diff --git a/ide/vs2022/mimalloc.vcxproj b/ide/vs2022/mimalloc.vcxproj
index 07a854ab..894c5030 100644
--- a/ide/vs2022/mimalloc.vcxproj
+++ b/ide/vs2022/mimalloc.vcxproj
@@ -116,7 +116,7 @@
       <SDLCheck>true</SDLCheck>
       <ConformanceMode>Default</ConformanceMode>
       <AdditionalIncludeDirectories>../../include</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>MI_DEBUG=4;MI_SECURE=0;%(PreprocessorDefinitions);</PreprocessorDefinitions>
+      <PreprocessorDefinitions>MI_DEBUG=0;MI_SECURE=0;%(PreprocessorDefinitions);</PreprocessorDefinitions>
       <CompileAs>CompileAsCpp</CompileAs>
       <SupportJustMyCode>false</SupportJustMyCode>
       <LanguageStandard>stdcpp20</LanguageStandard>
diff --git a/src/segment.c b/src/segment.c
index 1b73b19a..1e23bb1a 100644
--- a/src/segment.c
+++ b/src/segment.c
@@ -316,7 +316,9 @@ static uint8_t* _mi_segment_page_start_from_slice(const mi_segment_t* segment, c
   ptrdiff_t idx = slice - segment->slices;
   size_t psize = (size_t)slice->slice_count * MI_SEGMENT_SLICE_SIZE;
   // make the start not OS page aligned for smaller blocks to avoid page/cache effects
-  size_t start_offset = (xblock_size >= MI_INTPTR_SIZE && xblock_size <= 1024 ? 3*MI_MAX_ALIGN_GUARANTEE : 0); 
+  // note: the offset must always be an xblock_size multiple since we assume small allocations
+  // are aligned (see `mi_heap_malloc_aligned`).
+  size_t start_offset = (xblock_size >= MI_INTPTR_SIZE && xblock_size <= 512 ? xblock_size : 0); 
   if (page_size != NULL) { *page_size = psize - start_offset; }
   return (uint8_t*)segment + ((idx*MI_SEGMENT_SLICE_SIZE) + start_offset);
 }
diff --git a/test/main-override-static.c b/test/main-override-static.c
index 534c8849..5e8b7333 100644
--- a/test/main-override-static.c
+++ b/test/main-override-static.c
@@ -20,6 +20,7 @@ static void negative_stat(void);
 static void alloc_huge(void);
 static void test_heap_walk(void);
 static void test_heap_arena(void);
+static void test_align(void);
 
 int main() {
   mi_version();
@@ -37,6 +38,7 @@ int main() {
   // alloc_huge();
   // test_heap_walk();
   // test_heap_arena();
+  test_align();
   
   void* p1 = malloc(78);
   void* p2 = malloc(24);
@@ -68,6 +70,13 @@ int main() {
   return 0;
 }
 
+static void test_align() {
+  void* p = mi_malloc_aligned(256, 256);
+  if (((uintptr_t)p % 256) != 0) {
+    fprintf(stderr, "%p is not 256 alignend!\n", p);
+  }
+}
+
 static void invalid_free() {
   free((void*)0xBADBEEF);
   realloc((void*)0xBADBEEF,10);
diff --git a/test/test-api.c b/test/test-api.c
index c78e1972..1967dad7 100644
--- a/test/test-api.c
+++ b/test/test-api.c
@@ -212,6 +212,24 @@ int main(void) {
     result = mi_heap_contains_block(heap, p);
     mi_heap_destroy(heap);
   }
+  CHECK_BODY("malloc-aligned12") {
+    bool ok = true;
+    const size_t align = 256;
+    for (int j = 1; j < 1000; j++) {
+      void* ps[1000];
+      for (int i = 0; i < 1000 && ok; i++) {
+        ps[i] = mi_malloc_aligned(j  // size
+                                  , align);
+        if (ps[i] == NULL || ((uintptr_t)(ps[i]) % align) != 0) {
+          ok = false;
+        }
+      }
+      for (int i = 0; i < 1000 && ok; i++) {
+        mi_free(ps[i]);
+      }
+    }
+    result = ok;
+  };
   CHECK_BODY("malloc-aligned-at1") {
     void* p = mi_malloc_aligned_at(48,32,0); result = (p != NULL && ((uintptr_t)(p) + 0) % 32 == 0); mi_free(p);
   };