diff --git a/.gitignore b/.gitignore
index b95424e..38023be 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,6 +2,7 @@
 Switch-Fightstick
 GP2040-CE
 build
+build-aio
 debug
 .pycache
 *.egg-info
diff --git a/.gitmodules b/.gitmodules
new file mode 100644
index 0000000..18b03e8
--- /dev/null
+++ b/.gitmodules
@@ -0,0 +1,3 @@
+[submodule "external/bluepad32"]
+	path = external/bluepad32
+	url = https://github.com/ricardoquesada/bluepad32.git
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 10a816f..aea6efe 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -24,7 +24,20 @@ if (EXISTS ${picoVscode})
 endif()
 # ====================================================================================
 option(SWITCH_PICO_LOG "Enable UART debug logging" OFF)
+set(SWITCH_PICO_INPUT_BACKEND "UART" CACHE STRING "Controller input backend")
+set_property(CACHE SWITCH_PICO_INPUT_BACKEND PROPERTY STRINGS UART BLUEPAD32)
+if(NOT SWITCH_PICO_INPUT_BACKEND STREQUAL "UART"
+        AND NOT SWITCH_PICO_INPUT_BACKEND STREQUAL "BLUEPAD32")
+    message(FATAL_ERROR
+            "Unknown SWITCH_PICO_INPUT_BACKEND='${SWITCH_PICO_INPUT_BACKEND}'. "
+            "Expected UART or BLUEPAD32.")
+endif()
 set(PICO_BOARD pico CACHE STRING "Board type")
+if(SWITCH_PICO_INPUT_BACKEND STREQUAL "BLUEPAD32"
+        AND NOT PICO_BOARD STREQUAL "pico2_w")
+    message(FATAL_ERROR
+            "SWITCH_PICO_INPUT_BACKEND=BLUEPAD32 requires PICO_BOARD=pico2_w")
+endif()
 
 # Pull in Raspberry Pi Pico SDK (must be before project)
 include(pico_sdk_import.cmake)
@@ -34,12 +47,51 @@ project(switch-pico C CXX ASM)
 # Initialise the Raspberry Pi Pico SDK
 pico_sdk_init()
 
+# Configure BLUEPAD32 input backend if selected
+if(SWITCH_PICO_INPUT_BACKEND STREQUAL "BLUEPAD32")
+    # Ensure Python3 is available and execute patch preparation
+    find_package(Python3 REQUIRED COMPONENTS Interpreter)
+    
+    set(BLUEPAD32_PREP_SCRIPT ${CMAKE_CURRENT_LIST_DIR}/tools/prepare_bluepad32.py)
+    execute_process(
+        COMMAND ${Python3_EXECUTABLE} ${BLUEPAD32_PREP_SCRIPT}
+        WORKING_DIRECTORY ${CMAKE_CURRENT_LIST_DIR}
+        RESULT_VARIABLE BLUEPAD32_PREP_RESULT
+        OUTPUT_VARIABLE BLUEPAD32_PREP_OUTPUT
+        ERROR_VARIABLE BLUEPAD32_PREP_ERROR
+    )
+    
+    if(NOT BLUEPAD32_PREP_RESULT EQUAL 0)
+        message(FATAL_ERROR
+            "Failed to prepare Bluepad32: Patch application or validation failed. "
+            "Details: ${BLUEPAD32_PREP_ERROR}")
+    endif()
+    
+    # Configure Bluepad32 include paths and subdirectory
+    set(BLUEPAD32_ROOT ${CMAKE_CURRENT_LIST_DIR}/external/bluepad32)
+    set(BTSTACK_ROOT ${PICO_SDK_PATH}/lib/btstack)
+    include_directories(
+            ${CMAKE_CURRENT_LIST_DIR}/bluepad32_config
+            ${BTSTACK_ROOT}/3rd-party/bluedroid/encoder/include
+            ${BTSTACK_ROOT}/3rd-party/bluedroid/decoder/include
+            ${BTSTACK_ROOT}/src
+    )
+    add_subdirectory(
+            ${BLUEPAD32_ROOT}/src/components/bluepad32
+            ${CMAKE_CURRENT_BINARY_DIR}/libbluepad32
+    )
+endif()
+
 # Add executable. Default name is the project name, version 0.1
 
 add_executable(switch-pico
         switch-pico.cpp
         switch_pro_driver.cpp
 )
+if(SWITCH_PICO_INPUT_BACKEND STREQUAL "BLUEPAD32")
+    target_sources(switch-pico PRIVATE bluepad32_input_backend.cpp)
+    target_compile_definitions(switch-pico PRIVATE SWITCH_PICO_BLUEPAD32=1)
+endif()
 
 pico_set_program_name(switch-pico "switch-pico")
 pico_set_program_version(switch-pico "0.1")
@@ -57,6 +109,16 @@ target_link_libraries(switch-pico
         hardware_uart
         pico_rand
 )
+if(SWITCH_PICO_INPUT_BACKEND STREQUAL "BLUEPAD32")
+    target_link_libraries(switch-pico
+            bluepad32
+            pico_cyw43_arch_none
+            pico_btstack_ble
+            pico_btstack_classic
+            pico_btstack_cyw43
+            pico_multicore
+    )
+endif()
 
