diff --git a/src/engine/helper/HL_PoolAllocator.h b/src/engine/helper/HL_PoolAllocator.h
index 0e151723..f2ba70d8 100644
--- a/src/engine/helper/HL_PoolAllocator.h
+++ b/src/engine/helper/HL_PoolAllocator.h
@@ -11,6 +11,7 @@
 /// transparent passthrough to the global allocator (i.e. allocate_shared behaves like make_shared)
 /// for A/B benchmarking without touching the call sites.
 
+#include <atomic>
 #include <cstddef>
 #include <cstdint>
 #include <mutex>
@@ -45,6 +46,7 @@ namespace hl
 			{
 				uint8_t* raw = static_cast<uint8_t*>(::operator new(size + HeaderSize));
 				tagOf(raw) = TagGlobal;
+				s_activeGlobal.fetch_add(1, std::memory_order_relaxed);
 				return raw + HeaderSize;
 			}
 
@@ -54,6 +56,7 @@ namespace hl
 
 			uint8_t* block = static_cast<uint8_t*>(head);
 			head = nextOf(block);
+			s_activeBlocks.fetch_add(1, std::memory_order_relaxed);
 			return block + HeaderSize;
 		}
 
@@ -68,14 +71,49 @@ namespace hl
 				void*& head = freeListHead();
 				nextOf(block) = head;
 				head = block;
+				s_activeBlocks.fetch_sub(1, std::memory_order_relaxed);
 			}
 			else
 			{
 				::operator delete(static_cast<void*>(block));
+				s_activeGlobal.fetch_sub(1, std::memory_order_relaxed);
 			}
 		}
 
+		/// @name Diagnostics (relaxed gauges — not synchronization)
+		/// @{
+		/// Blocks currently handed out from the pool = pooled allocate() minus
+		/// pooled deallocate(). Unlike capacityBlocks() (the chunk high-water,
+		/// which only ever grows) this DROPS when objects are freed, so a rising
+		/// activeBlocks() is a genuine leak of referenced objects rather than a
+		/// traffic peak that merely carved extra chunks. Covers every pooled
+		/// shared_ptr node (allocate_shared<RTPPacket> + jitter-buffer packets).
+		static std::int64_t activeBlocks() noexcept
+		{
+			return s_activeBlocks.load(std::memory_order_relaxed);
+		}
+		/// Live oversized allocations that overflowed to ::operator new.
+		static std::int64_t activeGlobalAllocations() noexcept
+		{
+			return s_activeGlobal.load(std::memory_order_relaxed);
+		}
+		/// Total blocks ever carved (chunks x BlocksPerChunk) = high-water capacity.
+		static std::int64_t capacityBlocks() noexcept
+		{
+			return s_chunks.load(std::memory_order_relaxed) * static_cast<std::int64_t>(BlocksPerChunk);
+		}
+		/// Approximate live pooled payload bytes (excludes per-block header).
+		static std::int64_t activeBytes() noexcept
+		{
+			return activeBlocks() * static_cast<std::int64_t>(PayloadSize);
+		}
+		/// @}
+
 	private:
+		inline static std::atomic<std::int64_t> s_activeBlocks{0};
+		inline static std::atomic<std::int64_t> s_activeGlobal{0};
+		inline static std::atomic<std::int64_t> s_chunks{0};
+
 		static constexpr std::size_t HeaderSize =
 		    alignof(std::max_align_t) >= sizeof(uint64_t) ? alignof(std::max_align_t) : sizeof(uint64_t);
 		static constexpr std::size_t BlockSize = HeaderSize + PayloadSize;
@@ -114,6 +152,7 @@ namespace hl
 			{
 				const std::size_t chunkBytes = BlockSize * BlocksPerChunk;
 				uint8_t* chunk = static_cast<uint8_t*>(::operator new(chunkBytes));
+				s_chunks.fetch_add(1, std::memory_order_relaxed);
 
 				{
 					std::lock_guard<std::mutex> lock(m_Mutex);
@@ -144,6 +183,21 @@ namespace hl
 	};
 #endif  // HL_RTP_POOL
 
+	/// Mode-independent accessors for the pool diagnostics, so callers need not know
+	/// whether HL_RTP_POOL is compiled in. When pooling is disabled they return -1
+	/// ("not applicable") since allocate_shared then behaves like make_shared.
+#if HL_RTP_POOL
+	inline std::int64_t poolActiveBlocks()   noexcept { return FixedBlockPool::activeBlocks(); }
+	inline std::int64_t poolCapacityBlocks() noexcept { return FixedBlockPool::capacityBlocks(); }
+	inline std::int64_t poolActiveGlobal()   noexcept { return FixedBlockPool::activeGlobalAllocations(); }
+	inline std::int64_t poolActiveBytes()    noexcept { return FixedBlockPool::activeBytes(); }
+#else
+	inline std::int64_t poolActiveBlocks()   noexcept { return -1; }
+	inline std::int64_t poolCapacityBlocks() noexcept { return -1; }
+	inline std::int64_t poolActiveGlobal()   noexcept { return -1; }
+	inline std::int64_t poolActiveBytes()    noexcept { return -1; }
+#endif
+
 	/// @class PoolAllocator
 	/// A stateless, std-conforming Allocator suitable for std::allocate_shared. When HL_RTP_POOL is
 	/// enabled it serves single-node allocations from FixedBlockPool; otherwise (and for any request