diff --git a/apps/desktop/src/app/session/hooks/use-message-stream/index.ts b/apps/desktop/src/app/session/hooks/use-message-stream/index.ts
index 979bbc1e2f..90a6fcad9a 100644
--- a/apps/desktop/src/app/session/hooks/use-message-stream/index.ts
+++ b/apps/desktop/src/app/session/hooks/use-message-stream/index.ts
@@ -632,7 +632,7 @@ export function useMessageStream({
               nextMessages = prev.map((message, messageIndex) =>
                 messageIndex === index ? completeMessage(message) : message
               )
-            } else if (interimBoundaryPending && (responsePreviewed || finalContinuesInterim)) {
+            } else if (existing.interim && (responsePreviewed || finalContinuesInterim)) {
               // Settle the interim in place instead of creating a duplicate —
               // the DB has one row, so the live UI must agree. Previously this
               // was gated on `responsePreviewed` alone, so a NON-previewed
@@ -642,6 +642,15 @@ export function useMessageStream({
               // for ordinary tool-call turns while `responsePreviewed` still
               // covers the verify-on-stop continuation-budget case even when the
               // final text was rewritten and no longer shares a prefix.
+              //
+              // We key on the message's OWN durable `interim` state, not the
+              // session's volatile `interimBoundaryPending` flag: the flag is
+              // reset to false by a subsequent `message.start` (a chained turn,
+              // follow-up, or mid-turn compaction). When that reset lands
+              // between this turn's `message.interim` and `message.complete`, the
+              // flag-based gate wrongly fell through to appending a duplicate
+              // bubble (#74560). The message row still says `interim`, so the
+              // continuation merges correctly regardless of the flag.
               nextMessages = prev.map((message, messageIndex) =>
                 messageIndex === index ? completeMessage(message) : message
               )
diff --git a/apps/desktop/src/app/session/hooks/use-message-stream/interim-sealing.test.tsx b/apps/desktop/src/app/session/hooks/use-message-stream/interim-sealing.test.tsx
index 7825904d3d..90ac198a98 100644
--- a/apps/desktop/src/app/session/hooks/use-message-stream/interim-sealing.test.tsx
+++ b/apps/desktop/src/app/session/hooks/use-message-stream/interim-sealing.test.tsx
@@ -217,6 +217,23 @@ describe('useMessageStream interim text sealing', () => {
     expect(texts[0]).toBe('partial answer continued')
   })
 
+  it('settles final onto interim even after message.start reset the boundary flag (#74560)', async () => {
+    await mountStream()
+    await start()
+    await delta('partial')
+    await interim('partial')
+    // A chained turn / follow-up re-emits message.start, which resets
+    // interimBoundaryPending to false BEFORE the same turn's message.complete.
+    // The continuation must still settle onto the interim — not append a
+    // duplicate bubble. Regression for #74560.
+    await start()
+    await complete('partial answer continued')
+
+    const texts = assistantMessages()
+    expect(texts.filter(t => t.includes('partial'))).toHaveLength(1)
+    expect(texts[0]).toBe('partial answer continued')
+  })
+
   it('appends a genuinely different final as its own bubble (two real assistant segments)', async () => {
     await mountStream()
     await start()