 if (SWITCH_PICO_LOG)
     target_compile_definitions(switch-pico PRIVATE SWITCH_PICO_LOG=1)
diff --git a/bluepad32_config/btstack_config.h b/bluepad32_config/btstack_config.h
new file mode 100644
index 0000000..95072bf
--- /dev/null
+++ b/bluepad32_config/btstack_config.h
@@ -0,0 +1,82 @@
+#ifndef _PICO_BTSTACK_BTSTACK_CONFIG_H
+#define _PICO_BTSTACK_BTSTACK_CONFIG_H
+
+// Based on Bluepad32's official Pico W configuration. ENABLE_BLE and
+// ENABLE_CLASSIC are supplied by the corresponding Pico SDK BTstack targets.
+#define ENABLE_LOG_INFO
+#define ENABLE_LOG_ERROR
+#define ENABLE_PRINTF_HEXDUMP
+#define ENABLE_SCO_OVER_HCI
+
+#ifdef ENABLE_BLE
+#define ENABLE_GATT_CLIENT_PAIRING
+#define ENABLE_L2CAP_LE_CREDIT_BASED_FLOW_CONTROL_MODE
+#define ENABLE_LE_CENTRAL
+#define ENABLE_LE_DATA_LENGTH_EXTENSION
+#define ENABLE_LE_PERIPHERAL
+#define ENABLE_LE_PRIVACY_ADDRESS_RESOLUTION
+#define ENABLE_LE_SECURE_CONNECTIONS
+#else
+#error "BP32: ENABLE_BLE should be defined"
+#endif
+
+#ifdef ENABLE_CLASSIC
+#define ENABLE_L2CAP_ENHANCED_RETRANSMISSION_MODE
+#define ENABLE_GOEP_L2CAP
+#else
+#error "BP32: ENABLE_CLASSIC should be defined"
+#endif
+
+#if defined(ENABLE_CLASSIC) && defined(ENABLE_BLE)
+#define ENABLE_CROSS_TRANSPORT_KEY_DERIVATION
+#endif
+
+#define HCI_OUTGOING_PRE_BUFFER_SIZE 4
+#define HCI_ACL_PAYLOAD_SIZE (1691 + 4)
+#define HCI_ACL_CHUNK_SIZE_ALIGNMENT 4
+#define MAX_NR_AVDTP_CONNECTIONS 1
+#define MAX_NR_AVDTP_STREAM_ENDPOINTS 1
+#define MAX_NR_AVRCP_CONNECTIONS 2
+#define MAX_NR_BNEP_CHANNELS 1
+#define MAX_NR_BNEP_SERVICES 1
+#define MAX_NR_BTSTACK_LINK_KEY_DB_MEMORY_ENTRIES 2
+#define MAX_NR_GATT_CLIENTS 1
+#define MAX_NR_HCI_CONNECTIONS 4
+#define MAX_NR_HID_HOST_CONNECTIONS 1
+#define MAX_NR_HIDS_CLIENTS 1
+#define MAX_NR_HFP_CONNECTIONS 1
+#define MAX_NR_L2CAP_CHANNELS 6
+#define MAX_NR_L2CAP_SERVICES 5
+#define MAX_NR_RFCOMM_CHANNELS 1
+#define MAX_NR_RFCOMM_MULTIPLEXERS 1
+#define MAX_NR_RFCOMM_SERVICES 1
+#define MAX_NR_SERVICE_RECORD_ITEMS 4
+#define MAX_NR_SM_LOOKUP_ENTRIES 3
+#define MAX_NR_WHITELIST_ENTRIES 16
+#define MAX_NR_LE_DEVICE_DB_ENTRIES 16
+
+// Keep controller buffers and controller-to-host flow control enabled to avoid
+// overrunning the shared CYW43 bus.
+#define MAX_NR_CONTROLLER_ACL_BUFFERS 3
+#define MAX_NR_CONTROLLER_SCO_PACKETS 3
+#define ENABLE_HCI_CONTROLLER_TO_HOST_FLOW_CONTROL
+#define HCI_HOST_ACL_PACKET_LEN 1024
+#define HCI_HOST_ACL_PACKET_NUM 3
+#define HCI_HOST_SCO_PACKET_LEN 120
+#define HCI_HOST_SCO_PACKET_NUM 3
+
+// Persistent Classic and BLE pairing databases use Pico flash-backed TLV.
+#define NVM_NUM_DEVICE_DB_ENTRIES 16
+#define NVM_NUM_LINK_KEYS 16
+
+// Bluepad32 does not provide malloc to BTstack.
+#define MAX_ATT_DB_SIZE 512
+
+#define HAVE_EMBEDDED_TIME_MS
+#define HAVE_ASSERT
+#define HCI_RESET_RESEND_TIMEOUT_MS 1000
+#define ENABLE_SOFTWARE_AES128
+#define ENABLE_MICRO_ECC_FOR_LE_SECURE_CONNECTIONS
+#define HAVE_BTSTACK_STDIN
+
+#endif  // _PICO_BTSTACK_BTSTACK_CONFIG_H
diff --git a/bluepad32_config/sdkconfig.h b/bluepad32_config/sdkconfig.h
new file mode 100644
index 0000000..83002a9
--- /dev/null
+++ b/bluepad32_config/sdkconfig.h
@@ -0,0 +1,13 @@
+#pragma once
+
+// Bluepad32's Pico W example configuration, limited to one live controller.
+#define CONFIG_BLUEPAD32_MAX_DEVICES 1
+#define CONFIG_BLUEPAD32_MAX_ALLOWLIST 1
+#define CONFIG_BLUEPAD32_GAP_SECURITY 1
+#define CONFIG_BLUEPAD32_ENABLE_BLE_BY_DEFAULT 1
+
+#define CONFIG_BLUEPAD32_PLATFORM_CUSTOM
+#define CONFIG_TARGET_PICO_W
+
+// 2 == Info
+#define CONFIG_BLUEPAD32_LOG_LEVEL 2
diff --git a/bluepad32_input_backend.cpp b/bluepad32_input_backend.cpp
new file mode 100644
index 0000000..6bd3c2f
--- /dev/null
+++ b/bluepad32_input_backend.cpp
@@ -0,0 +1,394 @@
+#include "bluepad32_input_backend.h"
+
+#include <limits.h>
+#include <stddef.h>
+#include <string.h>
+
+#include <btstack_run_loop.h>
+#include <pico/critical_section.h>
+#include <pico/cyw43_arch.h>
+#include <pico/multicore.h>
+#include <pico/stdlib.h>
+#include <pico/util/queue.h>
+#include <uni.h>
+
+namespace {
+
+constexpr uint16_t kStickMidpoint = 32768;
+constexpr int32_t kAxisMinimum = -512;
+constexpr int32_t kAxisMaximum = 511;
+constexpr int32_t kTriggerMaximum = 1023;
+constexpr int32_t kTriggerThreshold = (kTriggerMaximum * 35) / 100;
+constexpr uint16_t kRumbleDurationMs = 50;
+constexpr uint32_t kRumblePollIntervalMs = 5;
+constexpr uint kRumbleQueueDepth = 8;
+
+struct RumblePacket {
+    uint8_t bytes[8];
+};
+
+critical_section_t g_state_lock;
+queue_t g_rumble_queue;
+SwitchInputState g_shared_state;
+bool g_shared_controller_active = false;
+uint32_t g_shared_generation = 0;
+
+// These generations are only read or written by Core 0.
+uint32_t g_consumed_generation = 0;
+uint32_t g_last_snapshot_generation = 0;
+bool g_initialized = false;
+bool g_started = false;
+
+// This pointer and the timer are only read or written by Core 1 / BTstack.
+uni_hid_device_t* g_active_device = nullptr;
+btstack_timer_source_t g_rumble_timer{};
+
+SwitchInputState make_neutral_state() {
+    SwitchInputState state{};
+    state.lx = kStickMidpoint;
+    state.ly = kStickMidpoint;
+    state.rx = kStickMidpoint;
+    state.ry = kStickMidpoint;
+    return state;
+}
+
+void publish_state(const SwitchInputState& state, bool controller_active) {
+    critical_section_enter_blocking(&g_state_lock);
+    g_shared_state = state;
+    g_shared_controller_active = controller_active;
+    ++g_shared_generation;
+    critical_section_exit(&g_state_lock);
+}
+
+int32_t clamp_axis(int32_t value) {
+    if (value < kAxisMinimum) {
+        return kAxisMinimum;
+    }
+    if (value > kAxisMaximum) {
+        return kAxisMaximum;
+    }
+    return value;
+}
+
+uint16_t scale_stick(int32_t value) {
+    value = clamp_axis(value);
+    if (value <= 0) {
+        return static_cast<uint16_t>(
+            (static_cast<int64_t>(value - kAxisMinimum) * kStickMidpoint) / -kAxisMinimum);
+    }
+    return static_cast<uint16_t>(
+        kStickMidpoint + (static_cast<int64_t>(value) * (UINT16_MAX - kStickMidpoint)) / kAxisMaximum);
+}
+
+int16_t clamp_int16(int64_t value) {
+    if (value < INT16_MIN) {
+        return INT16_MIN;
+    }
+    if (value > INT16_MAX) {
+        return INT16_MAX;
+    }
+    return static_cast<int16_t>(value);
+}
+
+int64_t divide_round_nearest(int64_t numerator, int64_t denominator) {
+    if (numerator >= 0) {
+        return (numerator + denominator / 2) / denominator;
+    }
+    return -((-numerator + denominator / 2) / denominator);
+}
+
+int16_t convert_accel(int64_t q13_value) {
+    return clamp_int16(q13_value / 2);
+}
+
+int16_t convert_gyro(int64_t q10_value) {
+    constexpr int64_t kNumeratorScale = 13371;
+    constexpr int64_t kDenominator = 1024 * 936;
+    return clamp_int16(divide_round_nearest(q10_value * kNumeratorScale, kDenominator));
+}
+
+bool has_motion(const uni_gamepad_t& gamepad) {
+    for (size_t i = 0; i < 3; ++i) {
+        if (gamepad.accel[i] != 0 || gamepad.gyro[i] != 0) {
+            return true;
+        }
+    }
+    return false;
+}
+
+SwitchInputState map_gamepad(const uni_gamepad_t& gamepad) {
+    SwitchInputState state = make_neutral_state();
+
+    state.dpad_up = (gamepad.dpad & DPAD_UP) != 0;
+    state.dpad_down = (gamepad.dpad & DPAD_DOWN) != 0;
+    state.dpad_left = (gamepad.dpad & DPAD_LEFT) != 0;
+    state.dpad_right = (gamepad.dpad & DPAD_RIGHT) != 0;
+
+    // Bluepad32's A/B/X/Y are positional: south/east/west/north.
+    state.button_b = (gamepad.buttons & BUTTON_A) != 0;
+    state.button_a = (gamepad.buttons & BUTTON_B) != 0;
+    state.button_y = (gamepad.buttons & BUTTON_X) != 0;
+    state.button_x = (gamepad.buttons & BUTTON_Y) != 0;
+    state.button_l = (gamepad.buttons & BUTTON_SHOULDER_L) != 0;
+    state.button_r = (gamepad.buttons & BUTTON_SHOULDER_R) != 0;
+    state.button_zl = (gamepad.buttons & BUTTON_TRIGGER_L) != 0 || gamepad.brake >= kTriggerThreshold;
+    state.button_zr = (gamepad.buttons & BUTTON_TRIGGER_R) != 0 || gamepad.throttle >= kTriggerThreshold;
+    state.button_l3 = (gamepad.buttons & BUTTON_THUMB_L) != 0;
+    state.button_r3 = (gamepad.buttons & BUTTON_THUMB_R) != 0;
+
+    state.button_minus = (gamepad.misc_buttons & MISC_BUTTON_SELECT) != 0;
+    state.button_plus = (gamepad.misc_buttons & MISC_BUTTON_START) != 0;
+    state.button_home = (gamepad.misc_buttons & MISC_BUTTON_SYSTEM) != 0;
+    state.button_capture = (gamepad.misc_buttons & MISC_BUTTON_CAPTURE) != 0;
+
+    state.lx = scale_stick(gamepad.axis_x);
+    state.ly = scale_stick(gamepad.axis_y);
+    state.rx = scale_stick(gamepad.axis_rx);
+    state.ry = scale_stick(gamepad.axis_ry);
+
+    if (has_motion(gamepad)) {
