diff --git a/src/telegram/bot-message-dispatch.test.ts b/src/telegram/bot-message-dispatch.test.ts
index 7caa7cc3af7..2f6b2b00e84 100644
--- a/src/telegram/bot-message-dispatch.test.ts
+++ b/src/telegram/bot-message-dispatch.test.ts
@@ -1903,4 +1903,60 @@ describe("dispatchTelegramMessage draft streaming", () => {
     expect(draftA.clear).toHaveBeenCalledTimes(1);
     expect(draftB.clear).toHaveBeenCalledTimes(1);
   });
+
+  it("swallows post-connect network timeout on preview edit to prevent duplicate messages", async () => {
+    const draftStream = createDraftStream(999);
+    createTelegramDraftStream.mockReturnValue(draftStream);
+    dispatchReplyWithBufferedBlockDispatcher.mockImplementation(
+      async ({ dispatcherOptions, replyOptions }) => {
+        await replyOptions?.onPartialReply?.({ text: "Streaming..." });
+        await dispatcherOptions.deliver({ text: "Final answer" }, { kind: "final" });
+        return { queuedFinal: true };
+      },
+    );
+    deliverReplies.mockResolvedValue({ delivered: true });
+    // Simulate a post-connect timeout: editMessageTelegram throws a network
+    // error even though Telegram's server already processed the edit.
+    editMessageTelegram.mockRejectedValue(new Error("timeout: request timed out after 30000ms"));
+
+    await dispatchWithContext({ context: createContext() });
+
+    // The edit was attempted (and may have succeeded server-side).
+    expect(editMessageTelegram).toHaveBeenCalledTimes(1);
+    // The fix: no fallback sendPayload via deliverReplies for the final text,
+    // because the editPreview callback swallowed the network timeout.
+    // deliverReplies should NOT have been called with the "Final answer" text
+    // (it would only be called if the fallback chain fired).
+    const deliverCalls = deliverReplies.mock.calls;
+    const finalTextSentViaDeliverReplies = deliverCalls.some((call: unknown[]) =>
+      (call[0] as { replies?: Array<{ text?: string }> })?.replies?.some(
+        (r: { text?: string }) => r.text === "Final answer",
+      ),
+    );
+    expect(finalTextSentViaDeliverReplies).toBe(false);
+  });
+
+  it("re-throws pre-connect errors on preview edit so fallback can send", async () => {
+    const draftStream = createDraftStream(999);
+    createTelegramDraftStream.mockReturnValue(draftStream);
+    dispatchReplyWithBufferedBlockDispatcher.mockImplementation(
+      async ({ dispatcherOptions, replyOptions }) => {
+        await replyOptions?.onPartialReply?.({ text: "Streaming..." });
+        await dispatcherOptions.deliver({ text: "Final answer" }, { kind: "final" });
+        return { queuedFinal: true };
+      },
+    );
+    deliverReplies.mockResolvedValue({ delivered: true });
+    // Simulate a pre-connect error: the edit never reached Telegram.
+    const preConnectErr = new Error("connect ECONNREFUSED 149.154.167.220:443");
+    (preConnectErr as NodeJS.ErrnoException).code = "ECONNREFUSED";
+    editMessageTelegram.mockRejectedValue(preConnectErr);
+
+    await dispatchWithContext({ context: createContext() });
+
+    expect(editMessageTelegram).toHaveBeenCalledTimes(1);
+    // Pre-connect error is re-thrown, so the fallback chain fires and
+    // deliverReplies sends the final text as a new message.
