feat(app): implement Phase 5 local UDP sender->receiver loopback

Wire the first end-to-end pipeline: capture -> encode -> packetize ->
UDP -> depacketize -> decode -> render.

- UdpRtpTransport: raw POSIX UDP sockets (IPv4 via getaddrinfo), a
  receive jthread woken by socket close on stop; port 0 skips binding
  so the sender uses an OS-assigned source port. ASIO stays deferred
  to the signaling phase per ARCHITECTURE.md.
- SdlRenderer: SDL3 window/renderer with RGBA texture upload; the
  texture is recreated on resolution change. RendererFactory now
  returns RendererResult so SDL init failures carry a message,
  mirroring the codec/capture error patterns.
- screencast binary: parse_cli plus SenderPipeline/ReceiverPipeline
  per the app scaffolds; the sender creates its encoder once capture
  reports real dimensions, the receiver keeps a bounded 3-frame queue
  to hold latency down and renders on its own thread until the window
  closes. cli argv signature fixed to 'const char* const*' so main's
  argv converts implicitly.
- Encoder: drop AV_CODEC_FLAG_GLOBAL_HEADER so libx264 repeats SPS/PPS
  in-band at each keyframe -- the receiver decodes from the bitstream
  alone, which also makes mid-stream joins and later PLI recovery
  work without out-of-band parameter negotiation. The round-trip test
  now exercises exactly that path.
- tests: new udp-loopback integration test pushes synthetic frames
  through a real localhost socket and decodes 10/10 frames with the
  right dimensions; valgrind clean (loopback + codec). meson test 4/4.

Manual validation on the desktop (receiver window shows the captured
desktop) is documented in docs/RUNBOOK.md.

.agents/MEMORY.md

@@ -1,51 +1,38 @@
 # Project Memory — screen_cast
 
-Last updated: Phase 4 (RTP framing) complete and tested; current phase is
-Phase 5.
+Last updated: Phase 5 implemented and automated tests green; manual windowed
+loopback validation pending. See bottom for the run commands.
 
 ## Project state
 
-- Phase 4 done: `sc_network` library implements `RtpHeader`/`RtpPacket`
-  (RFC 3550 serialize/parse; tolerates CSRC lists, extension headers, and
-  padding on the receive side) plus `H264Packetizer`/`H264Depacketizer`
-  (RFC 6184 single-NAL + FU-A; STAP-A never emitted, unsupported types mark
-  the frame damaged on receive).
-- Depacketized access units use **3-byte start codes**, and the splitter
-  keeps a zero byte preceding a start code with the previous NAL, so both
-  3- and 4-byte-start-code streams round-trip byte-exactly (tested).
-- Loss handling is drop-on-damage: a sequence gap or missing FU fragment
-  marks the frame damaged and it is dropped silently at its marker. Full
-  loss recovery / jitter handling is Phase 7.
-- `test_rtp` covers header/packet round-trips, malformed rejections, NAL
-  splitting (3- and 4-byte codes), FU-A chunking with MTU bounds, full
-  packetize→depacketize round-trip, gap dropping, frame separation by
-  marker alone, sequence wrap, and random default SSRC/sequence.
-  Valgrind-clean; `meson test` 3/3.
-- Phase 3 remains validated; earlier review fixes still in place.
-- The first smoke run emitted `impl_ext_end_proxy called from wrong context`
-  warnings: `pw_context_connect_fd` and `pw_core_disconnect` ran outside the
-  thread-loop lock. Fixed by holding the lock across all pw setup/teardown
-  proxy operations. Re-running the smoke tool should now be warning-free
-  (capture worked both ways; the warnings only meant the first two
-  marshaled messages were rejected and retried from the right context).
-- `src/capture/pipewire_capture.cpp` now implements the full backend:
-  libportal 0.10 handshake (`create_screencast_session` → `session_start` →
-  `open_pipewire_remote`), PipeWire 1.6 stream on the first portal node,
-  BGRx/BGRA/RGBx/RGBA enumeration, latest-frame slot with condvar handoff.
-  Portal handshake blocks on the caller thread; frames arrive on the pw
-  thread. `stop()` is thread-safe; teardown follows the pw-required order.
-- Encoder now accepts padded strides for packed RGB formats and the new
-  `PixelFormat::Bgrx` (mapped to `AV_PIX_FMT_BGRA`); planar Yuv420p still
-  requires a packed layout. This was the review note blocking Phase 3.
-- `tools/capture_smoke` (manual, not in `meson test`) captures N frames →
-  encodes → writes Annex-B including prepended SPS/PPS extradata (verified:
-  libx264 GLOBAL_HEADER extradata is Annex-B).
-- Earlier review fixes remain in place; valgrind on the codec round-trip
-  test is still clean after the encoder stride changes.
-- Encoder PTS caveat: the encoder time_base is derived from the configured
-  frame rate (default 25fps), but portal frames arrive at monitor refresh
-  (often 60Hz), so pts values quantize to 40ms units and can repeat. Harmless
-  for the smoke test; revisit when RTP timestamps matter (Phase 4/5).
+- **Phase 5 (local UDP sender→receiver loopback) is implemented**:
+  - `UdpRtpTransport` (src/network/udp_transport.cpp): raw POSIX sockets,
+    AF_INET, IPv4 via getaddrinfo; port 0 skips binding (sender side); stop()
+    closes the socket to unblock the receive jthread. ASIO was deliberately
+    deferred to Phase 6 (see decisions).
+  - `screencast` binary (src/app/): cli.cpp + main.cpp + pipelines.cpp wiring
+    SenderPipeline (capture→encode→packetize→send) and ReceiverPipeline
+    (recv→depacketize→decode→bounded 3-frame queue→render thread).
+  - `SdlRenderer` (src/render/sdl_renderer.cpp): SDL3 window/renderer/texture,
+    RGBA texture upload, texture recreated on resolution change.
+    `RendererFactory::create` now returns `RendererResult` (error channel
+    added, mirroring codec/capture patterns).
+  - Encoder change: **GLOBAL_HEADER removed** so libx264 repeats SPS/PPS
+    in-band at every keyframe; a receiver now decodes from the bitstream
+    alone (mid-stream join, PLI recovery-ready). `get_extradata()` is empty
+    in this mode; codec round-trip test updated to match the streaming path.
+- **Automated validation**: `meson test` 4/4 — new `udp loopback` test
+  encodes synthetic frames, packetizes, sends over a real localhost UDP
+  socket, depacketizes, and decodes 10/10 frames with correct dimensions.
+  Valgrind clean (loopback + codec tests).
+- **Manual validation pending (needs the desktop)**: run the two commands in
+  `docs/RUNBOOK.md` — receiver window should show the captured desktop. Tick
+  `docs/PHASES.md` Phase 5 after this works.
+- Phase 4 network framing done (RFC 3550 + RFC 6184 single-NAL/FU-A;
+  3-byte canonical start codes; drop-on-damage loss handling).
+- Phase 3 capture done and validated (PipeWire/portal backend; the
+  `impl_ext_end_proxy` wrong-context warnings were fixed by holding the
+  thread-loop lock across all pw proxy operations).
 
