": "healthy",
-      "last_success_at": "2026-05-15T01:41:07Z",
+      "last_success_at": "2026-05-15T01:46:37Z",
       "last_failure_at": null,
       "last_error": null,
-      "latency_ms": 258,
+      "latency_ms": 157,
       "consecutive_failures": 0,
-      "tested_at": "2026-05-15T01:41:07Z",
+      "tested_at": "2026-05-15T01:46:37Z",
       "probe_history": [
-        {
-          "ok": true,
-          "latency_ms": 348,
-          "ts": "2026-05-15T01:27:51Z",
-          "src": "probe"
-        },
         {
           "ok": true,
           "latency_ms": 332,
@@ -664,6 +658,12 @@
           "latency_ms": 258,
           "ts": "2026-05-15T01:41:07Z",
           "src": "probe"
+        },
+        {
+          "ok": true,
+          "latency_ms": 157,
+          "ts": "2026-05-15T01:46:37Z",
+          "src": "probe"
         }
       ],
       "degraded_until_monotonic": 0.0,

frontend/src/app/page.tsx

@@ -197,12 +197,13 @@ export default function Page() {
   // actual mic state, PTT temporarily flips it to false during a recording
   // even while userPrefersLive stays true (so Live resumes after PTT).
   const [userPrefersLive, setUserPrefersLive] = useState(false);
-  // Load persisted preference on mount. Only "on" persists; default and
-  // explicit "off" both render grey.
+  // KI-131 (2026-05-15) — voice is OFF by default. Previously persisted
+  // "on" preferences are wiped on next load so users with a stale green
+  // pill from before this change also see grey and have to re-opt in.
+  // This amends ADR-028 "Default ON" — see ADR-033 for rationale.
   useEffect(() => {
     if (typeof window === "undefined") return;
-    const pref = localStorage.getItem("insurance_live_pref");
-    if (pref === "on") setUserPrefersLive(true);
+    localStorage.removeItem("insurance_live_pref");
   }, []);
   // Persist + sync to the live hook whenever the user toggles preference.
   useEffect(() => {

frontend/src/lib/useLiveConversation.ts

@@ -536,8 +536,17 @@ export function useLiveConversation(opts: LiveConversationOptions): LiveConversa
           node.connect(silentSink);
           silentSink.connect(ctx.destination);
         } catch (e) {
+          // KI-131 (2026-05-15) — previously, a worklet-setup failure left
+          // setMicPermissionDenied=false and live.live=true so the UI pill
+          // stayed green ("Voice on — just speak") while no PCM frames were
+          // ever delivered — silent functional break. Now flip both states
+          // so the pill correctly switches to "🔇 Mic blocked" and the user
+          // gets a visible signal that voice is broken on their device.
           // eslint-disable-next-line no-console
-          console.error("[live-mode] AudioWorklet setup failed; fallback to silence", e);
+          console.error("[live-mode] AudioWorklet setup failed", e);
+          setMicPermissionDenied(true);
+          setLive(false);
+          return;
         }
 
         setMicPermissionDenied(false);