diff --git a/.gitignore b/.gitignore
index a7de0c6..c7ea5a8 100644
--- a/.gitignore
+++ b/.gitignore
@@ -3,3 +3,5 @@ __pycache__
 /dist
 /obj
 .ninja_log
+.ninja_deps
+/boost_*
\ No newline at end of file
diff --git a/.idea/customTargets.xml b/.idea/customTargets.xml
index a6033b3..861fa49 100644
--- a/.idea/customTargets.xml
+++ b/.idea/customTargets.xml
@@ -2,7 +2,7 @@
 <project version="4">
   <component name="CLionExternalBuildManager">
     <target id="5c1cb7ec-1aa4-4f6f-97c1-504abfc76bd3" name="main" defaultType="MAKE">
-      <configuration id="ea6404e0-1feb-4070-b1c1-4295409dace7" name="main">
+      <configuration id="ea6404e0-1feb-4070-b1c1-4295409dace7" name="main" toolchainName="Default">
         <build type="MAKE">
           <make targetName="runtime_lib/main.cpp" />
         </build>
diff --git a/.idea/makefile.xml b/.idea/makefile.xml
index 6278399..bde9212 100644
--- a/.idea/makefile.xml
+++ b/.idea/makefile.xml
@@ -4,7 +4,7 @@
     <user-build-targets>
       <build-target name="runtime_lib/main.cpp">
         <build-configurations>
-          <build-configuration>
+          <build-configuration toolchainName="Default">
             <make-targets>
               <make-target>runtime_lib/main.cpp</make-target>
             </make-targets>
@@ -13,7 +13,7 @@
       </build-target>
       <build-target name="clean">
         <build-configurations>
-          <build-configuration>
+          <build-configuration toolchainName="Default">
             <make-targets>
               <make-target>clean</make-target>
             </make-targets>
diff --git a/.idea/misc.xml b/.idea/misc.xml
index 1caec42..d13647f 100644
--- a/.idea/misc.xml
+++ b/.idea/misc.xml
@@ -4,6 +4,9 @@
     <sourceRoots>
       <file path="$PROJECT_DIR$/runtime_lib" />
     </sourceRoots>
+    <libraryRoots>
+      <file path="$PROJECT_DIR$/boost_1_79_0" />
+    </libraryRoots>
     <excludeRoots>
       <file path="$PROJECT_DIR$/dist" />
       <file path="$PROJECT_DIR$/obj" />
diff --git a/build.ninja b/build.ninja
index a871e52..cddde28 100644
--- a/build.ninja
+++ b/build.ninja
@@ -1,11 +1,18 @@
-cflags = -Wall -Wextra -g
+cflags = -Wall -Wextra -g -Iboost_1_79_0
 print_flags = -fdiagnostics-color=always
+link_args = -c
 
 libraries = -lsodium
 
 rule cc
-    command = g++ $print_flags $cflags $libraries $in -o $out
+    deps = gcc
+    depfile = $out.d
+    command = g++ -MD -MF $out.d $print_flags $link_args $cflags $libraries $in -o $out
 
-build dist/main: cc runtime_lib/main.cpp
+build obj/call_frame.o: cc runtime_lib/call_frame.cpp
+build obj/global_thread_pool.o: cc runtime_lib/global_thread_pool.cpp
+
+build dist/main: cc runtime_lib/main.cpp obj/call_frame.o obj/global_thread_pool.o
+    link_args =
 