 ## Decisions
 
@@ -55,13 +42,19 @@ Phase 5.
 - Capture: PipeWire + xdg-desktop-portal.
 - Encode/Decode: FFmpeg (libavcodec, libavutil, libswscale).
 - H.264 encoder path: software `libx264`, low-latency settings, Annex-B output.
-- Decoder receives SPS/PPS through `DecoderConfig.extradata`.
-- Transport: RTP over UDP; signaling via WebSocket/JSON.
+- **SPS/PPS are sent in-band ahead of every keyframe** (no GLOBAL_HEADER);
+  the decoder starts from the bitstream alone. `DecoderConfig.extradata`
+  remains available if signaling ever negotiates parameters out of band.
+- Transport: RTP over UDP via raw POSIX sockets for now; **ASIO stays a
+  Phase 6+ option** (the ARCHITECTURE.md dependency table places it with
+  signaling/discovery). IPv4 only at the transport level for now.
+- Rendering: **SDL3** (`sdl3` pkg-config, 3.4 installed); plain texture
+  upload, no GPU pipeline yet.
 - Discovery: mDNS/Avahi.
-- Rendering: SDL2 or SDL3 + OpenGL.
 - Namespace: `sc`.
 - Module error results use per-module `std::variant<T, XError>` types
-  (`CodecResult`, `CaptureResult`) since C++20 has no `std::expected`.
+  (`CodecResult`, `CaptureResult`, `RendererResult`) since C++20 has no
+  `std::expected`.
 
 ## Active blockers
 
@@ -72,20 +65,26 @@ None.
 - GUI framework (Qt6 vs. none / CLI only) — deferred to later phase.
 - Hardware acceleration strategy (VAAPI / Vulkan Video / NVENC) — evaluate after
   software encode path works.
+- IPv6 at the transport layer — revisit when LAN streaming lands (Phase 6+).
 
 ## Forward-looking review notes (for later phases)
 
-- `AV_CODEC_FLAG_GLOBAL_HEADER` suppresses in-band SPS/PPS; a receiver cannot
-  join mid-stream or recover after PLI without parameter sets. Phase 5 must
-  prepend SPS/PPS to keyframes or negotiate them in signaling.
 - Encoder sets no VBV (`maxrate`/`buffer_size`) — ABR only; add for smoother
   UDP streaming in Phase 7.
+- Encoder PTS caveat: time_base derives from the configured frame rate
+  (default 25fps) while portal frames arrive at monitor refresh (often 60Hz),
+  so pts values quantize and can repeat. RTP timestamps come from the capture
+  clock instead, so streaming is unaffected; revisit if decoder-side
+  presentation timing ever matters.
+- `RtpTransport::start/send` return plain bools (scaffold API); error
+  messages are lost — consider an error channel when signaling lands.
 - DMA-BUF-only portal streams are rejected with a clear message (hardware
   path is Phase 7).
 - No negative-path tests yet (bad config, bad stride, undersized buffer).
 - `to_annex_b_h264` sniffs AVCC vs Annex-B by content; if an AVCC-emitting
   encoder is ever added, prefer an explicit config flag over the heuristic.
 - Region targets are rejected: the desktop portal has no region capture.
-- `next_frame()` returns `nullopt` on stream error without surfacing the
-  reason (logged to stderr); consider an error channel when the receiver
-  pipeline lands.
+- `CaptureSession::next_frame()` returns `nullopt` on stream error without
+  surfacing the reason (logged to stderr).
+- Receiver ignores unknown packetization modes (STAP-A/MTAP/FU-B); senders
+  we control never emit them, but third-party interop would need support.

docs/RUNBOOK.md

@@ -49,6 +49,23 @@ PipeWire proxy operation ran without the thread-loop lock — all pw
 calls that send messages (connect, stream, core) must happen under
 `pw_thread_loop_lock`.
 