+        // Dependency patches normalize both arrays to SDL3 PlayStation axes.
+        SwitchImuSample sample{};
+        sample.accel_x = convert_accel(-static_cast<int64_t>(gamepad.accel[2]));
+        sample.accel_y = convert_accel(-static_cast<int64_t>(gamepad.accel[0]));
+        sample.accel_z = convert_accel(gamepad.accel[1]);
+        sample.gyro_x = convert_gyro(-static_cast<int64_t>(gamepad.gyro[2]));
+        sample.gyro_y = convert_gyro(-static_cast<int64_t>(gamepad.gyro[0]));
+        sample.gyro_z = convert_gyro(gamepad.gyro[1]);
+        state.imu_sample_count = 3;
+        for (SwitchImuSample& destination : state.imu_samples) {
+            destination = sample;
+        }
+    }
+
+    return state;
+}
+
+void decode_rumble(const uint8_t bytes[8], uint8_t* left_magnitude, uint8_t* right_magnitude) {
+    static constexpr uint8_t kNeutralPacket[8] = {0x00, 0x01, 0x40, 0x40, 0x00, 0x01, 0x40, 0x40};
+    if (memcmp(bytes, kNeutralPacket, sizeof(kNeutralPacket)) == 0) {
+        *left_magnitude = 0;
+        *right_magnitude = 0;
+        return;
+    }
+
+    uint16_t right_raw = static_cast<uint16_t>(((bytes[1] & 0x03) << 8) | bytes[0]);
+    uint16_t left_raw = static_cast<uint16_t>(((bytes[5] & 0x03) << 8) | bytes[4]);
+    if (left_raw < 8 && right_raw < 8) {
+        left_raw = 0;
+        right_raw = 0;
+    }
+
+    *left_magnitude = static_cast<uint8_t>((left_raw * UINT8_MAX + 511) / 1023);
+    *right_magnitude = static_cast<uint8_t>((right_raw * UINT8_MAX + 511) / 1023);
+}
+
+void process_rumble_timer(btstack_timer_source_t* timer) {
+    RumblePacket packet{};
+    RumblePacket latest{};
+    bool have_packet = false;
+    while (queue_try_remove(&g_rumble_queue, &packet)) {
+        latest = packet;
+        have_packet = true;
+    }
+
+    if (have_packet && g_active_device != nullptr &&
+        g_active_device->report_parser.play_dual_rumble != nullptr) {
+        uint8_t left_magnitude = 0;
+        uint8_t right_magnitude = 0;
+        decode_rumble(latest.bytes, &left_magnitude, &right_magnitude);
+        // Bluepad orders the weak (high-frequency) motor before the strong
+        // (low-frequency) motor; the project decoder names those right/left.
+        g_active_device->report_parser.play_dual_rumble(
+            g_active_device, 0, kRumbleDurationMs, right_magnitude, left_magnitude);
+    }
+
+    btstack_run_loop_set_timer(timer, kRumblePollIntervalMs);
+    btstack_run_loop_add_timer(timer);
+}
+
+void platform_init(int argc, const char** argv) {
+    (void)argc;
+    (void)argv;
+}
+
+void platform_on_init_complete() {
+    btstack_run_loop_set_timer_handler(&g_rumble_timer, process_rumble_timer);
+    btstack_run_loop_set_timer(&g_rumble_timer, kRumblePollIntervalMs);
+    btstack_run_loop_add_timer(&g_rumble_timer);
+
+    uni_bt_allow_incoming_connections(true);
+    uni_bt_start_scanning_and_autoconnect_unsafe();
+}
+
+uni_error_t platform_on_device_discovered(bd_addr_t addr, const char* name, uint16_t cod, uint8_t rssi) {
+    (void)addr;
+    (void)name;
+    (void)cod;
+    (void)rssi;
+    return g_active_device == nullptr ? UNI_ERROR_SUCCESS : UNI_ERROR_IGNORE_DEVICE;
+}
+
+void platform_on_device_connected(uni_hid_device_t* device) {
+    (void)device;
+}
+
+void resume_connections() {
+    uni_bt_allow_incoming_connections(true);
+    uni_bt_start_scanning_and_autoconnect_unsafe();
+}
+
+void platform_on_device_disconnected(uni_hid_device_t* device) {
+    if (device == g_active_device) {
+        g_active_device = nullptr;
+        publish_state(make_neutral_state(), false);
+        resume_connections();
+    } else if (g_active_device == nullptr) {
+        resume_connections();
+    }
+}
+
+uni_error_t platform_on_device_ready(uni_hid_device_t* device) {
+    if (!uni_hid_device_is_gamepad(device)) {
+        return UNI_ERROR_INVALID_CONTROLLER;
+    }
+    if (g_active_device != nullptr && g_active_device != device) {
+        return UNI_ERROR_NO_SLOTS;
+    }
+
+    g_active_device = device;
+    publish_state(make_neutral_state(), true);
+    uni_bt_stop_scanning_unsafe();
+    uni_bt_allow_incoming_connections(false);
+    return UNI_ERROR_SUCCESS;
+}
+
+void platform_on_controller_data(uni_hid_device_t* device, uni_controller_t* controller) {
+    if (device != g_active_device || controller == nullptr || controller->klass != UNI_CONTROLLER_CLASS_GAMEPAD) {
+        return;
+    }
+    publish_state(map_gamepad(controller->gamepad), true);
+}
+
+const uni_property_t* platform_get_property(uni_property_idx_t index) {
+    (void)index;
+    return nullptr;
+}
+
+void platform_on_oob_event(uni_platform_oob_event_t event, void* data) {
+    (void)event;
+    (void)data;
+}
+
+uni_platform* get_platform() {
+    static uni_platform platform = {
+        "Switch Pico",
+        platform_init,
+        platform_on_init_complete,
+        platform_on_device_discovered,
+        platform_on_device_connected,
+        platform_on_device_disconnected,
+        platform_on_device_ready,
+        nullptr,
+        platform_on_controller_data,
+        platform_get_property,
+        platform_on_oob_event,
+        nullptr,
+        nullptr,
+    };
+    return &platform;
+}
+
+[[noreturn]] void core1_main() {
+    if (cyw43_arch_init() != 0) {
+        publish_state(make_neutral_state(), false);
+        while (true) {
+            tight_loop_contents();
+        }
+    }
+
+    uni_platform_set_custom(get_platform());
+    if (uni_init(0, nullptr) != 0) {
+        publish_state(make_neutral_state(), false);
+        while (true) {
+            tight_loop_contents();
+        }
+    }
+
+    btstack_run_loop_execute();
+    while (true) {
+        tight_loop_contents();
+    }
+}
+
+}  // namespace
+
+void bluepad32_input_backend_init() {
+    if (g_initialized) {
+        return;
+    }
+
+    critical_section_init(&g_state_lock);
+    queue_init(&g_rumble_queue, sizeof(RumblePacket), kRumbleQueueDepth);
+    g_shared_state = make_neutral_state();
+    g_shared_controller_active = false;
+    g_shared_generation = 0;
+    g_consumed_generation = 0;
+    g_last_snapshot_generation = 0;
+    g_initialized = true;
+}
+
+void bluepad32_input_backend_start() {
+    if (!g_initialized) {
+        bluepad32_input_backend_init();
+    }
+    if (g_started) {
+        return;
+    }
+
+    g_started = true;
+    multicore_launch_core1(core1_main);
+}
+
+bool bluepad32_input_backend_snapshot(SwitchInputState* out) {
+    if (out == nullptr) {
+        return false;
+    }
+    if (!g_initialized) {
+        *out = make_neutral_state();
+        return false;
+    }
+
+    critical_section_enter_blocking(&g_state_lock);
+    *out = g_shared_state;
+    const bool controller_active = g_shared_controller_active;
+    const uint32_t generation = g_shared_generation;
+    critical_section_exit(&g_state_lock);
+
+    if (generation == g_consumed_generation) {
+        out->imu_sample_count = 0;
+    }
+    g_last_snapshot_generation = generation;
+    return controller_active;
+}
+
+void bluepad32_input_backend_report_sent() {
+    if (!g_initialized) {
+        return;
+    }
+    g_consumed_generation = g_last_snapshot_generation;
+}
+
+void bluepad32_input_backend_queue_rumble(const uint8_t rumble[8]) {
+    if (!g_initialized || rumble == nullptr) {
+        return;
+    }
+
+    RumblePacket packet{};
+    memcpy(packet.bytes, rumble, sizeof(packet.bytes));
+    if (!queue_try_add(&g_rumble_queue, &packet)) {
+        RumblePacket discarded{};
+        (void)queue_try_remove(&g_rumble_queue, &discarded);
+        (void)queue_try_add(&g_rumble_queue, &packet);
+    }
+}
diff --git a/bluepad32_input_backend.h b/bluepad32_input_backend.h
new file mode 100644
index 0000000..df5cc56
--- /dev/null
+++ b/bluepad32_input_backend.h
@@ -0,0 +1,11 @@
+#pragma once
+
+#include <stdint.h>
+
+#include "switch_pro_driver.h"
+
+void bluepad32_input_backend_init();
+void bluepad32_input_backend_start();
+bool bluepad32_input_backend_snapshot(SwitchInputState* out);
+void bluepad32_input_backend_report_sent();
+void bluepad32_input_backend_queue_rumble(const uint8_t rumble[8]);
diff --git a/build.py b/build.py
index 95c537c..8b728d3 100644
--- a/build.py
+++ b/build.py
@@ -12,9 +12,12 @@ from pathlib import Path
 SCRIPT_DIR = Path(__file__).resolve().parent
 CONFIG_FILE = SCRIPT_DIR / "controller_color_config.h"
 BUILD_DIR = SCRIPT_DIR / "build"
+AIO_BUILD_DIR = SCRIPT_DIR / "build-aio"
 FIRMWARE_DIR = SCRIPT_DIR / "firmware"
 FIRMWARE_ELF_PATH = FIRMWARE_DIR / "switch-pico.elf"
 FIRMWARE_UF2_PATH = FIRMWARE_DIR / "switch-pico.uf2"
+AIO_FIRMWARE_ELF_PATH = FIRMWARE_DIR / "switch-pico-aio.elf"
+AIO_FIRMWARE_UF2_PATH = FIRMWARE_DIR / "switch-pico-aio.uf2"
 