 
diff --git a/runtime_lib/call_frame.cpp b/runtime_lib/call_frame.cpp
new file mode 100644
index 0000000..79fb186
--- /dev/null
+++ b/runtime_lib/call_frame.cpp
@@ -0,0 +1,56 @@
+//
+// Created by anton on 6/24/22.
+//
+
+#include "call_frame.h"
+#include "helpers.h"
+
+
+namespace pmp::core {
+
+
+    // public
+
+    bool call_frame::evaluate() {
+        std::unique_lock<std::mutex> lock(evaluation_lock, std::try_to_lock);
+
+        // check if lock was acquired
+        if (!lock.owns_lock()) {
+            return false;
+        }
+        // we now hold the lock
+
+        // only run if we are in the initial state
+        if (state != initial)
+            return false;
+
+        // set housekeeping variables
+        state = running;
+
+        try {
+            result = target_function(arguments, scope);
+        } catch (std::exception& ex)
+        {
+            std::cerr << "Error evaluating " << id << ": " << ex.what() << std::endl;
+            m_has_error = true;
+        }
+        catch (...)
+        {
+            std::cerr << "Error evaluating " << id << std::endl;
+            m_has_error = true;
+        }
+
+        // children are empty, when the "end" of an execution chain has been reached
+        // this means that no further nodes were created during the execution
+        // we don't need to acquire any lock, since no children will ever be added
+        if (children.empty()) {
+            state = completed;
+            parent->mark_completion();
+        } else {
+            // we do have children, so we must wait until they are finished to move on.
+            state = waiting;
+        }
+
+        return true;
+    }
+}
diff --git a/runtime_lib/call_frame.h b/runtime_lib/call_frame.h
new file mode 100644
index 0000000..5333726
--- /dev/null
+++ b/runtime_lib/call_frame.h
@@ -0,0 +1,88 @@
+//
+// Created by anton on 6/24/22.
+//
+
+#ifndef PMP_CALL_FRAME_H
+#define PMP_CALL_FRAME_H
+
+#include <iostream>
+#include <cassert>
+#include <atomic>
+#include <utility>
+#include <vector>
+#include <mutex>
+#include <algorithm>
+#include <memory>
+#include <list>
+
+#include "helpers.h"
+
+
+namespace pmp::core {
+
+    typedef void *call_frame_target(void * args, void * scope);
+
+    typedef u64 frame_id_t;
+
+    enum call_frame_state {
+        initial = 0,
+        running = 1,
+        waiting = 2,
+        completed = 3
+    };
+
+    class call_frame {
+
+    public:
+        call_frame(void *argument, void *scope, call_frame_target *target, frame_id_t id, const std::weak_ptr<call_frame>&parent) :
+                state(initial), arguments(argument), scope(scope), result(nullptr),
+                target_function(target), id(id), children(), children_lock(), parent(parent),
+                evaluation_lock(), m_has_error(false) {
+
+        }
+
+        std::atomic<call_frame_state> state;
+        void *arguments;
+        void *scope;
+        void* result;
+        call_frame_target *target_function;
+        frame_id_t id;
+
+        bool evaluate();
+
+    private:
+
+        void mark_completion() {
+            // we can only be notified of child completion when we are waiting...
+            assert(state == waiting);
+
+            if (state == waiting && has_unfinished_children()) {
+                state = completed;
+                if (parent != nullptr) {
+                    parent->mark_completion();
+                }
+            }
+        }
+
+        bool has_unfinished_children() {
+            // make sure the list of children is not modified
+            WITH_LOCK(children_lock);
+            // return true as soon as any child is not complete
+            return std::any_of(children.begin(), children.end(), [](auto child) {
+                return child->state != completed;
+            });
+
+        }
+
+        std::list<std::shared_ptr<call_frame>> children;
+        std::mutex children_lock;
+        const std::shared_ptr<call_frame> parent;
+
+        // this lock is taken when the function is executed to prevent double-execution
+        std::mutex evaluation_lock;
+        std::atomic<bool> m_has_error;
+
+    };
+}
+
+#endif //PMP_CALL_FRAME_H
diff --git a/runtime_lib/global_thread_pool.cpp b/runtime_lib/global_thread_pool.cpp
new file mode 100644
index 0000000..5458885
--- /dev/null
+++ b/runtime_lib/global_thread_pool.cpp
@@ -0,0 +1,62 @@
+//
+// Created by anton on 6/24/22.
+//
+
+
+#include "global_thread_pool.h"
+
+
+namespace pmp::core {
+
+    // public
