diff --git a/deploy/reactor.py b/deploy/reactor.py
index 556b2ad..588abb9 100644
--- a/deploy/reactor.py
+++ b/deploy/reactor.py
@@ -173,17 +173,40 @@ async def _claude_call(messages: list, system: str = "") -> str:
                 raise RuntimeError(f"Claude API error {resp.status}: {data.get('error',{}).get('message','')}")
             return data["content"][0]["text"]
 
+def _parse_groq_reset(val: str) -> float:
+    """Parse Groq's Go-style duration strings ('229ms', '2.5s', '2m52.8s') into seconds."""
+    import re as _re
+    if not val:
+        return 0.0
+    total = 0.0
+    for num, unit in _re.findall(r'([\d.]+)(ms|s|m|h)', val):
+        n = float(num)
+        total += n/1000 if unit == 'ms' else n*60 if unit == 'm' else n*3600 if unit == 'h' else n
+    return total
+
 async def _groq_call(messages: list, system: str = "") -> str:
+    # Added 2026-07-07: retry once on 429 (rate limit) using Groq's own reset-time
+    # header, capped short — this account's tier has a low 12k-tokens/min ceiling that
+    # gmail_triage alone can exhaust, so transient contention here is common and often
+    # clears in well under a second. Capped so a genuinely long reset just falls
+    # through to Ollama instead of blocking the whole compose flow.
     all_msgs = ([{"role": "system", "content": system}] if system else []) + messages
     payload = {"model": GROQ_MODEL, "messages": all_msgs, "max_tokens": 4096}
     headers = {"Authorization": f"Bearer {GROQ_API_KEY}", "Content-Type": "application/json"}
-    async with aiohttp.ClientSession() as session:
-        async with session.post("https://api.groq.com/openai/v1/chat/completions", json=payload,
-                                headers=headers, timeout=aiohttp.ClientTimeout(total=45)) as resp:
-            data = await resp.json()
-            if resp.status != 200:
-                raise RuntimeError(f"Groq error {resp.status}")
-            return data["choices"][0]["message"]["content"]
+    for attempt in range(2):
+        async with aiohttp.ClientSession() as session:
+            async with session.post("https://api.groq.com/openai/v1/chat/completions", json=payload,
+                                    headers=headers, timeout=aiohttp.ClientTimeout(total=45)) as resp:
+                if resp.status == 429 and attempt == 0:
+                    wait = min(_parse_groq_reset(resp.headers.get("x-ratelimit-reset-tokens", "")) or
+                               _parse_groq_reset(resp.headers.get("x-ratelimit-reset-requests", "")) or 2.0, 8.0)
+                    log.info(f"[LLM] Groq 429 — retrying in {wait:.1f}s")
+                    await asyncio.sleep(wait)
+                    continue
+                data = await resp.json()
+                if resp.status != 200:
+                    raise RuntimeError(f"Groq error {resp.status}")
+                return data["choices"][0]["message"]["content"]
 
 async def _ollama_call(messages: list, system: str = "") -> str:
     prompt = (system + "\n\n" if system else "") + "\n".join(f"{m['role'].upper()}: {m['content']}" for m in messages)
diff --git a/public_html/admin/index.php b/public_html/admin/index.php
index 332964f..21f0ab9 100644
--- a/public_html/admin/index.php
+++ b/public_html/admin/index.php
@@ -4458,6 +4458,7 @@ function outboxCompose() {
       <input id="oc-to"      class="inp" placeholder="To: email address" type="email">
       <input id="oc-subject" class="inp" placeholder="Subject">
       <textarea id="oc-body" class="inp" rows="5" style="resize:vertical" placeholder="Describe what to say — AI will draft the full message"></textarea>
+      <div id="oc-status" style="display:none;font-size:0.65rem;color:var(--text-dim);padding:8px;background:#060a0e;border:1px solid var(--border);border-radius:3px"></div>
     </div>
   `, async () => {
     const to      = document.getElementById('oc-to')?.value.trim();
@@ -4465,14 +4466,39 @@ function outboxCompose() {
     const body    = document.getElementById('oc-body')?.value.trim();
     const account = document.getElementById('oc-account')?.value;
     if (!to || !body) { toast('Please fill To and message description', 'err'); return; }
+    const statusEl = document.getElementById('oc-status');
     const d = await api('compose_email', {to, subject, body, account});
-    if (d.job_id) {
-      toast('Compose job dispatched — Job #' + d.job_id, 'ok');
-      closeModal();
-      setTimeout(loadOutbox, 5000);
-    } else {
-      toast('Failed: ' + (d.error||'Arc Reactor offline'), 'err');
+    if (!d.job_id) { toast('Failed: ' + (d.error||'Arc Reactor offline'), 'err'); return; }
+    // Fixed 2026-07-07: this used to toast "dispatched" and close immediately, with
+    // no way to ever find out if the job later failed — it just silently vanished.
+    // Now polls the actual job status until it finishes, so failures are visible.
+    if (statusEl) { statusEl.style.display = 'block'; statusEl.textContent = 'Drafting — Job #' + d.job_id + '...'; }
+    document.getElementById('modalSave').disabled = true;
+    const jobId = d.job_id;
+    const deadline = Date.now() + 120000;
+    while (Date.now() < deadline) {
+      await new Promise(r => setTimeout(r, 3000));
+      const j = await api('arc_job_get', {id: jobId}).catch(() => null);
+      if (!j || j.error) continue;
+      if (j.status === 'done') {
+        toast('Draft ready: ' + (j.result?.subject || '(no subject)'), 'ok');
+        closeModal();
+        loadOutbox();
+        return;
+      } else if (j.status === 'failed') {
+        if (statusEl) {
+          statusEl.style.color = 'var(--red)';
+          statusEl.textContent = '✗ Failed: ' + (j.error || 'unknown error');
+        }
+        document.getElementById('modalSave').disabled = false;
+        document.getElementById('modalSave').textContent = 'CLOSE';
+        document.getElementById('modalSave').onclick = closeModal;
+        return;
+      }
+      if (statusEl) statusEl.textContent = 'Drafting — Job #' + jobId + ' (' + j.status + ')...';
     }
+    if (statusEl) { statusEl.style.color = 'var(--yellow)'; statusEl.textContent = '⚠ Still running in background — check outbox shortly.'; }
+    document.getElementById('modalSave').disabled = false;
   }, 'DISPATCH');
 }