+## Sender / receiver loopback (Phase 5, manual)
+
+Two terminals on the same desktop session:
+
+```sh
+./build/src/app/screencast --receive        # terminal 1: window appears
+./build/src/app/screencast --send           # terminal 2: portal source picker
+```
+
+Expected: the receiver window shows the captured desktop in near real time.
+Stop either side with Ctrl-C. Optional flags: `--port`, `--peer HOST[:PORT]`,
+`--bitrate KBPS`, `--target window`.
+
+The headless equivalent runs as part of `meson test` (`udp loopback` test):
+synthetic frames → encode → packetize → localhost UDP → depacketize →
+decode, no portal or window involved.
+
 ## Formatting
 
 ```sh

include/screencast/app/cli.h

@@ -20,6 +20,8 @@ struct ReceiveCommand {
 using Command = std::variant<SendCommand, ReceiveCommand>;
 
 // Parse command line arguments. Prints usage and returns std::nullopt on error.
-std::optional<Command> parse_cli(int argc, const char* argv[]);
+// `argv` is `char const* const*` so both `main`'s `char**` and const arrays
+// convert implicitly.
+std::optional<Command> parse_cli(int argc, const char* const argv[]);
 
 } // namespace sc

include/screencast/app/pipeline.h

@@ -16,6 +16,9 @@ struct SenderPipelineConfig {
     EncoderConfig encoder;
     Endpoint local_rtp_endpoint;
     std::optional<Endpoint> signaling_server;
+    // Where encoded RTP packets are sent. Defaults to the local loopback so
+    // a sender and receiver on one machine work without any configuration.
+    Endpoint peer_rtp_endpoint{"127.0.0.1", 5004};
 };
 
 struct ReceiverPipelineConfig {

include/screencast/render/error.h

@@ -0,0 +1,36 @@
+#pragma once
+
+#include <string>
+#include <variant>
+
+namespace sc {
+
+struct RendererError {
+    std::string message;
+};
+
+// C++20 does not provide std::expected. Use a variant-based result type so
+// fallible renderer operations do not rely on exceptions.
+template <typename T> using RendererResult = std::variant<T, RendererError>;
+
+template <typename T> constexpr bool is_renderer_error(const RendererResult<T>& result) noexcept {
+    return std::holds_alternative<RendererError>(result);
+}
+
+template <typename T> T& renderer_value(RendererResult<T>& result) {
+    return std::get<T>(result);
+}
+
+template <typename T> const T& renderer_value(const RendererResult<T>& result) {
+    return std::get<T>(result);
+}
+
+template <typename T> RendererError& renderer_error(RendererResult<T>& result) {
+    return std::get<RendererError>(result);
+}
+
+template <typename T> const RendererError& renderer_error(const RendererResult<T>& result) {
+    return std::get<RendererError>(result);
+}
+
+} // namespace sc

include/screencast/render/renderer.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include "screencast/codec/decoder.h"
+#include "screencast/render/error.h"
 
 #include <memory>
 #include <string>
@@ -29,7 +30,7 @@ class Renderer {
 
 class RendererFactory {
   public:
-    static std::unique_ptr<Renderer> create(const RendererConfig& config);
+    static RendererResult<std::unique_ptr<Renderer>> create(const RendererConfig& config);
 };
 
 } // namespace sc

src/app/cli.cpp

@@ -0,0 +1,109 @@
+#include "screencast/app/cli.h"
+
+#include <charconv>
+#include <cstdio>
+#include <string_view>
+
+namespace sc {
+namespace {
+
+void print_usage() {
+    std::fputs("usage: screencast --send [--target monitor|window] [--peer HOST[:PORT]] [--bitrate KBPS]\n"
+               "       screencast --receive [--port PORT]\n",
+               stderr);
+}
+
+bool parse_int(std::string_view text, int& value) {
+    const auto [pointer, error] = std::from_chars(text.data(), text.data() + text.size(), value);
+    return error == std::errc{} && pointer == text.data() + text.size();
+}
+
+bool next_argument(int argc, const char* const argv[], int& index, std::string_view& value) {
+    if (index + 1 >= argc) {
+        print_usage();
+        return false;
+    }
+    value = argv[++index];
+    return true;
+}
+
+} // namespace
+
+std::optional<Command> parse_cli(int argc, const char* const argv[]) {
+    enum class Mode {
+        None,
+        Send,
+        Receive,
+    };
+
+    Mode mode = Mode::None;
+    SendCommand send;
+    ReceiveCommand receive;
+
+    for (int index = 1; index < argc; ++index) {
+        const std::string_view argument = argv[index];
+
+        if (argument == "--send") {
+            if (mode != Mode::None) {
+                print_usage();
+                return std::nullopt;
+            }
+            mode = Mode::Send;
+        } else if (argument == "--receive") {
+            if (mode != Mode::None) {
+                print_usage();
+                return std::nullopt;
+            }
+            mode = Mode::Receive;
+        } else if (argument == "--target") {
+            std::string_view value;
+            if (!next_argument(argc, argv, index, value)) {
+                return std::nullopt;
+            }
+            if (value != "monitor" && value != "window") {
+                print_usage();
+                return std::nullopt;
+            }
+            send.target = value;
+        } else if (argument == "--peer") {
+            std::string_view value;
+            if (!next_argument(argc, argv, index, value)) {
+                return std::nullopt;
+            }
+            if (mode == Mode::Receive) {
+                receive.peer_address = value;
+            } else {
+                send.peer_address = value;
+            }
+        } else if (argument == "--bitrate") {
+            std::string_view value;
+            if (!next_argument(argc, argv, index, value) || !parse_int(value, send.bitrate_kbps) ||
+                send.bitrate_kbps <= 0) {
+                print_usage();
+                return std::nullopt;
+            }
+        } else if (argument == "--port") {
+            std::string_view value;
+            int port = 0;
+            if (!next_argument(argc, argv, index, value) || !parse_int(value, port) || port <= 0 || port > 65535) {
+                print_usage();
+                return std::nullopt;
