⚡ Bolt: [Replace .filter().collect() with manual loop and Vec::with_capacity]

.jules/bolt.md

@@ -4,3 +4,7 @@
 ## 2025-11-05 - [Replace HashMap Allocation with Flat Vec for Block Lookup]
 **Learning:** In the `flatten_blocks` function within `orch8-engine/src/evaluator.rs`, allocating a `HashMap<&BlockId, &BlockDefinition>` to map execution blocks introduces significant hashing and memory allocation overhead on the evaluation hot path, which is called every tick. Since the tree size is typically bounded and small, building a `Vec` and sorting it by `BlockId` allows for O(log N) lookups via `.binary_search_by_key()`, completely avoiding `HashMap` overhead.
 **Action:** In execution hot paths where a lookup map is created from a slice on every iteration, always prefer a flat `Vec` initialized with `Vec::with_capacity()`, sorted by key, and queried via `.binary_search_by_key()` over a `HashMap` to eliminate hashing and heap allocation costs.
+
+## 2025-11-06 - [Optimize children_of using Vec::with_capacity]
+**Learning:** In `orch8-engine/src/evaluator.rs`, the `children_of` function is called heavily on the hot path for all execution node evaluations. It previously chained `.filter(...).collect()` on the execution tree slice. Because `.filter()` creates an iterator of unknown size, `.collect()` causes the resulting `Vec` to allocate and potentially reallocate multiple times as elements are found.
+**Action:** Replace `.filter(...).collect()` chains in hot paths with a manual `for` loop pushing to a `Vec` instantiated with `Vec::with_capacity(N)` when a reasonable small bound (like 8 for most child node scenarios) is known. This significantly eliminates redundant iterator and allocator overhead.

orch8-engine/src/evaluator.rs

@@ -1250,12 +1250,19 @@ pub fn children_of(
     parent_id: ExecutionNodeId,
     branch_index: Option<i16>,
 ) -> Vec<&ExecutionNode> {
-    tree.iter()
-        .filter(|n| {
-            n.parent_id == Some(parent_id)
-                && (branch_index.is_none() || n.branch_index == branch_index)
-        })
-        .collect()
+    // ⚡ Bolt: Replace `.filter().collect()` with a manual loop and `Vec::with_capacity`
+    // to avoid multiple reallocation overheads when collecting an unknown-sized iterator.
+    // Most nodes have a small number of direct children, so preallocating a small capacity
+    // is highly efficient.
+    let mut children = Vec::with_capacity(8);
+    for n in tree {
+        if n.parent_id == Some(parent_id)
+            && (branch_index.is_none() || n.branch_index == branch_index)
+        {
+            children.push(n);
+        }
+    }
+    children
 }
 
 /// Check if all nodes in a set are in a terminal state.