 ELF_PATH = Path(os.environ.get("ELF_PATH", BUILD_DIR / "switch-pico.elf")).expanduser()
 UF2_PATH = Path(os.environ.get("UF2_PATH", BUILD_DIR / "switch-pico.uf2")).expanduser()
@@ -34,6 +37,11 @@ def parse_args():
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="Default behavior leaves controller_color_config.h unchanged.",
     )
+    parser.add_argument(
+        "--aio",
+        action="store_true",
+        help="Build and flash the Pico 2 W Bluepad32 all-in-one firmware.",
+    )
     group = parser.add_mutually_exclusive_group()
     group.add_argument(
         "--random-grip-color",
@@ -111,41 +119,68 @@ def resolve_picotool():
     sys.stderr.write("Error: picotool not found. Put it on your PATH or set PICOTOOL_PATH.\n")
     sys.exit(1)
 
-def build():
+def build(
+    aio,
+    build_dir,
+    elf_path,
+    uf2_path,
+    firmware_elf_path,
+    firmware_uf2_path,
+):
+    if aio:
+        run_cmd([sys.executable, str(SCRIPT_DIR / "tools" / "prepare_bluepad32.py")])
+        definitions = [
+            "-DSWITCH_PICO_LOG=OFF",
+            "-DPICO_BOARD=pico2_w",
+            "-DSWITCH_PICO_INPUT_BACKEND=BLUEPAD32",
+        ]
+    else:
+        definitions = [
+            "-DSWITCH_PICO_LOG=OFF",
+            "-DPICO_BOARD=pico",
+            "-DSWITCH_PICO_INPUT_BACKEND=UART",
+        ]
+
     run_cmd(
         [
             "cmake",
             "-S",
             str(SCRIPT_DIR),
             "-B",
-            str(BUILD_DIR),
-            "-DSWITCH_PICO_LOG=OFF",
+            str(build_dir),
+            *definitions,
         ]
     )
-    run_cmd(["cmake", "--build", str(BUILD_DIR)])
+    run_cmd(["cmake", "--build", str(build_dir)])
 
-    missing_artifacts = [path for path in (ELF_PATH, UF2_PATH) if not path.is_file()]
+    missing_artifacts = [
+        path for path in (elf_path, uf2_path) if not path.is_file()
+    ]
     if missing_artifacts:
         missing = ", ".join(str(path) for path in missing_artifacts)
         sys.stderr.write(f"Error: Build did not produce required artifact(s): {missing}\n")
         sys.exit(1)
     FIRMWARE_DIR.mkdir(parents=True, exist_ok=True)
-    shutil.copy2(ELF_PATH, FIRMWARE_ELF_PATH)
-    shutil.copy2(UF2_PATH, FIRMWARE_UF2_PATH)
+    shutil.copy2(elf_path, firmware_elf_path)
+    shutil.copy2(uf2_path, firmware_uf2_path)
 
-    print(f"Built ELF: {ELF_PATH}")
-    print(f"Built UF2: {UF2_PATH}")
-    print(f"Copied ELF: {FIRMWARE_ELF_PATH}")
-    print(f"Copied UF2: {FIRMWARE_UF2_PATH}")
+    print(f"Built ELF: {elf_path}")
+    print(f"Built UF2: {uf2_path}")
+    print(f"Copied ELF: {firmware_elf_path}")
+    print(f"Copied UF2: {firmware_uf2_path}")
 
-def flash():
+
+def flash(elf_path, allow_elf_override):
     picotool = resolve_picotool()
-    if not ELF_PATH.exists():
-        sys.stderr.write(
-            f"Error: Cannot find ELF at {ELF_PATH}. Set ELF_PATH to override.\n"
-        )
+    if not elf_path.exists():
+        if allow_elf_override:
+            sys.stderr.write(
+                f"Error: Cannot find ELF at {elf_path}. Set ELF_PATH to override.\n"
+            )
+        else:
+            sys.stderr.write(f"Error: Cannot find ELF at {elf_path}.\n")
         sys.exit(1)
-    run_cmd([str(picotool), "load", str(ELF_PATH), "-fx"])
+    run_cmd([str(picotool), "load", str(elf_path), "-fx"])
 
 def main():
     args = parse_args()
@@ -164,8 +199,28 @@ def main():
         update_grip_colors(color)
         print(f"Grip color set to #{color} in {CONFIG_FILE.name}")
 