+            }
+            receive.local_rtp_port = port;
+        } else {
+            print_usage();
+            return std::nullopt;
+        }
+    }
+
+    if (mode == Mode::Send) {
+        return Command{std::move(send)};
+    }
+    if (mode == Mode::Receive) {
+        return Command{std::move(receive)};
+    }
+    print_usage();
+    return std::nullopt;
+}
+
+} // namespace sc

src/app/main.cpp

@@ -0,0 +1,87 @@
+#include "screencast/app/cli.h"
+#include "screencast/app/pipeline.h"
+
+#include <atomic>
+#include <charconv>
+#include <chrono>
+#include <csignal>
+#include <string_view>
+#include <thread>
+
+namespace {
+
+std::atomic<bool> g_interrupted{false};
+
+void handle_interrupt(int) {
+    g_interrupted.store(true);
+}
+
+sc::Endpoint parse_endpoint(std::string_view address, uint16_t default_port) {
+    sc::Endpoint endpoint;
+    const std::size_t separator = address.rfind(':');
+    if (separator != std::string_view::npos) {
+        endpoint.address = std::string{address.substr(0, separator)};
+        int port = 0;
+        const auto [pointer, error] =
+            std::from_chars(address.data() + separator + 1, address.data() + address.size(), port);
+        endpoint.port = error == std::errc{} ? static_cast<uint16_t>(port) : default_port;
+    } else {
+        endpoint.address = std::string{address};
+        endpoint.port = default_port;
+    }
+    return endpoint;
+}
+
+int run_sender(const sc::SendCommand& command) {
+    sc::SenderPipelineConfig config;
+    if (command.target == "window") {
+        config.capture_target = sc::CaptureTargetWindow{};
+    } else {
+        config.capture_target = sc::CaptureTargetWholeScreen{};
+    }
+    config.peer_rtp_endpoint =
+        command.peer_address.empty() ? sc::Endpoint{"127.0.0.1", 5004} : parse_endpoint(command.peer_address, 5004);
+    config.encoder.bitrate_kbps = command.bitrate_kbps;
+
+    sc::SenderPipeline pipeline{std::move(config)};
+    if (!pipeline.start()) {
+        return 1;
+    }
+    while (!g_interrupted.load()) {
+        std::this_thread::sleep_for(std::chrono::milliseconds(100));
+    }
+    pipeline.stop();
+    return 0;
+}
+
+int run_receiver(const sc::ReceiveCommand& command) {
+    sc::ReceiverPipelineConfig config;
+    config.local_rtp_endpoint = sc::Endpoint{"0.0.0.0", static_cast<uint16_t>(command.local_rtp_port)};
+
+    sc::ReceiverPipeline pipeline{std::move(config)};
+    if (!pipeline.start()) {
+        return 1;
+    }
+    while (!g_interrupted.load()) {
+        std::this_thread::sleep_for(std::chrono::milliseconds(100));
+    }
+    pipeline.stop();
+    return 0;
+}
+
+} // namespace
+
+int main(int argc, char* argv[]) {
+    const std::optional<sc::Command> command = sc::parse_cli(argc, argv);
+    if (!command.has_value()) {
+        return 1;
+    }
+
+    std::signal(SIGINT, handle_interrupt);
+    std::signal(SIGTERM, handle_interrupt);
+
+    if (const sc::SendCommand* send = std::get_if<sc::SendCommand>(&*command)) {
+        return run_sender(*send);
+    }
+    return run_receiver(std::get<sc::ReceiveCommand>(*command));
+}

src/app/meson.build

@@ -0,0 +1,9 @@
+# The screencast application binary wiring every module together.
+
+executable('screencast',
+    [
+        'cli.cpp',
+        'main.cpp',
+        'pipelines.cpp',
+    ],
+    dependencies : [sc_capture_dep, sc_codec_dep, sc_network_dep, sc_render_dep])

src/app/pipelines.cpp

@@ -0,0 +1,224 @@
+#include "screencast/app/pipeline.h"
+
+#include "screencast/network/h264_packetizer.h"
+
+#include <chrono>
+#include <deque>
+#include <iostream>
+#include <memory>
+#include <mutex>
+#include <thread>
+
+namespace sc {
+
+class SenderPipeline::Impl {
+  public:
+    explicit Impl(SenderPipelineConfig config) : config_(std::move(config)) {}
+
+    ~Impl() {
+        stop();
+    }
+
+    bool start() {
+        auto capture_result = CaptureFactory::create(config_.capture_target);
+        if (is_capture_error(capture_result)) {
+            std::cerr << std::format("screencast: capture failed: {}\n", capture_error(capture_result).message);
+            return false;
+        }
+        capture_ = std::move(capture_value(capture_result));
+
+        if (!transport_->start(config_.local_rtp_endpoint, [](RtpPacket) {})) {
+            std::cerr << "screencast: failed to start the RTP transport\n";
+            return false;
+        }
+        transport_->set_peer(config_.peer_rtp_endpoint);
+
+        run_thread_ = std::jthread([this](std::stop_token stop_token) { run(std::move(stop_token)); });
+        return true;
+    }
+
+    void stop() {
+        if (capture_ != nullptr) {
+            capture_->stop();
+        }
+        run_thread_ = std::jthread{};
+        transport_->stop();
+    }
+
+  private:
+    void run(std::stop_token stop_token) {
+        while (!stop_token.stop_requested()) {
+            const std::optional<CapturedFrame> frame = capture_->next_frame();
+            if (!frame.has_value()) {
+                break;
+            }
+
+            if (encoder_ == nullptr && !create_encoder(*frame)) {
+                break;
+            }
+
+            auto encoded_result = encoder_->encode(*frame);
+            if (is_codec_error(encoded_result)) {
+                std::cerr << std::format("screencast: encode failed: {}\n", codec_error(encoded_result).message);
+                break;
+            }
+            for (const EncodedFrame& encoded : codec_value(encoded_result)) {
+                for (RtpPacket& packet : packetizer_.packetize(encoded)) {
+                    (void)transport_->send(packet);
+                }
+            }
+        }
+    }
+
+    bool create_encoder(const CapturedFrame& frame) {
+        EncoderConfig config = config_.encoder;
+        config.width = frame.width;
+        config.height = frame.height;
+
+        auto encoder_result = EncoderFactory::create(config);
+        if (is_codec_error(encoder_result)) {
+            std::cerr << std::format("screencast: encoder creation failed: {}\n", codec_error(encoder_result).message);
+            return false;