+    expect(deliverReplies).toHaveBeenCalled();
+  });
 });
diff --git a/src/telegram/bot-message-dispatch.ts b/src/telegram/bot-message-dispatch.ts
index fee56211ae5..58dab2f6568 100644
--- a/src/telegram/bot-message-dispatch.ts
+++ b/src/telegram/bot-message-dispatch.ts
@@ -39,6 +39,7 @@ import {
   type DraftLaneState,
   type LaneName,
 } from "./lane-delivery.js";
+import { isSafeToRetrySendError } from "./network-errors.js";
 import {
   createTelegramReasoningStepState,
   splitTelegramReasoningText,
@@ -478,13 +479,35 @@ export const dispatchTelegramMessage = async ({
       await lane.stream?.stop();
     },
     editPreview: async ({ messageId, text, previewButtons }) => {
-      await editMessageTelegram(chatId, messageId, text, {
-        api: bot.api,
-        cfg,
-        accountId: route.accountId,
-        linkPreview: telegramCfg.linkPreview,
-        buttons: previewButtons,
-      });
+      try {
+        await editMessageTelegram(chatId, messageId, text, {
+          api: bot.api,
+          cfg,
+          accountId: route.accountId,
+          linkPreview: telegramCfg.linkPreview,
+          buttons: previewButtons,
+        });
+      } catch (err) {
+        // Post-connect network errors (timeout, connection reset) mean the edit
+        // may have already landed on Telegram's server. Swallow these to prevent
+        // the fallback chain from sending a duplicate message via sendPayload.
+        // Only re-throw pre-connect errors (DNS, connection refused — edit
+        // definitely never reached Telegram) and API errors (400/500 — Telegram
+        // explicitly rejected the edit).
+        if (isSafeToRetrySendError(err)) {
+          throw err;
+        }
+        const isNetworkError =
+          err instanceof Error &&
+          /timeout|timed out|reset|closed|network|fetch failed|socket/i.test(err.message);
+        if (isNetworkError) {
+          logVerbose(
+            `telegram: preview edit may have succeeded despite network error; treating as delivered to avoid duplicate (${String(err)})`,
+          );
+          return;
+        }
+        throw err;
+      }
     },
     deletePreviewMessage: async (messageId) => {
       await bot.api.deleteMessage(chatId, messageId);
diff --git a/src/telegram/lane-delivery.test.ts b/src/telegram/lane-delivery.test.ts
index 1cd1d36cf4c..fb69133e69d 100644
--- a/src/telegram/lane-delivery.test.ts
+++ b/src/telegram/lane-delivery.test.ts
@@ -362,6 +362,54 @@ describe("createLaneTextDeliverer", () => {
     expect(harness.markDelivered).not.toHaveBeenCalled();
   });
 
+  // ── Duplicate message regression tests ──────────────────────────────────
+  // These document the fallback behavior at the lane-delivery layer.
+  // The real fix lives in bot-message-dispatch.ts: the editPreview callback
+  // swallows post-connect network errors (timeout, reset) so they never
+  // reach this layer. These tests verify that if a non-network API error
+  // (e.g. 500) still reaches tryEditPreviewMessage, the fallback to
+  // sendPayload still works correctly.
+
+  it("falls back to sendPayload on genuine API errors (not network timeout)", async () => {
+    const harness = createHarness({ answerMessageId: 999 });
+    // A real API error (not a network timeout) — edit definitely didn't land
+    harness.editPreview.mockRejectedValue(new Error("500: Internal Server Error"));
+
+    const result = await harness.deliverLaneText({
+      laneName: "answer",
+      text: "Hello final",
+      payload: { text: "Hello final" },
+      infoKind: "final",
+    });
+
+    expect(harness.editPreview).toHaveBeenCalledTimes(1);
+    expect(harness.sendPayload).toHaveBeenCalledTimes(1);
+    expect(result).toBe("sent");
+  });
+
+  it("falls back on archived preview API error and cleans up old preview", async () => {
+    const harness = createHarness();
+    harness.archivedAnswerPreviews.push({
+      messageId: 5555,
+      textSnapshot: "Partial streaming...",
+      deleteIfUnused: true,
+    });
+    // Genuine API error — edit definitely didn't land
+    harness.editPreview.mockRejectedValue(new Error("500: Internal Server Error"));
+
+    const result = await harness.deliverLaneText({
+      laneName: "answer",
+      text: "Complete final answer",
+      payload: { text: "Complete final answer" },
+      infoKind: "final",
+    });
+
+    expect(harness.editPreview).toHaveBeenCalledTimes(1);
+    expect(harness.sendPayload).toHaveBeenCalledTimes(1);
+    expect(harness.deletePreviewMessage).toHaveBeenCalledWith(5555);
+    expect(result).toBe("sent");
+  });
+
   it("deletes consumed boundary previews after fallback final send", async () => {
     const harness = createHarness();
     harness.archivedAnswerPreviews.push({