+    frame_id_t global_thread_pool::insert_frame(void *argument, void *scope, call_frame_target *target_function, frame_id_t parent) {
+        assert(("ThreadPool must be active when inserting! stop() or join() were called too early!", m_active));
+
+        frame_id_t id = m_last_id++;
+
+
+
+
+        return id;
+    }
+
+    void global_thread_pool::join() {
+        assert(m_active);
+
+        stop();
+
+        std::cout << "joining..." << std::endl;
+
+        for (size_t i = 0; i < m_thread_count; i++) {
+            if (m_threads[i].joinable()) {
+                m_threads[i].join();
+                std::cout << "thread " << i << " joined!" << std::endl;
+            } else {
+                std::cout << "thread " << i << " not joinable" << std::endl;
+            }
+        }
+        std::cout << "done!" << std::endl;
+
+
+        std::cout << "GlobalThreadPool finished!" << std::endl;
+    }
+
+    // private
+
+    void global_thread_pool::worker() {
+        // std::cout << "Hello from worker #" << std::this_thread::get_id() << std::endl;
+        using namespace std::chrono_literals;
+
+        call_frame* current = m_root_frame;
+
+        while (true) {
+            break;
+
+
+            if (!current->evaluate()) {
+                // someone else was faster
+
+            }
+        }
+    }
+}
\ No newline at end of file
diff --git a/runtime_lib/global_thread_pool.h b/runtime_lib/global_thread_pool.h
new file mode 100644
index 0000000..4836374
--- /dev/null
+++ b/runtime_lib/global_thread_pool.h
@@ -0,0 +1,92 @@
+//
+// Created by anton on 6/24/22.
+//
+
+#ifndef PMP_GLOBAL_THREAD_POOL_H
+#define PMP_GLOBAL_THREAD_POOL_H
+
+#include <iostream>
+#include <cassert>
+#include <functional>
+#include <unordered_map>
+#include <stack>
+#include <queue>
+#include <memory>
+#include <vector>
+#include <thread>
+#include <mutex>
+#include <chrono>
+#include <atomic>
+#include <random>
+#include <sodium.h>
+#include <unordered_set>
+#include <map>
+#include <optional>
+#include <boost/lockfree/stack.hpp>
+
+#include "call_frame.h"
+#include "helpers.h"
+
+namespace pmp::core {
+
+    class global_thread_pool {
+
+    private:
+
+        call_frame* m_root_frame;
+
+        // frame creation / lookup stuff
+        std::atomic<frame_id_t> m_last_id;
+        std::unordered_map<int, call_frame*> m_frames;
+        std::mutex m_frames_lock;
+
+
+        // thread stuff
+        std::atomic<bool> m_active;
+        const size_t m_thread_count;
+        std::vector<std::thread> m_threads;
+
+        boost::lockfree::stack<std::shared_ptr<call_frame>> stack;
+
+        void worker();
+
+    public:
+        explicit global_thread_pool(call_frame* root_frame) : m_root_frame(root_frame), m_last_id(0), m_frames(),
+        m_frames_lock(), m_active(true), m_thread_count(std::thread::hardware_concurrency()),
+        stack(100)
+        {
+            std::cout << "Init GlobalThreadPool" << std::endl;
+
+            m_threads = std::vector<std::thread>(m_thread_count);
+
+            for (size_t i = 0; i < m_thread_count; i++) {
+                m_threads[i] = std::thread([&] {
+                    worker();
+                });
+            }
+        }
+
+        ~global_thread_pool() {
+            if (m_active) {
+                std::cout << "GlobalThreadPool destructor called before join!" << std::endl;
+                join();
+            }
+
+            assert(!m_active);
+        }
+
+        frame_id_t insert_frame(void *argument, void *scope, call_frame_target *target_function, frame_id_t parent);
+
+        void stop() { m_active = false; }
+
+        void join();
+
+        size_t thread_count() const { return m_thread_count; }
+        size_t thread_count() { return m_thread_count; }
+
+
+    };
+
+}
+
+#endif //PMP_GLOBAL_THREAD_POOL_H
diff --git a/runtime_lib/helpers.h b/runtime_lib/helpers.h
new file mode 100644
index 0000000..ada5a41
--- /dev/null
+++ b/runtime_lib/helpers.h
@@ -0,0 +1,24 @@
+//
+// Created by anton on 6/24/22.
+//
+
+#ifndef PMP_HELPERS_H
+#define PMP_HELPERS_H
+
+#include <mutex>
+
+
+#define NOT_REACHED() assert(0)
+
+
+#define WITH_LOCK(lock_name) std::lock_guard<std::mutex> lock_name ## _guard(lock_name)
+
+
+
+namespace pmp::core {
+    typedef uint64_t u64;
+    typedef int64_t i64;
+
+}
+
+#endif //PMP_HELPERS_H
diff --git a/runtime_lib/main.cpp b/runtime_lib/main.cpp
index 133403a..900a57d 100644
--- a/runtime_lib/main.cpp
+++ b/runtime_lib/main.cpp
@@ -17,137 +17,17 @@
 #include <sodium.h>
 #include <unordered_set>
 #include <map>
+#include <optional>
 