+        }
+        encoder_ = std::move(codec_value(encoder_result));
+        return true;
+    }
+
+    SenderPipelineConfig config_;
+    std::unique_ptr<CaptureSession> capture_;
+    std::unique_ptr<Encoder> encoder_;
+    H264Packetizer packetizer_;
+    std::unique_ptr<RtpTransport> transport_ = RtpTransportFactory::create();
+    std::jthread run_thread_;
+};
+
+SenderPipeline::SenderPipeline(SenderPipelineConfig config) : impl_(std::make_unique<Impl>(std::move(config))) {}
+
+SenderPipeline::~SenderPipeline() = default;
+
+bool SenderPipeline::start() {
+    return impl_->start();
+}
+
+void SenderPipeline::stop() {
+    impl_->stop();
+}
+
+class ReceiverPipeline::Impl {
+  public:
+    explicit Impl(ReceiverPipelineConfig config) : config_(std::move(config)) {}
+
+    ~Impl() {
+        stop();
+    }
+
+    bool start() {
+        auto renderer_result = RendererFactory::create(config_.renderer);
+        if (is_renderer_error(renderer_result)) {
+            std::cerr << std::format("screencast: renderer creation failed: {}\n",
+                                     renderer_error(renderer_result).message);
+            return false;
+        }
+        renderer_ = std::move(renderer_value(renderer_result));
+
+        auto decoder_result = DecoderFactory::create(config_.decoder);
+        if (is_codec_error(decoder_result)) {
+            std::cerr << std::format("screencast: decoder creation failed: {}\n", codec_error(decoder_result).message);
+            return false;
+        }
+        decoder_ = std::move(codec_value(decoder_result));
+
+        if (!transport_->start(config_.local_rtp_endpoint,
+                               [this](RtpPacket packet) { on_packet(std::move(packet)); })) {
+            std::cerr << "screencast: failed to start the RTP transport\n";
+            return false;
+        }
+
+        render_thread_ = std::jthread([this](std::stop_token stop_token) { render_loop(std::move(stop_token)); });
+        return true;
+    }
+
+    void stop() {
+        render_thread_ = std::jthread{};
+        transport_->stop();
+        renderer_ = nullptr;
+        decoder_ = nullptr;
+        {
+            std::lock_guard lock(queue_mutex_);
+            queue_.clear();
+        }
+    }
+
+  private:
+    void on_packet(RtpPacket packet) {
+        std::optional<std::vector<std::byte>> access_unit = depacketizer_.depacketize(packet);
+        if (!access_unit.has_value()) {
+            return;
+        }
+
+        EncodedFrame encoded;
+        encoded.data = std::move(*access_unit);
+        encoded.rtp_timestamp = packet.header.timestamp;
+        encoded.is_keyframe = false;
+
+        auto decoded_result = decoder_->decode(encoded);
+        if (is_codec_error(decoded_result)) {
+            std::cerr << std::format("screencast: decode failed: {}\n", codec_error(decoded_result).message);
+            return;
+        }
+        for (DecodedFrame& decoded : codec_value(decoded_result)) {
+            std::lock_guard lock(queue_mutex_);
+            // Keep latency low: drop the oldest frame when the queue is full.
+            if (queue_.size() >= kMaxQueuedFrames) {
+                queue_.pop_front();
+            }
+            queue_.push_back(std::move(decoded));
+        }
+    }
+
+    void render_loop(std::stop_token stop_token) {
+        while (!stop_token.stop_requested()) {
+            if (!renderer_->poll_events()) {
+                break; // window closed
+            }
+            std::optional<DecodedFrame> frame;
+            {
+                std::unique_lock lock(queue_mutex_);
+                if (!queue_.empty()) {
+                    frame = std::move(queue_.front());
+                    queue_.pop_front();
+                }
+            }
+            if (frame.has_value()) {
+                (void)renderer_->present(*frame);
+            } else {
+                std::this_thread::sleep_for(std::chrono::milliseconds(5));
+            }
+        }
+    }
+
+    static constexpr std::size_t kMaxQueuedFrames = 3;
+
+    ReceiverPipelineConfig config_;
+    std::unique_ptr<Renderer> renderer_;
+    std::unique_ptr<Decoder> decoder_;
+    H264Depacketizer depacketizer_;
+    std::unique_ptr<RtpTransport> transport_ = RtpTransportFactory::create();
+    std::jthread render_thread_;
+    std::mutex queue_mutex_;
+    std::deque<DecodedFrame> queue_;
+};
+
+ReceiverPipeline::ReceiverPipeline(ReceiverPipelineConfig config) : impl_(std::make_unique<Impl>(std::move(config))) {}
+
+ReceiverPipeline::~ReceiverPipeline() = default;
+
+bool ReceiverPipeline::start() {
+    return impl_->start();
+}
+
+void ReceiverPipeline::stop() {
+    impl_->stop();
+}
+
+} // namespace sc

src/codec/ffmpeg_encoder.cpp

@@ -397,7 +397,10 @@ CodecResult<std::unique_ptr<Encoder>> EncoderFactory::create(const EncoderConfig
     ctx->max_b_frames = 0;
     ctx->thread_count = 1;
     ctx->profile = AV_PROFILE_H264_MAIN;
-    ctx->flags |= AV_CODEC_FLAG_LOW_DELAY | AV_CODEC_FLAG_GLOBAL_HEADER;
+    // No GLOBAL_HEADER: SPS/PPS are repeated in-band at every keyframe so a
+    // receiver that joins mid-stream (or recovers after loss) can decode
+    // without out-of-band parameter negotiation.
+    ctx->flags |= AV_CODEC_FLAG_LOW_DELAY;
 
     if (av_opt_set(ctx->priv_data, "preset", "ultrafast", 0) < 0) {
         return CodecError{"failed to set libx264 preset"};

src/meson.build

@@ -24,3 +24,6 @@ sc_codec_dep = declare_dependency(
     dependencies : [dep_avcodec, dep_avutil, dep_swscale])
 
 subdir('network')
+subdir('render')
+
+subdir('app')

src/network/meson.build

@@ -3,6 +3,7 @@
 sc_network_sources = files(
     'rtp_packet.cpp',
     'h264_packetizer.cpp',
+    'udp_transport.cpp',
 )
 
 sc_network = static_library('sc_network',

src/network/udp_transport.cpp

@@ -0,0 +1,163 @@
+#include "screencast/network/transport.h"
+
+#include <arpa/inet.h>
+#include <netdb.h>
+#include <netinet/in.h>
+#include <sys/socket.h>
+#include <unistd.h>
+
+#include <array>
+#include <atomic>
+#include <cstring>
+#include <mutex>
+#include <optional>
+#include <string>
