diff --git a/CMakeLists.txt b/CMakeLists.txt
index 97ea97d..6525f9a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -709,6 +709,7 @@ set(APP_HEADERS
     src/data/exchange_info.h
     src/util/logger.h
     src/util/async_task_manager.h
+    src/util/serial_job_lane.h
     src/util/fetch_backoff.h
     src/util/coalescing_writer.h
     src/util/background_sampler.h
diff --git a/src/util/serial_job_lane.h b/src/util/serial_job_lane.h
new file mode 100644
index 0000000..182f464
--- /dev/null
+++ b/src/util/serial_job_lane.h
@@ -0,0 +1,144 @@
+// DragonX Wallet - ImGui Edition
+// Copyright 2024-2026 The Hush Developers
+// Released under the GPLv3
+//
+// util::SerialJobLane - a FIFO of blocking jobs run one at a time on a lazily started background thread that can be
+// ABANDONED. Each job returns a completion; completions run only inside pump(), on the owner's thread. Unlike
+// RPCWorker / AsyncTaskManager nothing here ever joins: a job stuck in a blocking call (a lite-backend save waiting
+// out a scan) is abandoned when the lane is destroyed, and its thread finishes on its own holding only the shared State.
+// post() is meant to be called from ONE producer thread (the owner); pump() from the owner thread.
+#pragma once
+
+#include <chrono>
+#include <condition_variable>
+#include <deque>
+#include <functional>
+#include <memory>
+#include <mutex>
+#include <thread>
+#include <utility>
+
+namespace dragonx {
+namespace util {
+
+class SerialJobLane {
+public:
+    using Completion = std::function<void()>;   // runs on the thread that calls pump()
+    using Job = std::function<Completion()>;    // runs on the lane thread; may return an empty Completion
+
+    SerialJobLane() : st_(std::make_shared<State>()) {}
+    ~SerialJobLane() { abandon(); }
+    SerialJobLane(const SerialJobLane&) = delete;
+    SerialJobLane& operator=(const SerialJobLane&) = delete;
+
+    // Owner thread. False if the job was dropped: the lane is abandoned, or no thread could be started.
+    bool post(Job job)
+    {
+        if (!job) return false;
+        bool spawn = false;
+        {
+            std::lock_guard<std::mutex> lk(st_->m);
+            if (st_->closed) return false;
+            st_->queue.push_back(std::move(job));
+            if (!st_->threadRunning) { st_->threadRunning = true; spawn = true; }
+        }
+        if (!spawn) return true;
+        try {
+            std::thread([st = st_] { run(st); }).detach();
+        } catch (...) {
+            std::deque<Job> dropped;               // destroyed outside the lock (may wipe captured secrets)
+            {
+                std::lock_guard<std::mutex> lk(st_->m);
+                st_->threadRunning = false;
+                dropped.swap(st_->queue);          // single producer: holds only the job just posted
+            }
+            return false;
+        }
+        return true;
+    }
+
+    // Owner thread: run every completion produced so far, in FIFO order. Returns how many ran.
+    int pump()
+    {
+        std::deque<Completion> ready;
+        {
+            std::lock_guard<std::mutex> lk(st_->m);
+            ready.swap(st_->done);
+        }
+        int n = 0;
+        for (auto& c : ready) { if (c) { c(); ++n; } }
+        return n;
+    }
+
+    bool hasCompletions() const { std::lock_guard<std::mutex> lk(st_->m); return !st_->done.empty(); }
+    bool busy() const
+    {
+        std::lock_guard<std::mutex> lk(st_->m);
+        return !st_->queue.empty() || st_->executing || !st_->done.empty();
+    }
+
+    // Bounded: true once nothing is queued or executing (completions may still await pump()).
+    bool waitIdle(std::chrono::milliseconds timeout)
+    {
+        std::unique_lock<std::mutex> lk(st_->m);
+        return st_->cv.wait_for(lk, timeout, [this] { return st_->queue.empty() && !st_->executing; });
+    }
+
+    // Drop queued jobs and undelivered completions; refuse later posts. A running job finishes by itself and its
+    // completion is discarded. Idempotent.
+    void abandon()
+    {
+        std::deque<Job> q;
+        std::deque<Completion> d;
+        {
+            std::lock_guard<std::mutex> lk(st_->m);
+            st_->closed = true;
+            q.swap(st_->queue);
+            d.swap(st_->done);
+        }   // q/d destroyed here, outside the lock
+    }
+
+private:
+    struct State {
+        mutable std::mutex m;
+        std::condition_variable cv;
+        std::deque<Job> queue;
+        std::deque<Completion> done;
+        bool threadRunning = false;
+        bool executing = false;
+        bool closed = false;
+    };
+
+    static void run(std::shared_ptr<State> st)
+    {
+        for (;;) {
+            Job job;
+            {
+                std::lock_guard<std::mutex> lk(st->m);
+                if (st->closed || st->queue.empty()) {
+                    st->threadRunning = false;
+                    st->cv.notify_all();
+                    return;
+                }
+                job = std::move(st->queue.front());
+                st->queue.pop_front();
+                st->executing = true;
+            }
+            Completion done;
+            try { done = job(); } catch (...) {}
+            job = nullptr;   // release captured state (secret holders) BEFORE reporting idle
+            {
+                std::lock_guard<std::mutex> lk(st->m);
+                st->executing = false;
+                if (done && !st->closed) st->done.push_back(std::move(done));
+                st->cv.notify_all();
+            }
+            // A completion that was not queued (closed lane) is destroyed here, on this thread, never invoked.