-    build()
-    flash()
+    if args.aio:
+        build_dir = AIO_BUILD_DIR
+        elf_path = AIO_BUILD_DIR / "switch-pico.elf"
+        uf2_path = AIO_BUILD_DIR / "switch-pico.uf2"
+        firmware_elf_path = AIO_FIRMWARE_ELF_PATH
+        firmware_uf2_path = AIO_FIRMWARE_UF2_PATH
+    else:
+        build_dir = BUILD_DIR
+        elf_path = ELF_PATH
+        uf2_path = UF2_PATH
+        firmware_elf_path = FIRMWARE_ELF_PATH
+        firmware_uf2_path = FIRMWARE_UF2_PATH
+
+    build(
+        args.aio,
+        build_dir,
+        elf_path,
+        uf2_path,
+        firmware_elf_path,
+        firmware_uf2_path,
+    )
+    flash(elf_path, allow_elf_override=not args.aio)
 
 if __name__ == "__main__":
     main()
diff --git a/external/bluepad32 b/external/bluepad32
new file mode 160000
index 0000000..6efa712
--- /dev/null
+++ b/external/bluepad32
@@ -0,0 +1 @@
+Subproject commit 6efa7123fe8badf5a40ad1205743a80b31c00ea4
diff --git a/firmware/switch-pico-aio.elf b/firmware/switch-pico-aio.elf
new file mode 100755
index 0000000..166e1b1
Binary files /dev/null and b/firmware/switch-pico-aio.elf differ
diff --git a/firmware/switch-pico-aio.uf2 b/firmware/switch-pico-aio.uf2
new file mode 100644
index 0000000..8e4968d
Binary files /dev/null and b/firmware/switch-pico-aio.uf2 differ
diff --git a/patches/bluepad32-sdl3-imu.patch b/patches/bluepad32-sdl3-imu.patch
new file mode 100644
index 0000000..7060f78
--- /dev/null
+++ b/patches/bluepad32-sdl3-imu.patch
@@ -0,0 +1,154 @@
+diff --git a/src/components/bluepad32/parser/uni_hid_parser_ds4.c b/src/components/bluepad32/parser/uni_hid_parser_ds4.c
+index ea063b8..7670caf 100644
+--- a/src/components/bluepad32/parser/uni_hid_parser_ds4.c
++++ b/src/components/bluepad32/parser/uni_hid_parser_ds4.c
+@@ -297,17 +297,17 @@ void uni_hid_parser_ds4_parse_feature_report(uni_hid_device_t* d, const uint8_t*
+             // Set gyroscope calibration and normalization parameters.
+             // Data values will be normalized to 1/DS_GYRO_RES_PER_DEG_S degree/s.
+             speed_2x = r->gyro_speed_plus + r->gyro_speed_minus;
+-            ins->gyro_calib_data[0].bias = 0;
++            ins->gyro_calib_data[0].bias = r->gyro_pitch_bias;
+             ins->gyro_calib_data[0].sens_numer = speed_2x * DS4_GYRO_RES_PER_DEG_S;
+             ins->gyro_calib_data[0].sens_denom =
+                 abs(r->gyro_pitch_plus - r->gyro_pitch_bias) + abs(r->gyro_pitch_minus + r->gyro_pitch_bias);
+ 
+-            ins->gyro_calib_data[1].bias = 0;
++            ins->gyro_calib_data[1].bias = r->gyro_yaw_bias;
+             ins->gyro_calib_data[1].sens_numer = speed_2x * DS4_GYRO_RES_PER_DEG_S;
+             ins->gyro_calib_data[1].sens_denom =
+                 abs(r->gyro_yaw_plus - r->gyro_yaw_bias) + abs(r->gyro_yaw_minus - r->gyro_yaw_bias);
+ 
+-            ins->gyro_calib_data[2].bias = 0;
++            ins->gyro_calib_data[2].bias = r->gyro_roll_bias;
+             ins->gyro_calib_data[2].sens_numer = speed_2x * DS4_GYRO_RES_PER_DEG_S;
+             ins->gyro_calib_data[2].sens_denom =
+                 abs(r->gyro_roll_plus - r->gyro_roll_bias) + abs(r->gyro_roll_minus - r->gyro_roll_bias);
+@@ -476,7 +476,7 @@ static void ds4_parse_input_report_11(uni_hid_device_t* d, const ds4_input_repor
+ 
+     // Gyro
+     for (size_t i = 0; i < ARRAY_SIZE(r->gyro); i++) {
+-        int32_t raw_data = (int16_t)r->gyro[i];
++        int32_t raw_data = (int16_t)r->gyro[i] - ins->gyro_calib_data[i].bias;
+         int32_t calib_data =
+             mult_frac(ins->gyro_calib_data[i].sens_numer, raw_data, ins->gyro_calib_data[i].sens_denom);
+         ctl->gamepad.gyro[i] = calib_data;
+@@ -484,7 +484,7 @@ static void ds4_parse_input_report_11(uni_hid_device_t* d, const ds4_input_repor
+ 
+     // Accel
+     for (size_t i = 0; i < ARRAY_SIZE(r->accel); i++) {
+-        int32_t raw_data = (int16_t)r->accel[i];
++        int32_t raw_data = (int16_t)r->accel[i] - ins->accel_calib_data[i].bias;
+         int32_t calib_data =
+             mult_frac(ins->accel_calib_data[i].sens_numer, raw_data, ins->accel_calib_data[i].sens_denom);
+         ctl->gamepad.accel[i] = calib_data;
+diff --git a/src/components/bluepad32/parser/uni_hid_parser_ds5.c b/src/components/bluepad32/parser/uni_hid_parser_ds5.c
+index a22ef26..3d5ecef 100644
+--- a/src/components/bluepad32/parser/uni_hid_parser_ds5.c
++++ b/src/components/bluepad32/parser/uni_hid_parser_ds5.c
+@@ -487,17 +487,17 @@ void uni_hid_parser_ds5_parse_feature_report(uni_hid_device_t* d, const uint8_t*
+             // Set gyroscope calibration and normalization parameters.
+             // Data values will be normalized to 1/DS_GYRO_RES_PER_DEG_S degree/s.
+             speed_2x = r->gyro_speed_plus + r->gyro_speed_minus;
+-            ins->gyro_calib_data[0].bias = 0;
++            ins->gyro_calib_data[0].bias = r->gyro_pitch_bias;
+             ins->gyro_calib_data[0].sens_numer = speed_2x * DS5_GYRO_RES_PER_DEG_S;
+             ins->gyro_calib_data[0].sens_denom =
+                 abs(r->gyro_pitch_plus - r->gyro_pitch_bias) + abs(r->gyro_pitch_minus + r->gyro_pitch_bias);
+ 
+-            ins->gyro_calib_data[1].bias = 0;
++            ins->gyro_calib_data[1].bias = r->gyro_yaw_bias;
+             ins->gyro_calib_data[1].sens_numer = speed_2x * DS5_GYRO_RES_PER_DEG_S;
+             ins->gyro_calib_data[1].sens_denom =
+                 abs(r->gyro_yaw_plus - r->gyro_yaw_bias) + abs(r->gyro_yaw_minus - r->gyro_yaw_bias);
+ 
+-            ins->gyro_calib_data[2].bias = 0;
++            ins->gyro_calib_data[2].bias = r->gyro_roll_bias;
+             ins->gyro_calib_data[2].sens_numer = speed_2x * DS5_GYRO_RES_PER_DEG_S;
+             ins->gyro_calib_data[2].sens_denom =
+                 abs(r->gyro_roll_plus - r->gyro_roll_bias) + abs(r->gyro_roll_minus - r->gyro_roll_bias);
+@@ -622,7 +622,7 @@ void uni_hid_parser_ds5_parse_input_report(uni_hid_device_t* d, const uint8_t* r
+ 
+     // Gyro
+     for (size_t i = 0; i < ARRAY_SIZE(r->gyro); i++) {
+-        int32_t raw_data = (int16_t)r->gyro[i];
++        int32_t raw_data = (int16_t)r->gyro[i] - ins->gyro_calib_data[i].bias;
+         int32_t calib_data =
+             mult_frac(ins->gyro_calib_data[i].sens_numer, raw_data, ins->gyro_calib_data[i].sens_denom);
+         ctl->gamepad.gyro[i] = calib_data;
+@@ -630,7 +630,7 @@ void uni_hid_parser_ds5_parse_input_report(uni_hid_device_t* d, const uint8_t* r
+ 
+     // Accel
+     for (size_t i = 0; i < ARRAY_SIZE(r->accel); i++) {
+-        int32_t raw_data = (int16_t)r->accel[i];
++        int32_t raw_data = (int16_t)r->accel[i] - ins->accel_calib_data[i].bias;
+         int32_t calib_data =
+             mult_frac(ins->accel_calib_data[i].sens_numer, raw_data, ins->accel_calib_data[i].sens_denom);
+         ctl->gamepad.accel[i] = calib_data;
+diff --git a/src/components/bluepad32/parser/uni_hid_parser_switch.c b/src/components/bluepad32/parser/uni_hid_parser_switch.c
+index 599fc35..c72f056 100644
+--- a/src/components/bluepad32/parser/uni_hid_parser_switch.c
++++ b/src/components/bluepad32/parser/uni_hid_parser_switch.c
+@@ -51,7 +51,8 @@ static const int16_t DEFAULT_ACCEL_OFFSET = 0;
+ static const int16_t DEFAULT_ACCEL_SCALE = 16384;
+ static const int16_t DEFAULT_GYRO_OFFSET = 0;
+ static const int16_t DEFAULT_GYRO_SCALE = 13371;
+-#define SWITCH_IMU_PREC_RANGE_SCALE 1000
++#define SWITCH_IMU_GYRO_RES_PER_DEG_S 1024
++#define SWITCH_IMU_ACCEL_RES_PER_G 8192
+ 
+ #define SWITCH_FACTORY_IMU_CAL_DATA_SIZE 24
+ static const uint16_t SWITCH_FACTORY_IMU_CAL_DATA_ADDR = 0x6020;
+@@ -823,19 +824,26 @@ static void parse_imu(uni_hid_device_t* d, const struct switch_imu_data_s* r) {
+     switch_instance_t* ins = get_switch_instance(d);
+     uni_controller_t* ctl = &d->controller;
+ 
+-    int accel[3];
+-    int gyro[3];
++    int32_t accel[3];
++    int32_t gyro[3];
+ 
+     for (int i = 0; i < 3; i++) {
+-        if (ins->imu_cal_accel_divisor[i] == 0)
+-            accel[i] = r->accel[i];
+-        else
+-            accel[i] = (r->accel[i] * ins->cal_accel.scale[i]) / ins->imu_cal_accel_divisor[i];
+-        gyro[i] = mult_frac((SWITCH_IMU_PREC_RANGE_SCALE * (r->gyro[i] - ins->cal_gyro.offset[i])),
+-                            ins->cal_gyro.scale[i], ins->imu_cal_gyro_divisor[i]);
++        if (ins->imu_cal_accel_divisor[i] == 0) {
++            accel[i] = r->accel[i] * 2;
++        } else {
++            accel[i] = mult_frac(r->accel[i], 4 * SWITCH_IMU_ACCEL_RES_PER_G, ins->imu_cal_accel_divisor[i]);
++        }
++
++        if (ins->imu_cal_gyro_divisor[i] == 0) {
++            gyro[i] = mult_frac(r->gyro[i], 936 * SWITCH_IMU_GYRO_RES_PER_DEG_S, DEFAULT_GYRO_SCALE);
++        } else {
++            gyro[i] = mult_frac(r->gyro[i] - ins->cal_gyro.offset[i],
++                                936 * SWITCH_IMU_GYRO_RES_PER_DEG_S,
++                                ins->imu_cal_gyro_divisor[i]);
++        }
+     }
+ 
+-    // Right joycon has Y and Z axes negated.
++    // Right Joy-Con has native Y and Z axes negated.
+     if (ins->controller_type == SWITCH_CONTROLLER_TYPE_JCR) {
+         accel[1] = -accel[1];
+         accel[2] = -accel[2];
+@@ -843,10 +851,13 @@ static void parse_imu(uni_hid_device_t* d, const struct switch_imu_data_s* r) {
+         gyro[2] = -gyro[2];
+     }
+ 
+-    for (int i = 0; i < 3; i++) {
+-        ctl->gamepad.accel[i] = accel[i];
+-        ctl->gamepad.gyro[i] = gyro[i];
+-    }
++    // Match SDL3's PlayStation-oriented sensor coordinate convention.
++    ctl->gamepad.accel[0] = -accel[1];
++    ctl->gamepad.accel[1] = accel[2];
++    ctl->gamepad.accel[2] = -accel[0];
++    ctl->gamepad.gyro[0] = -gyro[1];
++    ctl->gamepad.gyro[1] = gyro[2];
++    ctl->gamepad.gyro[2] = -gyro[0];
+ }
+ 
+ // Process 0x30 input report: SWITCH_INPUT_IMU_DATA
diff --git a/switch-pico.cpp b/switch-pico.cpp
index 22beaef..22a3083 100644
--- a/switch-pico.cpp
+++ b/switch-pico.cpp
@@ -1,10 +1,14 @@
 #include <stdio.h>
 #include <string.h>
 #include "bsp/board.h"
-#include "hardware/uart.h"
 #include "pico/stdlib.h"
 #include "tusb.h"
 #include "switch_pro_driver.h"
+#ifndef SWITCH_PICO_BLUEPAD32
+#include "hardware/uart.h"
+#else
+#include "bluepad32_input_backend.h"
+#endif
 
