diff --git a/packages/engine/src/services/chunkEncoder.test.ts b/packages/engine/src/services/chunkEncoder.test.ts
index f0ce7adab..cc6325333 100644
--- a/packages/engine/src/services/chunkEncoder.test.ts
+++ b/packages/engine/src/services/chunkEncoder.test.ts
@@ -389,6 +389,7 @@ describe("muxVideoWithAudio audio codec handling", () => {
     );
 
     await flushMuxCodecResolution();
+    expect(calls[0]!.args).toContain("-shortest");
     expect(calls[0]!.args).toContain("-t");
     expect(calls[0]!.args).toContain("16.066667");
     emitClose(calls[0]!.proc, 0);
diff --git a/packages/engine/src/services/chunkEncoder.ts b/packages/engine/src/services/chunkEncoder.ts
index a2d301781..e373520a7 100644
--- a/packages/engine/src/services/chunkEncoder.ts
+++ b/packages/engine/src/services/chunkEncoder.ts
@@ -709,6 +709,13 @@ export async function muxVideoWithAudio(
     args.push("-r", fpsToFfmpegArg(fps));
   }
   if (config?.durationSeconds !== undefined) {
+    // Stream-copying a normalized AAC sidecar can preserve a longer packet
+    // timeline than its container/edit-list duration.  `-t` alone limits the
+    // output timestamp window but does not prevent the copied audio stream's
+    // tail from extending the MP4 timeline.  When an explicit video-derived
+    // duration is supplied, also stop at the shortest stream so the final
+    // mux boundary is deterministic.
+    args.push("-shortest");
     args.push("-t", String(config.durationSeconds));
   }
   args.push("-y", outputPath);