+        }
+    }
+
+    std::shared_ptr<State> st_;
+};
+
+} // namespace util
+} // namespace dragonx
diff --git a/tests/test_phase4.cpp b/tests/test_phase4.cpp
index 843dcb2..1c3677b 100644
--- a/tests/test_phase4.cpp
+++ b/tests/test_phase4.cpp
@@ -16,6 +16,7 @@
 #include "daemon/embedded_daemon.h"
 #include "util/connect_stall.h"
 #include "util/fetch_backoff.h"
+#include "util/serial_job_lane.h"
 #include "util/logger.h"
 #include "data/transaction_history_cache.h"
 #include "data/transaction_history_store_policy.h"
@@ -7978,6 +7979,81 @@ void testLiteWalletControllerShutdownDoesNotHangDuringSync()
     std::this_thread::sleep_for(std::chrono::milliseconds(50));  // let it unwind cleanly
 }
 
+// Tier 4 (g1-lite): SerialJobLane — completions only via pump(), FIFO, abandonable, exception-safe.
+void testSerialJobLane()
+{
+    using dragonx::util::SerialJobLane;
+    {   // completions run only inside pump(), on the pumping thread, in FIFO order
+        SerialJobLane lane;
+        std::vector<int> order;
+        const auto mainId = std::this_thread::get_id();
+        bool rightThread = true;
+        for (int i = 0; i < 3; ++i)
+            EXPECT_TRUE(lane.post([i, &order, &rightThread, mainId]() -> SerialJobLane::Completion {
+                return [i, &order, &rightThread, mainId]() {
+                    order.push_back(i);
+                    if (std::this_thread::get_id() != mainId) rightThread = false;
+                };
+            }));
+        EXPECT_TRUE(lane.waitIdle(std::chrono::milliseconds(3000)));
+        EXPECT_TRUE(order.empty());
+        EXPECT_EQ(lane.pump(), 3);
+        EXPECT_TRUE(order == std::vector<int>({0, 1, 2}));
+        EXPECT_TRUE(rightThread);
+    }
+    {   // post() never waits for a blocked job; waitIdle is bounded
+        SerialJobLane lane;
+        auto gate = std::make_shared<std::atomic<bool>>(true);
+        const auto t0 = std::chrono::steady_clock::now();
+        lane.post([gate]() -> SerialJobLane::Completion {
+            while (gate->load()) std::this_thread::sleep_for(std::chrono::milliseconds(2));
+            return {};
+        });
+        EXPECT_TRUE(std::chrono::steady_clock::now() - t0 < std::chrono::milliseconds(500));
+        EXPECT_TRUE(lane.busy());
+        EXPECT_FALSE(lane.waitIdle(std::chrono::milliseconds(50)));
+        gate->store(false);
+        EXPECT_TRUE(lane.waitIdle(std::chrono::milliseconds(3000)));
+    }
+    {   // destroying the lane with a blocked job returns at once; queued job dropped; late completion never runs
+        auto gate = std::make_shared<std::atomic<bool>>(true);
+        auto ran = std::make_shared<std::atomic<int>>(0);
+        auto entered = std::make_shared<std::atomic<bool>>(false);
+        std::chrono::steady_clock::time_point t0;
+        {
+            SerialJobLane lane;
+            lane.post([gate, ran, entered]() -> SerialJobLane::Completion {
+                entered->store(true);
+                while (gate->load()) std::this_thread::sleep_for(std::chrono::milliseconds(2));
+                ran->fetch_add(1);
+                return [ran]() { ran->fetch_add(100); };
+            });
+            lane.post([ran]() -> SerialJobLane::Completion { ran->fetch_add(10); return {}; });
+            // The first job must be RUNNING when the lane dies: a job the lane thread hasn't picked up yet is
+            // dropped unrun like the queued one.
+            for (int i = 0; i < 400 && !entered->load(); ++i) std::this_thread::sleep_for(std::chrono::milliseconds(5));
+            EXPECT_TRUE(entered->load());
+            t0 = std::chrono::steady_clock::now();
+        }
+        EXPECT_TRUE(std::chrono::steady_clock::now() - t0 < std::chrono::milliseconds(500));
+        gate->store(false);
+        for (int i = 0; i < 400 && ran->load() == 0; ++i) std::this_thread::sleep_for(std::chrono::milliseconds(5));
+        std::this_thread::sleep_for(std::chrono::milliseconds(30));
+        EXPECT_EQ(ran->load(), 1);
+    }
+    {   // a throwing job doesn't kill the lane; post() after abandon() is refused
+        SerialJobLane lane;
+        int got = 0;
+        lane.post([]() -> SerialJobLane::Completion { throw std::runtime_error("boom"); });
+        lane.post([&got]() -> SerialJobLane::Completion { return [&got]() { got = 7; }; });
+        EXPECT_TRUE(lane.waitIdle(std::chrono::milliseconds(3000)));
+        EXPECT_EQ(lane.pump(), 1);
+        EXPECT_EQ(got, 7);
+        lane.abandon();
+        EXPECT_FALSE(lane.post([]() -> SerialJobLane::Completion { return {}; }));
+    }
+}
+
 // M5b: lite-wallet rollout / kill-switch policy (wallet/lite_rollout_policy.h).
 void testLiteRolloutVersionCompare()
 {
@@ -13107,6 +13183,7 @@ int main()
     testLiteWalletGatewayRefreshSkipsFailedCommand();
     testLiteWalletControllerWorkerProducesModel();
     testLiteWalletControllerShutdownDoesNotHangDuringSync();
+    testSerialJobLane();
     testLiteRolloutVersionCompare();
     testLiteRolloutBucketAndOverrideHelpers();
     testLiteRolloutPolicyOverridesAndKillSwitch();