diff --git a/packages/telegram-remote/lib/whisper.js b/packages/telegram-remote/lib/whisper.js
index 0dcce6c..91c3833 100644
--- a/packages/telegram-remote/lib/whisper.js
+++ b/packages/telegram-remote/lib/whisper.js
@@ -6,8 +6,8 @@
  *   POST {base}/api/transcriptions   multipart fields: file, language,
  *                                    modelSize, device (cuda|cpu), sourceUrl
  *   GET  {base}/api/transcriptions/{id}
- * Job status -1 = queued/running; statuses 0/1/2 = terminal; a successful
- * terminal job has a non-empty result.text.
+ * Job status 0 = created/queued, 1 = processing, 2/3 = finished; a finished
+ * job has a non-empty result.text on success.
  */
 
 const POLL_INTERVAL_MS = 1500;
@@ -78,8 +78,12 @@ export async function transcribeAudio({ baseUrl, model, device, language, timeou
     }
     const text = job?.result?.text;
     if (typeof text === "string" && text.trim().length > 0) return text;
-    if (job?.status !== -1) {
-      throw new Error("whisper: job finished without text (status " + String(job?.status) + ")");
+    // Whishper lifecycle (verified on this server): 0 = created/queued,
+    // 1 = processing (can last tens of seconds), 2/3 = finished.
+    const status = job?.status;
+    const running = status === -1 || status === 0 || status === 1;
+    if (!running) {
+      throw new Error("whisper: job finished without text (status " + String(status) + ")");
     }
   }
 }