diff --git a/schwab_sort.h b/schwab_sort.h
index 36bd2d6..169cb90 100644
--- a/schwab_sort.h
+++ b/schwab_sort.h
@@ -3,7 +3,7 @@
 
 /* A fast quicksort-like new alg created in Csolnok, Hungary with:
  * - 4-way partitioning with 0..5 copies (not swaps) per elem per run
- * - ensured O(log2(n)) worst recursion depth
+ * - "Practically ensures" O(log2(n)) recursion depth
  *
  * LICENCE: CC-BY, 2025 May 08-09
  * Author: Richárd István Thier (also author of the Magyarsort)
@@ -164,23 +164,20 @@ static inline int schwab_partition(
 		 * This is likely faster than calling a memcpy if we code this for not just uint32s!
 		 */
 		if(curr < klo) {
-			arr[b3] = arr[b2];
-			arr[b2] = arr[b1];
-			arr[b1] = arr[b0];
-			arr[b0] = curr;
-			++b0; ++b1; ++b2;
+			arr[b3] = arr[b2++];
+			arr[b2 - 1] = arr[b1++];
+			arr[b1 - 1] = arr[b0++];
+			arr[b0 - 1] = curr;
 			continue;
 		}
 
 		if(curr < kmid) {
-			arr[b3] = arr[b2];
-			arr[b2] = arr[b1];
-			arr[b1] = curr;
-			++b1; ++b2;
+			arr[b3] = arr[b2++];
+			arr[b2 - 1] = arr[b1++];
+			arr[b1 - 1] = curr;
 		} else {
-			arr[b3] = arr[b2];
-			arr[b2] = curr;
-			++b2;
+			arr[b3] = arr[b2++];
+			arr[b2 - 1] = curr;
 		}
 	}