diff --git a/Cargo.toml b/Cargo.toml
index a858059..8e93a29 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -92,7 +92,7 @@ winres = "0.1"
 
 [profile.release]
 opt-level = 3
-lto = "fat"
+lto = "thin"
 codegen-units = 1
 panic = "abort"
 strip = true
diff --git a/plugins/cagire-plugins/src/editor.rs b/plugins/cagire-plugins/src/editor.rs
index 8289074..0c9e760 100644
--- a/plugins/cagire-plugins/src/editor.rs
+++ b/plugins/cagire-plugins/src/editor.rs
@@ -234,7 +234,7 @@ pub fn create_editor(
             // Read live snapshot from the audio thread
             let shared = editor.bridge.shared_state.load();
             editor.snapshot = SequencerSnapshot::from(shared.as_ref());
-            editor.app.playback.playing = editor.snapshot.playing;
+            editor.app.playback.playing = editor.playing.load(std::sync::atomic::Ordering::Relaxed);
 
             // Sync host tempo into LinkState so title bar shows real tempo
             if shared.tempo > 0.0 {
diff --git a/scripts/__pycache__/build.cpython-314.pyc b/scripts/__pycache__/build.cpython-314.pyc
index e3c1119..a4c4da3 100644
Binary files a/scripts/__pycache__/build.cpython-314.pyc and b/scripts/__pycache__/build.cpython-314.pyc differ
diff --git a/scripts/build.py b/scripts/build.py
index 796fff2..cac90a4 100755
--- a/scripts/build.py
+++ b/scripts/build.py
@@ -655,6 +655,55 @@ def _build_display(
     return Group(table, log_text)
 
 
+def _prebuild_cross_images(root: Path, platforms: list[Platform]) -> None:
+    """Pre-build Docker images for cross-compiled targets in parallel."""
+    cross_platforms = [p for p in platforms if p.cross]
+    if not cross_platforms:
+        return
+
+    cross_toml = root / "Cross.toml"
+    if not cross_toml.exists():
+        return
+
+    with open(cross_toml, "rb") as f:
+        cross_config = tomllib.load(f)
+
+    def build_image(p: Platform) -> None:
+        target_cfg = cross_config.get("target", {}).get(p.triple, {})
+        dockerfile = target_cfg.get("dockerfile")
+        if not dockerfile:
+            return
+        tag = f"cross-custom-{p.triple}:local"
+        try:
+            subprocess.run(
+                ["docker", "build", "-t", tag, "-f", str(root / dockerfile), str(root)],
+                capture_output=True, timeout=600,
+            )
+        except Exception:
+            pass  # non-critical, cross will build if needed
+
+    console.print("[dim]Pre-building cross-compilation Docker images...[/]")
+    with ThreadPoolExecutor(max_workers=len(cross_platforms)) as pool:
+        pool.map(build_image, cross_platforms)
+
+
+def _build_native_sequential(
+    root: Path,
+    native_platforms: list[Platform],
+    config: BuildConfig,
+    completed: dict[str, PlatformResult],
+    start_times: dict[str, float],
+) -> list[PlatformResult]:
+    """Build native platforms sequentially (they share the cargo target/ lock)."""
+    native_results = []
+    for p in native_platforms:
+        start_times[p.alias] = time.monotonic()
+        r = build_platform(root, p, config)
+        completed[r.platform.alias] = r
+        native_results.append(r)
+    return native_results
+
+
 def run_builds(
     root: Path, platforms: list[Platform], config: BuildConfig, version: str, verbose: bool = False,
 ) -> list[PlatformResult]:
@@ -665,6 +714,10 @@ def run_builds(
     for p in platforms:
         _update_phase(p.alias, "waiting", 0)
 
+    # Split into native (share cargo lock) and cross (independent Docker builds)
+    native_platforms = [p for p in platforms if not p.cross]
+    cross_platforms = [p for p in platforms if p.cross]
+
     results: list[PlatformResult] = []
     completed: dict[str, PlatformResult] = {}
     start_times: dict[str, float] = {p.alias: time.monotonic() for p in platforms}
@@ -676,15 +729,31 @@ def run_builds(
         return _build_display(platforms, config, completed, start_times, log_max_lines)
 
     with Live(make_display(), console=console, refresh_per_second=4) as live:
-        with ThreadPoolExecutor(max_workers=len(platforms)) as pool:
-            futures = {pool.submit(build_platform, root, p, config): p for p in platforms}
+        # Native builds run sequentially in one thread (they contend on cargo lock).
+        # Cross builds run in parallel (each in its own Docker container).
+        with ThreadPoolExecutor(max_workers=max(len(cross_platforms) + 1, 1)) as pool:
+            futures = {}
+
+            if native_platforms:
+                f = pool.submit(
+                    _build_native_sequential, root, native_platforms, config,
+                    completed, start_times,
+                )
+                futures[f] = "native"
+
+            for p in cross_platforms:
+                f = pool.submit(build_platform, root, p, config)
+                futures[f] = "cross"
+
             pending = set(futures.keys())
             while pending:
                 done = {f for f in pending if f.done()}
                 for f in done:
-                    r = f.result()
-                    completed[r.platform.alias] = r
-                    results.append(r)
+                    tag = futures[f]
+                    if tag == "native":
+                        results.extend(f.result())
+                    else:
+                        results.append(f.result())
                     pending.discard(f)
                 live.update(make_display())
                 if pending:
@@ -936,6 +1005,8 @@ def main() -> None:
 
     check_prerequisites(platforms, config)
 
+    _prebuild_cross_images(root, platforms)
+
     t0 = time.monotonic()
     results = run_builds(root, platforms, config, version, verbose=args.verbose)
     wall_time = time.monotonic() - t0