+#include <thread>
+
+namespace sc {
+namespace {
+
+constexpr int kInvalidSocket = -1;
+
+// A UDP datagram cannot exceed 64 KiB on IPv4; one buffer fits any RTP
+// packet the receiver will ever see.
+constexpr std::size_t kReceiveBufferSize = 65536;
+
+std::optional<sockaddr_in> resolve_ipv4(const Endpoint& endpoint) {
+    addrinfo hints{};
+    hints.ai_family = AF_INET;
+    hints.ai_socktype = SOCK_DGRAM;
+
+    addrinfo* result = nullptr;
+    const std::string port = std::to_string(endpoint.port);
+    const char* node = endpoint.address.empty() ? nullptr : endpoint.address.c_str();
+    if (getaddrinfo(node, port.c_str(), &hints, &result) != 0) {
+        return std::nullopt;
+    }
+
+    std::optional<sockaddr_in> address;
+    if (result != nullptr && result->ai_family == AF_INET && result->ai_addrlen >= sizeof(sockaddr_in)) {
+        sockaddr_in resolved{};
+        std::memcpy(&resolved, result->ai_addr, sizeof(sockaddr_in));
+        address = resolved;
+    }
+    freeaddrinfo(result);
+    return address;
+}
+
+} // namespace
+
+class UdpRtpTransport final : public RtpTransport {
+  public:
+    UdpRtpTransport() = default;
+
+    ~UdpRtpTransport() override {
+        stop();
+    }
+
+    UdpRtpTransport(const UdpRtpTransport&) = delete;
+    UdpRtpTransport& operator=(const UdpRtpTransport&) = delete;
+
+    bool start(const Endpoint& local_endpoint, ReceiveCallback on_receive) override {
+        if (running_.load()) {
+            return false;
+        }
+        socket_ = ::socket(AF_INET, SOCK_DGRAM, 0);
+        if (socket_ < 0) {
+            return false;
+        }
+        int reuse = 1;
+        (void)::setsockopt(socket_, SOL_SOCKET, SO_REUSEADDR, &reuse, sizeof(reuse));
+
+        // A zero port skips binding: the OS picks the source port on send.
+        if (local_endpoint.port != 0) {
+            const std::optional<sockaddr_in> address = resolve_ipv4(local_endpoint);
+            if (!address.has_value()) {
+                (void)::close(socket_);
+                socket_ = kInvalidSocket;
+                return false;
+            }
+            if (::bind(socket_, reinterpret_cast<const sockaddr*>(&*address), sizeof(*address)) < 0) {
+                (void)::close(socket_);
+                socket_ = kInvalidSocket;
+                return false;
+            }
+        }
+
+        running_.store(true);
+        receive_thread_ = std::jthread([this, callback = std::move(on_receive)]() mutable { receive_loop(callback); });
+        return true;
+    }
+
+    bool send(const RtpPacket& packet) override {
+        if (!running_.load() || !has_peer_.load()) {
+            return false;
+        }
+        const std::vector<std::byte> bytes = packet.serialize();
+        if (bytes.empty()) {
+            return false;
+        }
+        sockaddr_in peer{};
+        {
+            // Snapshot the peer so set_peer() can be called concurrently.
+            std::lock_guard lock(peer_mutex_);
+            peer = peer_;
+        }
+        const ssize_t sent =
+            ::sendto(socket_, bytes.data(), bytes.size(), 0, reinterpret_cast<const sockaddr*>(&peer), sizeof(peer));
+        return sent == static_cast<ssize_t>(bytes.size());
+    }
+
+    void set_peer(const Endpoint& peer) override {
+        const std::optional<sockaddr_in> address = resolve_ipv4(peer);
+        if (!address.has_value()) {
+            return;
+        }
+        std::lock_guard lock(peer_mutex_);
+        peer_ = *address;
+        has_peer_.store(true);
+    }
+
+    void stop() override {
+        if (!running_.exchange(false)) {
+            return;
+        }
+        if (socket_ >= 0) {
+            (void)::shutdown(socket_, SHUT_RDWR);
+            (void)::close(socket_);
+            socket_ = kInvalidSocket;
+        }
+        // Closing the socket unblocks recvfrom; joining happens implicitly
+        // when the jthread assignment destroys the previous thread.
+        receive_thread_ = std::jthread{};
+    }
+
+  private:
+    void receive_loop(ReceiveCallback& callback) {
+        std::array<std::byte, kReceiveBufferSize> buffer{};
+        while (running_.load()) {
+            const ssize_t received = ::recvfrom(socket_, buffer.data(), buffer.size(), 0, nullptr, nullptr);
+            if (received <= 0) {
+                continue; // closed socket while running_ still true, or error
+            }
+            const std::optional<RtpPacket> packet =
+                RtpPacket::parse(std::span{buffer.data(), static_cast<std::size_t>(received)});
+            if (packet.has_value()) {
+                callback(*packet);
+            }
+        }
+    }
+
+    int socket_ = kInvalidSocket;
+    std::atomic<bool> running_{false};
+    std::atomic<bool> has_peer_{false};
+    std::mutex peer_mutex_;
+    sockaddr_in peer_{};
+    std::jthread receive_thread_;
+};
+
+std::unique_ptr<RtpTransport> RtpTransportFactory::create() {
+    return std::make_unique<UdpRtpTransport>();
+}
+
+} // namespace sc

src/render/meson.build

@@ -0,0 +1,15 @@
+# Phase 5 renderer: SDL3 software texture upload, no GPU pipeline yet.
+
+dep_sdl3 = dependency('sdl3')
+
+sc_render_sources = files('sdl_renderer.cpp')
+
+sc_render = static_library('sc_render',
+    sc_render_sources,
+    include_directories : sc_core_inc,
+    dependencies : [dep_sdl3])
+
+sc_render_dep = declare_dependency(
+    link_with : sc_render,
+    include_directories : sc_core_inc,
+    dependencies : [dep_sdl3])

src/render/sdl_renderer.cpp

@@ -0,0 +1,147 @@
+#include "screencast/render/renderer.h"
+
+#include <SDL3/SDL.h>
+
+#include <string>
+
+namespace sc {
+namespace {
+
+// DecodedFrame pixels are AV_PIX_FMT_RGBA: memory order R, G, B, A, which
+// matches SDL_PIXELFORMAT_RGBA8888.
+constexpr SDL_PixelFormat kSdlPixelFormat = SDL_PIXELFORMAT_RGBA8888;
+
+class SdlRenderer final : public Renderer {
+  public:
+    explicit SdlRenderer(const RendererConfig& config) : config_(config) {}
+
+    ~SdlRenderer() override {
+        shutdown();
+    }
+
+    SdlRenderer(const SdlRenderer&) = delete;
