diff --git a/dsv4/kernels/attention/fmha.py b/dsv4/kernels/attention/fmha.py
index a2d0a43f..699f6d47 100644
--- a/dsv4/kernels/attention/fmha.py
+++ b/dsv4/kernels/attention/fmha.py
@@ -24,7 +24,7 @@ class FmhaKernel:
         self.use_smem_p = use_smem_p if use_smem_p is not None else (head_dim > 64)
         self.debug_p_one = False  # DEBUG: write constant P=1.0 to verify mapping
         self.debug_swap_mn = False  # DEBUG: try swapping m and n0 in coordinate mapping
-        self.debug_permute = 0  # DEBUG: try different coordinate permutations
+        self.debug_permute = 1  # DEBUG: try different coordinate permutations (0=original,1=swap m↔n0)
         self.acc_dtype = Float32; self.qk_acc_dtype = Float32
         self.q_dtype = BFloat16; self.o_dtype = BFloat16; self.c_dtype = BFloat16
         self.use_2cta_instrs = False; self.epilog_sync_bar_id = 1