diff --git a/ROADMAP.md b/ROADMAP.md
index 7e075718d..dc9a2dcf2 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -107,12 +107,14 @@ sub-100 ms.
 After `FileOpened`, `bump_geometry()` fires; the first frame tessellates
 **every** model-space wire. Measurable hitch at ~100 k entities.
 
-### 2.1 Parallelize block-definition build
+### 2.1 Parallelize block-definition build ✅ DONE
 
-[`block_cache::build`](src/scene/block_cache.rs#L238) is single-threaded —
-the comment says "nested expansion is fiddly". Fix: stratify blocks in
-**topological order** (leaf blocks → callers → …) and build each layer in
-parallel via rayon. Dependency order is preserved.
+[`block_cache::build`](src/scene/block_cache.rs#L127) was single-threaded.
+No topological stratification was needed after all: `build_defn` stores
+nested INSERTs as by-name references (`LocalSub::Nested`) and never expands
+them at build time, so each defn depends only on the read-only `doc` — the
+builds are embarrassingly parallel. Now a plain rayon `par_iter().collect()`.
+`compute_block_aabbs` stays a serial post-pass (resolves nested refs, cheap).
 
 ### 2.2 Incremental wire cache (delta tessellation)
 
diff --git a/src/scene/block_cache.rs b/src/scene/block_cache.rs
index 0ea0d1741..d354f2264 100644
--- a/src/scene/block_cache.rs
+++ b/src/scene/block_cache.rs
@@ -125,12 +125,19 @@ impl BlockCache {
     /// because their entities are emitted as top-level wires, not via the
     /// cache.
     pub fn build(doc: &CadDocument, anno_scale: f32, bg_color: [f32; 4]) -> Self {
+        use rayon::prelude::*;
         let mut cache = Self::new();
         let referenced = collect_referenced_blocks(doc);
-        for name in &referenced {
-            let defn = build_defn(doc, name, anno_scale, bg_color);
-            cache.defns.insert(name.clone(), Arc::new(defn));
-        }
+        // Each defn is built independently: nested INSERTs are stored as
+        // by-name references (`LocalSub::Nested`), never expanded here, so a
+        // block's build never depends on another block's defn. That makes the
+        // builds embarrassingly parallel over the read-only `doc` — no
+        // topological ordering required. `compute_block_aabbs` stays a serial
+        // post-pass (it resolves nested references and is comparatively cheap).
+        cache.defns = referenced
+            .par_iter()
+            .map(|name| (name.clone(), Arc::new(build_defn(doc, name, anno_scale, bg_color))))
+            .collect();
         cache.compute_block_aabbs(&referenced);
         cache
     }