+#include "call_frame.h"
+#include "global_thread_pool.h"
+#include "helpers.h"
 
-typedef void *call_frame_target(void *, void *);
+using namespace pmp::core;
 
 
 
 
-struct call_frame {
-    call_frame(void *argument, void *scope, call_frame_target *target, int id) :
-            arguments(argument), scope(scope),
-            target_function(target), id(id) {
-
-    }
-
-    void *arguments;
-    void *scope;
-    call_frame_target *target_function;
-    int id;
-};
-
-
-class GlobalThreadPool {
-public:
-    GlobalThreadPool() : m_queue(), m_queue_lock(), m_last_id(0) {
-        std::cout << "Init GlobalThreadPool" << std::endl;
-
-        m_thread_count = std::thread::hardware_concurrency();
-
-        m_threads = std::vector<std::thread>(m_thread_count);
-
-        for (size_t i = 0; i < m_thread_count; i++) {
-            m_threads[i] = std::thread([&] {
-                worker();
-            });
-        }
-    }
-
-    ~GlobalThreadPool() {
-        if (m_active) {
-            std::cout << "GlobalThreadPool destructor called before join!" << std::endl;
-            join();
-        }
-
-        assert(!m_active);
-    }
-
-    int insert(void *arguments, void *scope, call_frame_target *target_function) {
-        assert(("ThreadPool must be active when inserting! stop() or join() were called too early!", m_active));
-
-        int id = m_last_id++;
-        m_queue_lock.lock();
-        m_queue.emplace(arguments, scope, target_function, id);
-        m_queue_lock.unlock();
-
-        return id;
-    }
-
-    void stop() {
-        m_active = false;
-    }
-
-    void join() {
-        assert(m_active);
-
-        stop();
-
-        std::cout << "joining..." << std::endl;
-
-        for (size_t i = 0; i < m_thread_count; i++) {
-            if (m_threads[i].joinable()) {
-                m_threads[i].join();
-                std::cout << "thread " << i << " joined!" << std::endl;
-            } else {
-                std::cout << "thread " << i << " not joinable" << std::endl;
-            }
-        }
-        std::cout << "done!" << std::endl;
-
-
-        std::cout << "GlobalThreadPool finished!" << std::endl;
-    }
-
-    size_t thread_count() const { return m_thread_count; }
-
-private:
-
-    std::queue<struct call_frame> m_queue;
-    std::mutex m_queue_lock;
-    std::atomic<int> m_last_id;
-    std::vector<std::thread> m_threads;
-
-    std::atomic<bool> m_active = true;
-
-    size_t m_thread_count;
-
-
-    void worker() {
-        // std::cout << "Hello from worker #" << std::this_thread::get_id() << std::endl;
-        using namespace std::chrono_literals;
-
-        while (true) {
-            m_queue_lock.lock();
-
-            if (m_queue.empty()) {
-                m_queue_lock.unlock();
-                if (!m_active) {
-                    break;
-                }
-                std::this_thread::sleep_for(10ms);
-                continue;
-            }
-
-            struct call_frame item = m_queue.front();
-            m_queue.pop();
-            m_queue_lock.unlock();
-
-            try {
-                item.target_function(item.arguments, item.scope);
-            } catch (const std::exception& t) {
-                std::cerr << "Error in thread " << std::this_thread::get_id() << ": " << t.what() << "\n"
-                    << "  in work package #" << item.id << std::endl;
-            } catch (...) {
-                std::cerr << "Error in thread " << std::this_thread::get_id() << ": " << "unknown" << "\n"
-                          << "  in work package #" << item.id << std::endl;
-            }
-        }
-    }
-
-};
-
-
 void *task1(void *args, void *scope) {
     //std::cout << "hello from task 1" << std::endl;
 
@@ -216,16 +96,16 @@ int main() {
 
     std::cout << "Hello, World!" << std::endl;
 
-    auto pool = GlobalThreadPool();
+    auto pool = global_thread_pool(nullptr);
     std::mutex stdout_lock;
 
     std::cout << "Using " << pool.thread_count() << " workers" << std::endl;
 
     for (int i = 0; i < 2; i++) {
-        pool.insert(nullptr, &stdout_lock, task1);
+        pool.insert_frame(nullptr, &stdout_lock, task1, 0);
     }
 
-    std::cout << pool.insert(nullptr, &stdout_lock, task1) << std::endl;
+    std::cout << pool.insert_frame(nullptr, &stdout_lock, task1,0) << std::endl;
 
     pool.join();