 #ifdef SWITCH_PICO_LOG
 #define LOG_PRINTF(...) printf(__VA_ARGS__)
@@ -12,6 +16,7 @@
 #define LOG_PRINTF(...) ((void)0)
 #endif
 
+#ifndef SWITCH_PICO_BLUEPAD32
 // UART1 is reserved for external input frames from the host PC.
 #define UART_ID uart1
 #define BAUD_RATE 921600
@@ -19,6 +24,7 @@
 #define UART_RX_PIN 5
 #define UART_RUMBLE_HEADER 0xBB
 #define UART_RUMBLE_RUMBLE_TYPE 0x01
+#endif
 
 static bool g_last_mounted = false;
 static bool g_last_ready = false;
@@ -26,12 +32,14 @@ static bool g_last_ready = false;
 // Track the latest state provided by UART or the autopilot.
 static SwitchInputState g_user_state;
 
+#ifndef SWITCH_PICO_BLUEPAD32
 static void init_uart_input() {
     uart_init(UART_ID, BAUD_RATE);
     gpio_set_function(UART_TX_PIN, GPIO_FUNC_UART);
     gpio_set_function(UART_RX_PIN, GPIO_FUNC_UART);
     uart_set_format(UART_ID, 8, 1, UART_PARITY_NONE);
 }
+#endif
 
 static SwitchInputState neutral_input() {
     SwitchInputState state{};
@@ -42,6 +50,7 @@ static SwitchInputState neutral_input() {
     return state;
 }
 
+#ifndef SWITCH_PICO_BLUEPAD32
 static void send_rumble_uart_frame(const uint8_t rumble[8]) {
     uint8_t frame[11];
     frame[0] = UART_RUMBLE_HEADER;
@@ -55,11 +64,17 @@ static void send_rumble_uart_frame(const uint8_t rumble[8]) {
     frame[10] = checksum;
     uart_write_blocking(UART_ID, frame, sizeof(frame));
 }
+#endif
 
 static void on_rumble_from_switch(const uint8_t rumble[8]) {
+#ifdef SWITCH_PICO_BLUEPAD32
+    bluepad32_input_backend_queue_rumble(rumble);
+#else
     send_rumble_uart_frame(rumble);
+#endif
 }
 
+#ifndef SWITCH_PICO_BLUEPAD32
 // Consume UART bytes and forward complete frames to the Switch Pro driver.
 static bool poll_uart_frames() {
     static uint8_t buffer[64];
@@ -134,6 +149,7 @@ static bool poll_uart_frames() {
 
     return new_data;
 }
+#endif
 
 static void log_usb_state() {
     bool mounted = tud_mounted();
@@ -153,7 +169,11 @@ int main() {
     board_init();
     stdio_init_all();
 
+#ifdef SWITCH_PICO_BLUEPAD32
+    bluepad32_input_backend_init();
+#else
     init_uart_input();
+#endif
 
     tusb_init();
     switch_pro_init();
@@ -161,17 +181,32 @@ int main() {
     g_user_state = neutral_input();
     switch_pro_set_input(g_user_state);
 