+    SdlRenderer& operator=(const SdlRenderer&) = delete;
+
+    const std::string& last_error() const {
+        return last_error_;
+    }
+
+    bool initialize() {
+        if (!SDL_Init(SDL_INIT_VIDEO)) {
+            last_error_ = std::string{"SDL_Init failed: "} + SDL_GetError();
+            return false;
+        }
+        sdl_inited_ = true;
+
+        window_ = SDL_CreateWindow(config_.window_title.c_str(), config_.initial_width, config_.initial_height, 0);
+        if (window_ == nullptr) {
+            last_error_ = std::string{"SDL_CreateWindow failed: "} + SDL_GetError();
+            shutdown();
+            return false;
+        }
+
+        renderer_ = SDL_CreateRenderer(window_, nullptr);
+        if (renderer_ == nullptr) {
+            last_error_ = std::string{"SDL_CreateRenderer failed: "} + SDL_GetError();
+            shutdown();
+            return false;
+        }
+
+        return true;
+    }
+
+    bool present(const DecodedFrame& frame) override {
+        if (frame.width <= 0 || frame.height <= 0) {
+            return false;
+        }
+        if (frame.width != texture_width_ || frame.height != texture_height_) {
+            if (!recreate_texture(frame.width, frame.height)) {
+                return false;
+            }
+        }
+
+        const int pitch = frame.width * 4;
+        if (!SDL_UpdateTexture(texture_, nullptr, frame.rgba_pixels.data(), pitch)) {
+            return false;
+        }
+        if (!SDL_RenderTexture(renderer_, texture_, nullptr, nullptr)) {
+            return false;
+        }
+        SDL_RenderPresent(renderer_);
+        return true;
+    }
+
+    bool poll_events() override {
+        if (closed_) {
+            return false;
+        }
+        SDL_Event event{};
+        while (SDL_PollEvent(&event)) {
+            if (event.type == SDL_EVENT_QUIT ||
+                (event.type == SDL_EVENT_WINDOW_CLOSE_REQUESTED && event.window.windowID == SDL_GetWindowID(window_))) {
+                closed_ = true;
+                return false;
+            }
+        }
+        return true;
+    }
+
+    void shutdown() override {
+        if (texture_ != nullptr) {
+            SDL_DestroyTexture(texture_);
+            texture_ = nullptr;
+        }
+        if (renderer_ != nullptr) {
+            SDL_DestroyRenderer(renderer_);
+            renderer_ = nullptr;
+        }
+        if (window_ != nullptr) {
+            SDL_DestroyWindow(window_);
+            window_ = nullptr;
+        }
+        if (sdl_inited_) {
+            SDL_Quit();
+            sdl_inited_ = false;
+        }
+    }
+
+  private:
+    bool recreate_texture(int width, int height) {
+        if (texture_ != nullptr) {
+            SDL_DestroyTexture(texture_);
+            texture_ = nullptr;
+        }
+        texture_ = SDL_CreateTexture(renderer_, kSdlPixelFormat, SDL_TEXTUREACCESS_STREAMING, width, height);
+        if (texture_ == nullptr) {
+            last_error_ = std::string{"SDL_CreateTexture failed: "} + SDL_GetError();
+            texture_width_ = 0;
+            texture_height_ = 0;
+            return false;
+        }
+        texture_width_ = width;
+        texture_height_ = height;
+        return true;
+    }
+
+    RendererConfig config_;
+    std::string last_error_;
+    bool sdl_inited_ = false;
+    bool closed_ = false;
+    SDL_Window* window_ = nullptr;
+    SDL_Renderer* renderer_ = nullptr;
+    SDL_Texture* texture_ = nullptr;
+    int texture_width_ = 0;
+    int texture_height_ = 0;
+};
+
+} // namespace
+
+RendererResult<std::unique_ptr<Renderer>> RendererFactory::create(const RendererConfig& config) {
+    auto renderer = std::make_unique<SdlRenderer>(config);
+    if (!renderer->initialize()) {
+        return RendererError{renderer->last_error()};
+    }
+    return std::unique_ptr<Renderer>(std::move(renderer));
+}
+
+} // namespace sc

tests/app/test_loopback.cpp

@@ -0,0 +1,170 @@
+// Phase 5 integration test: the local UDP sender -> receiver loopback without
+// the interactive parts. Synthetic frames go through the real chain —
+// encode, RTP packetization, a localhost UDP socket, depacketization, and
+// decode (with in-band SPS/PPS, since there is no signaling channel).
+
+#include "screencast/codec/decoder.h"
+#include "screencast/codec/encoder.h"
+#include "screencast/network/h264_packetizer.h"
+#include "screencast/network/transport.h"
+
+#include <atomic>
+#include <chrono>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <memory>
+#include <thread>
+#include <vector>
+
+namespace {
+
+[[noreturn]] void fail(const char* what) {
+    std::fprintf(stderr, "test_loopback: FAIL: %s\n", what);
+    std::abort();
+}
+
+void check(bool condition, const char* what) {
+    if (!condition) {
+        fail(what);
+    }
+}
+
+template <typename T> void expect_result(const char* what, const sc::CodecResult<T>& result) {
+    if (sc::is_codec_error(result)) {
+        std::fprintf(stderr, "test_loopback: FAIL: %s: %s\n", what, sc::codec_error(result).message.c_str());
+        std::abort();
+    }
+}
+
+constexpr int kWidth = 64;
+constexpr int kHeight = 64;
+constexpr int kFrameCount = 10;
+constexpr uint64_t kNsPerFrame = 40'000'000; // 25 fps
+
+sc::CapturedFrame make_frame(uint32_t index) {
+    sc::CapturedFrame frame;
+    frame.width = kWidth;
+    frame.height = kHeight;
+    frame.timestamp_ns = static_cast<uint64_t>(index) * kNsPerFrame;
+    frame.pixel_format = sc::PixelFormat::Rgba;
+    frame.stride = kWidth * 4;
+    frame.pixels.resize(static_cast<std::size_t>(kWidth) * kHeight * 4);
+
+    auto* pixels = reinterpret_cast<std::uint8_t*>(frame.pixels.data());
+    for (int y = 0; y < kHeight; ++y) {
+        for (int x = 0; x < kWidth; ++x) {
+            const std::size_t offset = (static_cast<std::size_t>(y) * kWidth + x) * 4;
+            pixels[offset + 0] = static_cast<std::uint8_t>((x + static_cast<int>(index) * 4) & 0xFF);
+            pixels[offset + 1] = static_cast<std::uint8_t>((y + static_cast<int>(index) * 2) & 0xFF);
+            pixels[offset + 2] = static_cast<std::uint8_t>(((x ^ y) + static_cast<int>(index)) & 0xFF);
