diff --git a/include/mimalloc/internal.h b/include/mimalloc/internal.h
index b846a233..88404f98 100644
--- a/include/mimalloc/internal.h
+++ b/include/mimalloc/internal.h
@@ -320,7 +320,7 @@ void __mi_stat_counter_increase_mt(mi_stat_counter_t* stat, size_t amount);
 #define mi_heap_stat_counter_increase(heap,stat,amount)         mi_tld_stat_counter_increase(heap->tld, stat, amount)
 #define mi_heap_stat_increase(heap,stat,amount)                 mi_tld_stat_increase( heap->tld, stat, amount)
 #define mi_heap_stat_decrease(heap,stat,amount)                 mi_tld_stat_decrease( heap->tld, stat, amount)
-
+#define mi_heap_stat_adjust_decrease(heap,stat,amount)          mi_tld_stat_adjust_decrease( heap->tld, stat, amount)
 
 /* -----------------------------------------------------------
   Options (exposed for the debugger)
diff --git a/src/alloc-aligned.c b/src/alloc-aligned.c
index 937cbdb8..2ed92146 100644
--- a/src/alloc-aligned.c
+++ b/src/alloc-aligned.c
@@ -193,10 +193,11 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t
       const bool is_aligned = (((uintptr_t)page->free + offset) & align_mask)==0;
       if mi_likely(is_aligned)
       {
+        void* p = (zero ? _mi_page_malloc_zeroed(heap,page,padsize) : _mi_page_malloc(heap,page,padsize)); // call specific page malloc for better codegen
         #if MI_STAT>1
+        mi_heap_stat_adjust_decrease(heap, malloc_requested, padsize);
         mi_heap_stat_increase(heap, malloc_requested, size);
         #endif
-        void* p = (zero ? _mi_page_malloc_zeroed(heap,page,padsize) : _mi_page_malloc(heap,page,padsize)); // call specific page malloc for better codegen
         mi_assert_internal(p != NULL);
         mi_assert_internal(((uintptr_t)p + offset) % alignment == 0);
         mi_track_malloc(p,size,zero);
diff --git a/src/alloc.c b/src/alloc.c
index 9a77b3dc..3e912726 100644
--- a/src/alloc.c
+++ b/src/alloc.c
@@ -92,6 +92,7 @@ extern inline void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_
     #if (MI_STAT>1)
     const size_t bin = _mi_bin(bsize);
     mi_heap_stat_increase(heap, malloc_bins[bin], 1);
+    mi_heap_stat_increase(heap, malloc_requested, size - MI_PADDING_SIZE);
     #endif
   }
   #endif
@@ -150,12 +151,6 @@ static inline mi_decl_restrict void* mi_heap_malloc_small_zero(mi_heap_t* heap,
   void* const p = _mi_page_malloc_zero(heap, page, size + MI_PADDING_SIZE, zero);
   mi_track_malloc(p,size,zero);
 
-  #if MI_STAT>1
-  if (p != NULL) {
-    if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
-    mi_heap_stat_increase(heap, malloc_requested, mi_usable_size(p));
-  }
-  #endif
   #if MI_DEBUG>3
   if (p != NULL && zero) {
     mi_assert_expensive(mi_mem_is_zero(p, size));
@@ -192,12 +187,6 @@ extern inline void* _mi_heap_malloc_zero_ex(mi_heap_t* heap, size_t size, bool z
     void* const p = _mi_malloc_generic(heap, size + MI_PADDING_SIZE, zero, huge_alignment);  // note: size can overflow but it is detected in malloc_generic
     mi_track_malloc(p,size,zero);
 
-    #if MI_STAT>1
-    if (p != NULL) {
-      if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
-      mi_heap_stat_increase(heap, malloc_requested, mi_usable_size(p));
-    }
-    #endif
     #if MI_DEBUG>3
     if (p != NULL && zero) {
       mi_assert_expensive(mi_mem_is_zero(p, size));
@@ -672,7 +661,8 @@ mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, boo
   if (p != NULL) {
     if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
     #if MI_STAT>1
-    mi_heap_stat_increase(heap, malloc_requested, mi_usable_size(p));
+    mi_heap_stat_adjust_decrease(heap, malloc_requested, req_size);
+    mi_heap_stat_increase(heap, malloc_requested, size);
     #endif
     mi_heap_stat_counter_increase(heap, malloc_guarded_count, 1);
   }
diff --git a/src/free.c b/src/free.c
index d63a430e..00ecf34d 100644
--- a/src/free.c
+++ b/src/free.c
@@ -532,16 +532,14 @@ static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) {
 