+#ifdef SWITCH_PICO_BLUEPAD32
+    bluepad32_input_backend_start();
+    LOG_PRINTF("[BOOT] switch-pico starting (Bluepad32 wireless @ 115200)\n");
+#else
     LOG_PRINTF("[BOOT] switch-pico starting (UART0 log @ 115200)\n");
     LOG_PRINTF("[INFO] UART1 pins TX=%d RX=%d baud=%d\n",
            UART_TX_PIN, UART_RX_PIN, BAUD_RATE);
+#endif
 
     while (true) {
         tud_task();          // USB device tasks
+#ifdef SWITCH_PICO_BLUEPAD32
+        bluepad32_input_backend_snapshot(&g_user_state);
+#else
         bool new_data = poll_uart_frames();  // Pull controller state from UART1
         (void)new_data;
+#endif
         SwitchInputState state = g_user_state;
         switch_pro_set_input(state);
-        switch_pro_task();   // Push state to the Switch host
+#ifdef SWITCH_PICO_BLUEPAD32
+        if (switch_pro_task()) {
+            bluepad32_input_backend_report_sent();
+        }
+#else
+        (void)switch_pro_task();
+#endif
         log_usb_state();
     }
 }
diff --git a/switch_pro_driver.cpp b/switch_pro_driver.cpp
index 4d26922..6417c68 100644
--- a/switch_pro_driver.cpp
+++ b/switch_pro_driver.cpp
@@ -753,9 +753,10 @@ void switch_pro_set_input(const SwitchInputState& state) {
     g_input_state = state;
 }
 
-void switch_pro_task() {
+bool switch_pro_task() {
     uint32_t now = to_ms_since_boot(get_absolute_time());
     report_sent = false;
+    bool regular_report_sent = false;
 
     update_switch_report_from_state();
 
@@ -784,6 +785,7 @@ void switch_pro_task() {
                 memcpy(last_report, inputReport, report_size);
                 g_input_state.imu_sample_count = 0;
                 report_sent = true;
+                regular_report_sent = true;
             }
 
             last_report_timer = now;
@@ -799,6 +801,7 @@ void switch_pro_task() {
             last_report_timer = now;
         }
     }
+    return regular_report_sent;
 }
 
 bool switch_pro_apply_uart_packet(const uint8_t* packet, uint8_t length, SwitchInputState* out_state) {
diff --git a/switch_pro_driver.h b/switch_pro_driver.h
index 951d68b..6a4f8bb 100644
--- a/switch_pro_driver.h
+++ b/switch_pro_driver.h
@@ -55,8 +55,9 @@ void switch_pro_init();
 // Update the desired controller state for the next USB report.
 void switch_pro_set_input(const SwitchInputState& state);
 
-// Drive the Switch Pro USB state machine; call this frequently in the main loop.
-void switch_pro_task();
+// Drive the Switch Pro USB state machine; returns true only when a regular
+// 0x30 input report was successfully queued.
+bool switch_pro_task();
 
