diff --git a/core/box/warm_pool_prefill_box.h b/core/box/warm_pool_prefill_box.h
index 89769c65..dc3f764a 100644
--- a/core/box/warm_pool_prefill_box.h
+++ b/core/box/warm_pool_prefill_box.h
@@ -24,7 +24,7 @@ extern SuperSlab* superslab_refill(int class_idx);
 // Prefill budget: How many additional SuperSlabs to load when pool is empty
 // - If pool is empty, load PREFILL_BUDGET extra slabs to build working set
 // - This avoids repeated registry scans on rapid cache misses
-// - Set to 2 to balance between prefill lock overhead and pool depletion
+// - Phase 2: Keep at 2 (increasing to 4 caused contention regression -1.5%)
 #define WARM_POOL_PREFILL_BUDGET 2
 
 // ============================================================================
diff --git a/core/front/tiny_warm_pool.h b/core/front/tiny_warm_pool.h
index 2df32d86..03b3d75b 100644
--- a/core/front/tiny_warm_pool.h
+++ b/core/front/tiny_warm_pool.h
@@ -39,10 +39,10 @@
 
 // Maximum warm SuperSlabs per thread per class (tunable)
 // Trade-off: Working set size vs warm pool effectiveness
-//   - 4: Original (90% hit rate expected, but broken implementation)
-//   - 16: Increased to compensate for suboptimal push logic
+//   - 4: Original (90% hit rate expected, but broken implementation - hardcoded prefill threshold)
+//   - 12: Optimized capacity with matching prefill threshold (Phase 1)
 //   - Higher values: More memory but better locality
-#define TINY_WARM_POOL_MAX_PER_CLASS 16
+#define TINY_WARM_POOL_MAX_PER_CLASS 12
 
 typedef struct {
     SuperSlab* slabs[TINY_WARM_POOL_MAX_PER_CLASS];
@@ -107,16 +107,16 @@ static inline int tiny_warm_pool_count(int class_idx) {
 // ============================================================================
 
 // Get warm pool capacity from environment (configurable at runtime)
-// ENV: HAKMEM_WARM_POOL_SIZE=N (default: 4)
+// ENV: HAKMEM_WARM_POOL_SIZE=N (default: 12)
 static inline int warm_pool_max_per_class(void) {
     static int g_max = -1;
     if (__builtin_expect(g_max == -1, 0)) {
         const char* env = getenv("HAKMEM_WARM_POOL_SIZE");
         if (env && *env) {
             int v = atoi(env);
-            // Clamp to valid range [1, 16]
+            // Clamp to valid range [1, 12]
             if (v < 1) v = 1;
-            if (v > 16) v = 16;
+            if (v > 12) v = 12;
             g_max = v;
         } else {
             g_max = TINY_WARM_POOL_MAX_PER_CLASS;
diff --git a/core/hakmem_shared_pool_acquire.c b/core/hakmem_shared_pool_acquire.c
index 3b44f948..c74fc91e 100644
--- a/core/hakmem_shared_pool_acquire.c
+++ b/core/hakmem_shared_pool_acquire.c
@@ -83,7 +83,8 @@ sp_acquire_from_empty_scan(int class_idx, SuperSlab** ss_out, int* slab_idx_out,
 
         // WARM POOL PREFILL: Add HOT SuperSlabs to warm pool (if not already primary result)
         // This is low-cost during registry scan and avoids future expensive scans
-        if (ss != primary_result && tiny_warm_pool_count(class_idx) < 4) {
+        // Phase 1: Increase threshold from 4 to 12 to match TINY_WARM_POOL_MAX_PER_CLASS
+        if (ss != primary_result && tiny_warm_pool_count(class_idx) < 12) {
             tiny_warm_pool_push(class_idx, ss);
             // Track prefilled SuperSlabs for metrics
             g_warm_pool_stats[class_idx].prefilled++;