 // only maintain stats for smaller objects if requested
 #if (MI_STAT>0)
-void mi_stat_free(const mi_page_t* page, const mi_block_t* block) {
-  #if (MI_STAT < 2)
+static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) {
   MI_UNUSED(block);
-  #endif
   mi_heap_t* const heap = mi_heap_get_default();
   const size_t bsize = mi_page_usable_block_size(page);
-  #if (MI_STAT>1)
-  const size_t usize = mi_page_usable_size_of(page, block);
-  mi_heap_stat_decrease(heap, malloc_requested, usize);
-  #endif
+  // #if (MI_STAT>1)
+  // const size_t usize = mi_page_usable_size_of(page, block);
+  // mi_heap_stat_decrease(heap, malloc_requested, usize);
+  // #endif
   if (bsize <= MI_LARGE_MAX_OBJ_SIZE) {
     mi_heap_stat_decrease(heap, malloc_normal, bsize);
     #if (MI_STAT > 1)
diff --git a/src/heap.c b/src/heap.c
index 8655bd27..971aad68 100644
--- a/src/heap.c
+++ b/src/heap.c
@@ -342,7 +342,7 @@ static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_
   if (bsize > MI_LARGE_MAX_OBJ_SIZE) {
     mi_heap_stat_decrease(heap, malloc_huge, bsize);
   }
-  #if (MI_STAT)
+  #if (MI_STAT>0)
   _mi_page_free_collect(page, false);  // update used count
   const size_t inuse = page->used;
   if (bsize <= MI_LARGE_MAX_OBJ_SIZE) {
@@ -351,7 +351,7 @@ static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_
     mi_heap_stat_decrease(heap, malloc_bins[_mi_bin(bsize)], inuse);
     #endif
   }
-  mi_heap_stat_decrease(heap, malloc_requested, bsize * inuse);  // todo: off for aligned blocks...
+  // mi_heap_stat_decrease(heap, malloc_requested, bsize * inuse);  // todo: off for aligned blocks...
   #endif
 
   /// pretend it is all free now
diff --git a/src/stats.c b/src/stats.c
index 7a761d82..95eefac8 100644
--- a/src/stats.c
+++ b/src/stats.c
@@ -227,6 +227,13 @@ static void mi_stat_peak_print(const mi_stat_count_t* stat, const char* msg, int
   _mi_fprintf(out, arg, "\n");
 }
 
+static void mi_stat_total_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) {
+  _mi_fprintf(out, arg, "%10s:", msg);
+  _mi_fprintf(out, arg, "%12s", " ");  // no peak
+  mi_print_amount(stat->total, unit, out, arg);
+  _mi_fprintf(out, arg, "\n");
+}
+
 static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg ) {
   _mi_fprintf(out, arg, "%10s:", msg);
   mi_print_amount(stat->total, -1, out, arg);
@@ -243,7 +250,7 @@ static void mi_stat_counter_print_avg(const mi_stat_counter_t* stat, const char*
 
 
 static void mi_print_header(mi_output_fun* out, void* arg ) {
-  _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s\n", "heap stats", "peak   ", "total   ", "current   ", "unit   ", "total#   ");
+  _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s\n", "heap stats", "peak   ", "total   ", "current   ", "block   ", "total#   ");
 }
 
 #if MI_STAT>1
@@ -312,10 +319,10 @@ static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0)
   // and print using that
   mi_print_header(out,arg);
   #if MI_STAT>1
-  mi_stats_print_bins(stats->malloc_bins, MI_BIN_HUGE, "normal",out,arg);
+  mi_stats_print_bins(stats->malloc_bins, MI_BIN_HUGE, "bin",out,arg);
   #endif
   #if MI_STAT
-  mi_stat_print(&stats->malloc_normal, "normal", (stats->malloc_normal_count.total == 0 ? 1 : -1), out, arg);
+  mi_stat_print(&stats->malloc_normal, "binned", (stats->malloc_normal_count.total == 0 ? 1 : -1), out, arg);
   mi_stat_print(&stats->malloc_huge, "huge", (stats->malloc_huge_count.total == 0 ? 1 : -1), out, arg);
   mi_stat_count_t total = { 0,0,0 };
   mi_stat_count_add_mt(&total, &stats->malloc_normal);
@@ -323,7 +330,7 @@ static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0)
   mi_stat_print_ex(&total, "total", 1, out, arg, "");
   #endif
   #if MI_STAT>1
-  mi_stat_print_ex(&stats->malloc_requested, "malloc req", 1, out, arg, "");
+  mi_stat_total_print(&stats->malloc_requested, "malloc req", 1, out, arg);
   _mi_fprintf(out, arg, "\n");
   #endif
   mi_stat_print_ex(&stats->reserved, "reserved", 1, out, arg, "");