+            pixels[offset + 3] = 0xFF;
+        }
+    }
+    return frame;
+}
+
+// Receiver state driven entirely by the transport's receive thread.
+struct ReceiverSink {
+    sc::H264Depacketizer depacketizer;
+    std::unique_ptr<sc::Decoder> decoder;
+    std::atomic<int> decoded_frames{0};
+    bool saw_frame = false;
+    int last_width = 0;
+    int last_height = 0;
+
+    void on_packet(sc::RtpPacket packet) {
+        std::optional<std::vector<std::byte>> access_unit = depacketizer.depacketize(packet);
+        if (!access_unit.has_value()) {
+            return;
+        }
+
+        sc::EncodedFrame encoded;
+        encoded.data = std::move(*access_unit);
+        encoded.rtp_timestamp = packet.header.timestamp;
+
+        auto decoded_result = decoder->decode(encoded);
+        if (sc::is_codec_error(decoded_result)) {
+            return;
+        }
+        for (const sc::DecodedFrame& decoded : sc::codec_value(decoded_result)) {
+            last_width = decoded.width;
+            last_height = decoded.height;
+            saw_frame = decoded.width > 0 && !decoded.rgba_pixels.empty();
+            decoded_frames.fetch_add(1);
+        }
+    }
+};
+
+} // namespace
+
+int main() {
+    // Encoder on the sender side.
+    sc::EncoderConfig encoder_config;
+    encoder_config.codec_name = "h264";
+    encoder_config.width = kWidth;
+    encoder_config.height = kHeight;
+    encoder_config.frame_rate_num = 25;
+    encoder_config.frame_rate_den = 1;
+    encoder_config.bitrate_kbps = 2000;
+
+    auto encoder_result = sc::EncoderFactory::create(encoder_config);
+    expect_result("encoder create", encoder_result);
+    auto encoder = std::move(sc::codec_value(encoder_result));
+
+    // The receiver decodes purely from the bitstream (in-band SPS/PPS).
+    sc::DecoderConfig decoder_config;
+    decoder_config.codec_name = "h264";
+    auto decoder_result = sc::DecoderFactory::create(decoder_config);
+    expect_result("decoder create", decoder_result);
+
+    ReceiverSink sink;
+    sink.decoder = std::move(sc::codec_value(decoder_result));
+
+    // Bind the receiver to the first free port in a small range.
+    auto receiver_transport = sc::RtpTransportFactory::create();
+    uint16_t bound_port = 0;
+    for (uint16_t port = 45904; port < 45934; ++port) {
+        if (receiver_transport->start(sc::Endpoint{"0.0.0.0", port},
+                                      [&sink](sc::RtpPacket packet) { sink.on_packet(std::move(packet)); })) {
+            bound_port = port;
+            break;
+        }
+    }
+    check(bound_port != 0, "receiver transport bound");
+
+    // Sender transport: no bind (OS picks the source port), peer is loopback.
+    auto sender_transport = sc::RtpTransportFactory::create();
+    check(sender_transport->start(sc::Endpoint{"", 0}, [](sc::RtpPacket) {}), "sender transport started");
+    sender_transport->set_peer(sc::Endpoint{"127.0.0.1", bound_port});
+
+    sc::H264Packetizer packetizer;
+    for (uint32_t index = 0; index < kFrameCount; ++index) {
+        const sc::CapturedFrame frame = make_frame(index);
+
+        auto encoded_result = encoder->encode(frame);
+        expect_result("encode", encoded_result);
+
+        for (const sc::EncodedFrame& encoded : sc::codec_value(encoded_result)) {
+            for (const sc::RtpPacket& packet : packetizer.packetize(encoded)) {
+                check(sender_transport->send(packet), "send packet");
+            }
+        }
+    }
+
+    // Wait (bounded) for the receiver to decode the frames.
+    const auto deadline = std::chrono::steady_clock::now() + std::chrono::seconds(5);
+    while (sink.decoded_frames.load() < kFrameCount && std::chrono::steady_clock::now() < deadline) {
+        std::this_thread::sleep_for(std::chrono::milliseconds(10));
+    }
+
+    check(sink.decoded_frames.load() >= kFrameCount / 2, "most frames decoded");
+    check(sink.saw_frame, "decoded frame has pixels");
+    check(sink.last_width == kWidth && sink.last_height == kHeight, "decoded dimensions");
+
+    sender_transport->stop();
+    receiver_transport->stop();
+
+    std::printf("test_loopback: %d frames decoded over UDP loopback\n", sink.decoded_frames.load());
+    return 0;
+}

tests/codec/test_roundtrip.cpp

@@ -85,14 +85,16 @@ int main() {
     expect_codec_result("encoder create", encoder_result);
     auto encoder = std::move(sc::codec_value(encoder_result));
 
+    // Without GLOBAL_HEADER the encoder carries no extradata; SPS/PPS are
+    // emitted in-band ahead of every keyframe. This mirrors the streaming
+    // path, where a receiver starts decoding from the bitstream alone.
     const auto extradata = encoder->get_extradata();
-    assert(!extradata.empty());
+    assert(extradata.empty());
 
     sc::DecoderConfig decoder_config;
     decoder_config.codec_name = "h264";
     decoder_config.width = kWidth;
     decoder_config.height = kHeight;
-    decoder_config.extradata = extradata;
 
     auto decoder_result = sc::DecoderFactory::create(decoder_config);
     expect_codec_result("decoder create", decoder_result);

tests/meson.build

@@ -15,3 +15,9 @@ test_rtp = executable('test_rtp',
     dependencies : sc_network_dep)
 
 test('rtp framing', test_rtp)
+
+test_loopback = executable('test_loopback',
+    'app/test_loopback.cpp',
+    dependencies : [sc_codec_dep, sc_network_dep])
+
+test('udp loopback', test_loopback)

tools/capture_smoke.cpp

@@ -97,9 +97,8 @@ int main(int argc, char** argv) {
             }
             encoder = std::move(sc::codec_value(encoder_result));
 
-            // The encoder is configured with AV_CODEC_FLAG_GLOBAL_HEADER, so
-            // parameter sets live in extradata; prepend them so the output
-            // file is a self-contained Annex-B stream.
+            // The encoder repeats SPS/PPS in-band at each keyframe, so the
+            // file is a self-contained Annex-B stream without a prefix.
             const auto extradata = encoder->get_extradata();
             if (!extradata.empty()) {
                 if (!write_bytes(output, extradata)) {