diff --git a/src/core/benchmark/fast_cal.cc b/src/core/benchmark/fast_cal.cc
index 52109ce..bdaf120 100644
--- a/src/core/benchmark/fast_cal.cc
+++ b/src/core/benchmark/fast_cal.cc
@@ -11,9 +11,9 @@ static void FastCalBenchmark(benchmark::State &state) {
     auto code = CommonCode::unsafe_create(0x1A9BF0C00).to_raw_code();
 
     for (auto _ : state) {
-        // auto fc = FastCal(code);
-        // benchmark::DoNotOptimize(fc.solve());
-        benchmark::DoNotOptimize(FastCal_demo(code));
+        auto fc = FastCal(code);
+        benchmark::DoNotOptimize(fc.demo());
+//        benchmark::DoNotOptimize(FastCal_demo(code));
     }
 
 }
diff --git a/src/core/fast_cal/fast_cal.h b/src/core/fast_cal/fast_cal.h
index 9111d0d..59aee58 100644
--- a/src/core/fast_cal/fast_cal.h
+++ b/src/core/fast_cal/fast_cal.h
@@ -44,6 +44,8 @@ public:
     RawCode target(const match_t &match);
     std::vector<RawCode> target_multi(const match_t &match);
 
+    RawCode demo();
+
     /// static BFS search functions
     static std::vector<RawCode> resolve(const RawCode &start);
     static std::vector<std::vector<RawCode>> to_furthest(const RawCode &start);
diff --git a/src/core/fast_cal/internal/fast_cal.cc b/src/core/fast_cal/internal/fast_cal.cc
index 85479e2..cdb9791 100644
--- a/src/core/fast_cal/internal/fast_cal.cc
+++ b/src/core/fast_cal/internal/fast_cal.cc
@@ -1,9 +1,13 @@
 #include <algorithm>
 
+#include <iostream>
+
 #include "utils/common.h"
 #include "fast_cal/fast_cal.h"
 #include "raw_code/raw_code.h"
 
+#include "group/group.h"
+
 FastCal::FastCal(const RawCode &code) {
     this->root = (uint64_t)code;
 }
@@ -21,6 +25,51 @@ RawCode FastCal::solve() {
     return FastCal::target(resolved);
 }
 
+RawCode FastCal::demo() {
+
+    // max_group_size = 25955
+//    auto reserve = klotski::cases::GroupUnion::from_raw_code(RawCode::unsafe_create(root)).max_group_size();
+
+    cases.reserve(25955); // FAST !!! (about 5ms)
+//    cases.reserve(32768); // SLOW !!! (about 50ms)
+
+    // TODO: using prime number!
+
+    std::queue<fast_cal_t*>{}.swap(cache);
+
+    cache.emplace(&cases.emplace(root, fast_cal_t {
+        .code = root,
+        .mask = 0,
+        .last = nullptr, // without parent node
+    }).first->second);
+
+    auto core = MaskMover(
+        [this](uint64_t code, uint64_t mask) { // lambda as function pointer
+
+            auto current = cases.find(code);
+            if (current != cases.end()) { // find existed case
+                current->second.mask |= mask; // update mask info
+                return;
+            }
+            cache.emplace(&cases.emplace(code, fast_cal_t { // record new case
+                .code = code,
+                .mask = mask,
+                .last = cache.front(), // link parent case
+            }).first->second);
+
+        }
+    );
+
+    while (!cache.empty()) {
+        if (((cache.front()->code >> (3 * 0xD)) & 0b111) == BLOCK_2x2) {
+            return RawCode::unsafe_create(cache.front()->code); // match target
+        }
+        core.next_cases(cache.front()->code, cache.front()->mask);
+        cache.pop();
+    }
+    return FC_NOT_FOUND; // target not found
+}
+
 std::vector<RawCode> FastCal::solve_multi() {
     return FastCal::target_multi(resolved);
 }