 // Convert a packed UART message into controller state (returns true if parsed).
 // If out_state is null the parsed state is written directly to the driver.
diff --git a/tests/test_prepare_bluepad32.py b/tests/test_prepare_bluepad32.py
new file mode 100644
index 0000000..3e4673f
--- /dev/null
+++ b/tests/test_prepare_bluepad32.py
@@ -0,0 +1,339 @@
+"""
+Tests for prepare_bluepad32.py patch preparation tool.
+
+Tests cover:
+- Fresh patch application
+- Idempotence (second invocation succeeds without changing content)
+- Missing paths validation
+- Diverged/ambiguous repository states
+"""
+
+import subprocess
+import tempfile
+import pytest
+from pathlib import Path
+from unittest.mock import patch as mock_patch
+
+import sys
+sys.path.insert(0, str(Path(__file__).parent.parent / "tools"))
+
+from prepare_bluepad32 import (
+    PatchError,
+    resolve_paths,
+    check_paths,
+    is_patch_applied,
+    apply_patch,
+    prepare_bluepad32,
+)
+
+
+@pytest.fixture
+def temp_repo_structure():
+    """Create a temporary directory structure with git repositories."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        root = Path(tmpdir)
+        
+        # Create bluepad32 repo
+        bp_dir = root / "external" / "bluepad32"
+        bp_dir.mkdir(parents=True)
+        subprocess.run(["git", "init"], cwd=bp_dir, check=True, capture_output=True)
+        subprocess.run(["git", "config", "user.email", "test@example.com"], cwd=bp_dir, check=True, capture_output=True)
+        subprocess.run(["git", "config", "user.name", "Test User"], cwd=bp_dir, check=True, capture_output=True)
+        
+        # Create a file to patch
+        test_file = bp_dir / "test.txt"
+        test_file.write_text("line 1\n")
+        subprocess.run(["git", "add", "test.txt"], cwd=bp_dir, check=True, capture_output=True)
+        subprocess.run(["git", "commit", "-m", "initial"], cwd=bp_dir, check=True, capture_output=True)
+        
+        # Create patches dir
+        patches_dir = root / "patches"
+        patches_dir.mkdir()
+        
+        yield root, bp_dir, patches_dir
+
+
+def create_simple_patch(repo_path: Path, patch_path: Path, file_to_patch: str = "test.txt") -> str:
+    """
+    Create a simple patch file that modifies a file in the repository.
+    
+    Returns the patch content as a string.
+    """
+    # Create the modification
+    test_file = repo_path / file_to_patch
+    original_content = test_file.read_text()
+    modified_content = original_content + "line 2\n"
+    
+    # Generate patch using git diff
+    test_file.write_text(modified_content)
+    result = subprocess.run(
+        ["git", "diff", file_to_patch],
+        cwd=repo_path,
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    patch_content = result.stdout
+    
+    # Reset the file to original state
+    test_file.write_text(original_content)
+    
+    # Write patch to file
+    patch_path.write_text(patch_content)
+    return patch_content
+
+
+def test_resolve_paths_with_defaults():
+    """Test that resolve_paths returns expected default paths."""
+    with tempfile.TemporaryDirectory() as tmpdir:
+        bp_path, patch_path = resolve_paths(Path(tmpdir))
+        assert bp_path == Path(tmpdir) / "external" / "bluepad32"
+        assert patch_path == Path(tmpdir) / "patches" / "bluepad32-sdl3-imu.patch"
+
+
+def test_resolve_paths_no_root():
+    """Test resolve_paths with no root uses current directory."""
+    bp_path, patch_path = resolve_paths()
+    assert bp_path.is_absolute()
+    assert patch_path.is_absolute()
+
+
+def test_check_paths_missing_bluepad32(temp_repo_structure):
+    """Test that check_paths fails if bluepad32 dir is missing."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    # Remove bluepad32
+    import shutil
+    shutil.rmtree(bp_dir)
+    
+    patch_file = patches_dir / "test.patch"
+    patch_file.write_text("dummy")
+    
+    with pytest.raises(PatchError, match="bluepad32 directory does not exist"):
+        check_paths(bp_dir, patch_file)
+
+
+def test_check_paths_missing_patch(temp_repo_structure):
+    """Test that check_paths fails if patch file is missing."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "nonexistent.patch"
+    
+    with pytest.raises(PatchError, match="patch file does not exist"):
+        check_paths(bp_dir, patch_file)
+
+
+def test_check_paths_bluepad32_not_git_repo(temp_repo_structure):
+    """Test that check_paths fails if bluepad32 is not a git repo."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    # Remove .git to make it not a git repo
+    import shutil
+    shutil.rmtree(bp_dir / ".git")
+    
+    patch_file = patches_dir / "test.patch"
+    patch_file.write_text("dummy")
+    
+    with pytest.raises(PatchError, match="not a git repository"):
+        check_paths(bp_dir, patch_file)
+
+
+def test_fresh_patch_application(temp_repo_structure):
+    """Test applying a fresh patch to a clean repository."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # Verify test.txt before patch
+    test_file = bp_dir / "test.txt"
+    original = test_file.read_text()
+    assert "line 2" not in original
+    
+    # Apply patch
+    apply_patch(bp_dir, patch_file)
+    
+    # Verify test.txt after patch
+    patched = test_file.read_text()
+    assert "line 2" in patched
+
+
+def test_idempotent_patch_application(temp_repo_structure):
+    """Test that applying the same patch twice succeeds (idempotence)."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # First application
+    apply_patch(bp_dir, patch_file)
+    test_file = bp_dir / "test.txt"
+    after_first = test_file.read_text()
+    
+    # Second application should succeed without changing content
+    apply_patch(bp_dir, patch_file)
+    after_second = test_file.read_text()
+    
+    assert after_first == after_second
+
+
+def test_is_patch_applied_not_applied(temp_repo_structure):
+    """Test is_patch_applied returns False for unapplied patch."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # Patch not applied yet
+    assert is_patch_applied(bp_dir, patch_file) is False
+
+
+def test_is_patch_applied_already_applied(temp_repo_structure):
+    """Test is_patch_applied returns True for already applied patch."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # Apply patch first
+    apply_patch(bp_dir, patch_file)
+    
+    # Now check should detect it's applied
+    assert is_patch_applied(bp_dir, patch_file) is True
+
+
+def test_diverged_repository_state(temp_repo_structure):
+    """Test that diverged repository (patch doesn't apply cleanly) is rejected."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # Diverge the repository by modifying the file such that the patch conflicts
+    test_file = bp_dir / "test.txt"
+    test_file.write_text("completely different line 1\n")
+    subprocess.run(["git", "add", "test.txt"], cwd=bp_dir, check=True, capture_output=True)
+    subprocess.run(["git", "commit", "-m", "divergence"], cwd=bp_dir, check=True, capture_output=True)
+    
+    # Try to apply patch - should fail because file content doesn't match
+    with pytest.raises(PatchError, match="Patch validation failed"):
+        apply_patch(bp_dir, patch_file)
+
+
+def test_missing_bluepad32_path(temp_repo_structure):
+    """Test prepare_bluepad32 fails gracefully with missing bluepad32."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    import shutil
+    shutil.rmtree(bp_dir)
+    
+    patch_file = patches_dir / "test.patch"
+    patch_file.write_text("dummy")
+    
+    with pytest.raises(PatchError, match="bluepad32 directory does not exist"):
+        prepare_bluepad32(bp_dir, patch_file)
+
+
+def test_missing_patch_file(temp_repo_structure):
+    """Test prepare_bluepad32 fails gracefully with missing patch file."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "nonexistent.patch"
+    
+    with pytest.raises(PatchError, match="patch file does not exist"):
+        prepare_bluepad32(bp_dir, patch_file)
+
+
+def test_prepare_bluepad32_full_workflow(temp_repo_structure):
+    """Test complete prepare_bluepad32 workflow: apply then idempotent re-apply."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    test_file = bp_dir / "test.txt"
+    original = test_file.read_text()
+    
+    # First prepare (should apply patch)
+    prepare_bluepad32(bp_dir, patch_file)
+    after_first = test_file.read_text()
+    assert after_first != original
+    assert "line 2" in after_first
+    
+    # Second prepare (should be idempotent)
+    prepare_bluepad32(bp_dir, patch_file)
+    after_second = test_file.read_text()
+    assert after_first == after_second
+
+
+def test_prepare_bluepad32_with_defaults(temp_repo_structure):
+    """Test prepare_bluepad32 uses correct defaults when paths not provided."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "bluepad32-sdl3-imu.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # Change to root directory and call with defaults
+    import os
+    original_cwd = os.getcwd()
+    try:
+        os.chdir(root)
+        prepare_bluepad32()  # Use defaults
+    finally:
+        os.chdir(original_cwd)
+    
+    # Verify patch was applied
+    test_file = bp_dir / "test.txt"
+    assert "line 2" in test_file.read_text()
+
+
+def test_patch_application_with_conflicting_content(temp_repo_structure):
+    """Test that patch with conflicting content is rejected."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    # Create a patch that adds a specific change
+    patch_content = """--- a/test.txt
++++ b/test.txt
+@@ -1 +1,3 @@
+ line 1
++line 2
++line 3
+"""
+    
+    patch_file = patches_dir / "conflict.patch"
+    patch_file.write_text(patch_content)
+    
+    # Modify the file to have different content that won't match the patch context
+    test_file = bp_dir / "test.txt"
+    test_file.write_text("modified line 1\n")
+    subprocess.run(["git", "add", "test.txt"], cwd=bp_dir, check=True, capture_output=True)
+    subprocess.run(["git", "commit", "-m", "modify"], cwd=bp_dir, check=True, capture_output=True)
+    
+    # Try to apply patch - should fail due to context mismatch
+    with pytest.raises(PatchError):
+        apply_patch(bp_dir, patch_file)
+
+
+def test_cli_with_explicit_paths(temp_repo_structure):
+    """Test CLI argument parsing with explicit paths."""
+    root, bp_dir, patches_dir = temp_repo_structure
+    
+    patch_file = patches_dir / "test.patch"
+    create_simple_patch(bp_dir, patch_file, "test.txt")
+    
+    # Simulate CLI call
+    sys.argv = [
+        "prepare_bluepad32.py",
+        "--bluepad32", str(bp_dir),
+        "--patch", str(patch_file),
+    ]
+    
+    from prepare_bluepad32 import main
+    
+    # Should not raise
+    try:
+        main()
+    except SystemExit as e:
+        # main() calls sys.exit on success, which we need to catch
+        if e.code != 0:
+            raise
diff --git a/tools/prepare_bluepad32.py b/tools/prepare_bluepad32.py
new file mode 100755
index 0000000..981d40f
--- /dev/null
+++ b/tools/prepare_bluepad32.py
@@ -0,0 +1,85 @@
+#!/usr/bin/env python3
+"""Apply the project Bluepad32 patch exactly once."""
+from __future__ import annotations
+
+import argparse
+import subprocess
+import sys
+from pathlib import Path
+
+
+class PatchError(RuntimeError):
+    pass
+
+
+def resolve_paths(repo_root: Path | None = None) -> tuple[Path, Path]:
+    root = Path.cwd() if repo_root is None else Path(repo_root)
+    return (
+        root / "external" / "bluepad32",
+        root / "patches" / "bluepad32-sdl3-imu.patch",
+    )
+
+
+def check_paths(bluepad32_path: Path, patch_path: Path) -> None:
+    if not bluepad32_path.is_dir():
+        raise PatchError(f"bluepad32 directory does not exist: {bluepad32_path}")
+    if not (bluepad32_path / ".git").exists():
+        raise PatchError(f"Bluepad32 is not a git repository: {bluepad32_path}")
+    if not patch_path.is_file():
+        raise PatchError(f"patch file does not exist: {patch_path}")
+
+
+def git_apply(bluepad32_path: Path, patch_path: Path, *args: str) -> subprocess.CompletedProcess[str]:
+    return subprocess.run(
+        ["git", "-C", str(bluepad32_path), "apply", *args, str(patch_path)],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+
+
+def is_patch_applied(bluepad32_path: Path, patch_path: Path) -> bool:
+    return git_apply(bluepad32_path, patch_path, "--reverse", "--check").returncode == 0
+
+
+def apply_patch(bluepad32_path: Path, patch_path: Path) -> None:
+    if is_patch_applied(bluepad32_path, patch_path):
+        return
+
+    check = git_apply(bluepad32_path, patch_path, "--check")
+    if check.returncode != 0:
+        detail = check.stderr.strip() or "patch does not apply"
+        raise PatchError(f"Patch validation failed (repository may be diverged):\n{detail}")
+
+    result = git_apply(bluepad32_path, patch_path)
+    if result.returncode != 0:
+        detail = result.stderr.strip() or "git apply failed"
+        raise PatchError(f"Could not patch Bluepad32: {detail}")
+
+
+def prepare_bluepad32(
+    bluepad32_path: Path | None = None,
+    patch_path: Path | None = None,
+) -> None:
+    default_bluepad32, default_patch = resolve_paths()
+    dependency = Path(bluepad32_path or default_bluepad32)
+    patch = Path(patch_path or default_patch)
+    check_paths(dependency, patch)
+    apply_patch(dependency, patch)
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--bluepad32", type=Path)
+    parser.add_argument("--patch", type=Path)
+    args = parser.parse_args()
+    try:
+        prepare_bluepad32(args.bluepad32, args.patch)
+    except PatchError as exc:
+        print(f"error: {exc}", file=sys.stderr)
+        return 1
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())