diff --git a/src/rechunk.jl b/src/rechunk.jl
index 01d05c0..a105e79 100644
--- a/src/rechunk.jl
+++ b/src/rechunk.jl
@@ -293,6 +293,17 @@ output = rechunk(input, 1024)
 ```
 """
 function rechunk(in::SignalChannel{T,N}, chunk_size::Integer, channel_size=16) where {T<:Number,N}
+    # No-op passthrough: when input and output chunk sizes already match, return
+    # the input channel directly. This avoids creating an intermediate channel
+    # and task, which fixes a race condition: with zero-copy passthrough the same
+    # buffer references flow through both the input and output channels, extending
+    # the pipeline depth beyond the upstream producer's buffer pool size. Under
+    # slow-consumer conditions (e.g. JIT compilation on first run), the producer
+    # can wrap around and overwrite buffers still queued in the downstream channel.
+    if in.num_samples == chunk_size
+        return in
+    end
+
     out = SignalChannel{T,N}(chunk_size, channel_size)
 
     # Estimate max outputs per input: input can complete partial + produce full chunks
diff --git a/test/rechunk.jl b/test/rechunk.jl
index 8bd9da4..8bda023 100644
--- a/test/rechunk.jl
+++ b/test/rechunk.jl
@@ -279,6 +279,17 @@ using FixedSizeArrays: FixedSizeMatrixDefault
         @test results[3] === inputs[3]
     end
 
+    @testset "rechunk channel - same size passthrough" begin
+        # When input and output chunk sizes match, rechunk should return
+        # the input channel directly (no intermediate channel or task).
+        # This prevents buffer-aliasing race conditions in pipelines where
+        # the upstream producer uses a fixed-size buffer pool.
+        input_chan = SignalChannel{ComplexF32,2}(1024)
+        output_chan = rechunk(input_chan, 1024)
+
+        @test output_chan === input_chan
+    end
+
     @testset "rechunk channel - upsampling" begin
         # Convert 100-sample chunks to 250-sample chunks
         input_chan = SignalChannel{ComplexF32,2}(100)