diff --git a/tests/test_dag.py b/tests/test_dag.py
index cb292a6..2a4e26f 100644
--- a/tests/test_dag.py
+++ b/tests/test_dag.py
@@ -348,3 +348,63 @@ class TestDAGComplexScenarios:
         assert len(sub) == 2
         assert set(sub["D"]) == {"E"}
         assert set(sub["E"]) == {"F"}
+
+        nested = sub.subgraph("E")
+        assert all_nodes(nested) == {"E", "F"}
+        assert set(nested.reverse["E"]) == set()
+        nested.remove_edge("E", "F")
+        assert "F" in sub["E"]
+        assert "F" in g["E"]
+
+    @pytest.mark.parametrize(
+        ("seeds", "reverse", "expected_nodes", "expected_edges"),
+        [
+            ({"C", "D"}, False, {"C", "D", "E", "F"}, {("D", "E"), ("E", "F")}),
+            ("D", True, {"A", "B", "D"}, {("D", "B"), ("B", "A")}),
+            ("F", False, {"F"}, set()),
+        ],
+        ids=["multiple-seeds", "reverse-extraction", "sink-only"],
+    )
+    def test_subgraph_boundaries(
+        self,
+        seeds: str | set[str],
+        reverse: bool,
+        expected_nodes: set[str],
+        expected_edges: set[tuple[str, str]],
+    ) -> None:
+        graph = DAG[str]()
+        graph["A"] = "B"
+        graph["B"] = {"C", "D"}
+        graph["D"] = "E"
+        graph["E"] = "F"
+
+        subgraph = (graph.reverse if reverse else graph).subgraph(seeds)
+
+        assert all_nodes(subgraph) == expected_nodes
+        assert set(subgraph.topo_sort()) == expected_nodes
+        assert len(subgraph) == len(expected_edges)
+        assert {(source, target) for source in subgraph for target in subgraph[source]} == expected_edges
+        assert {
+            (source, target) for target in subgraph.reverse for source in subgraph.reverse[target]
+        } == expected_edges
+
+    def test_subgraph_preserves_callbacks_without_firing_during_extraction(self) -> None:
+        added: list[tuple[str, str]] = []
+        removed: list[tuple[str, str]] = []
+        graph = DAG[str]()
+        graph.add_edge("A", "B")
+        graph.on_add = lambda source, target: added.append((source, target))
+        graph.on_remove = lambda source, target: removed.append((source, target))
+
+        subgraph = graph.subgraph("A")
+
+        assert subgraph.on_add is graph.on_add
+        assert subgraph.on_remove is graph.on_remove
+        assert added == []
+        assert removed == []
+        subgraph["A"].add("C")
+        subgraph["A"].discard("B")
+        assert added == [("A", "C")]
+        assert removed == [("A", "B")]
+        assert set(graph["A"]) == {"B"}
+        assert set(graph.reverse["B"]) == {"A"}
diff --git a/tests/test_example_build-graph.py b/tests/test_example_build-graph.py
new file mode 100644
index 0000000..c6a1776
--- /dev/null
+++ b/tests/test_example_build-graph.py
@@ -0,0 +1,76 @@
+import pytest
+
+from daglib.dag import DAG
+from daglib.operations import all_nodes
+
+
+@pytest.fixture
+def build_graph() -> DAG[str]:
+    graph = DAG[str]()
+    graph.add_edge("checkout", "compile")
+    graph.add_edge("toolchain", "compile")
+    graph.add_edge("compile", "package")
+    graph.add_edge("package", "deploy")
+    return graph
+
+
+def test_subgraph_reverse_mutation_preserves_parent_edge_consistency(build_graph: DAG[str]) -> None:
+    assert "compile" in build_graph["toolchain"]
+    assert "toolchain" in build_graph.reverse["compile"]
+
+    pipeline = build_graph.subgraph("checkout")
+    pipeline.reverse["compile"].discard("toolchain")
+
+    forward_has_edge = "compile" in build_graph["toolchain"]
+    reverse_has_edge = "toolchain" in build_graph.reverse["compile"]
+    assert forward_has_edge == reverse_has_edge, "Subgraph mutation desynchronized the parent DAG's edge indexes"
+    assert forward_has_edge
+
+
+def test_subgraph_reverse_mutation_preserves_parent_topological_dependencies(build_graph: DAG[str]) -> None:
+    assert "toolchain" in build_graph.topo_sort()
+
+    pipeline = build_graph.subgraph("checkout")
+    pipeline.reverse["compile"].discard("toolchain")
+
+    assert "compile" in build_graph["toolchain"]
+    order = build_graph.topo_sort()
+    assert "toolchain" in order, "Topological sorting omitted a node with an outgoing edge in the parent DAG"
+    assert order.index("toolchain") < order.index("compile")
+
+
+def test_build_subgraph_excludes_external_dependencies(build_graph: DAG[str]) -> None:
+    pipeline = build_graph.subgraph("checkout")
+
+    assert all_nodes(pipeline) == {"checkout", "compile", "package", "deploy"}
+    assert set(pipeline.reverse["compile"]) == {"checkout"}
+    assert len(pipeline) == 3
+    order = pipeline.topo_sort()
+    assert set(order) == all_nodes(pipeline)
+    assert order.index("checkout") < order.index("compile") < order.index("package") < order.index("deploy")
+
+
+@pytest.mark.parametrize("mutate_parent", [False, True], ids=["child-mutation", "parent-mutation"])
+@pytest.mark.parametrize("mutation", ["add", "remove", "reverse-remove", "discard-node"])
+def test_build_subgraph_mutations_are_isolated(build_graph: DAG[str], mutate_parent: bool, mutation: str) -> None:
+    pipeline = build_graph.subgraph("checkout")
+    changed, protected = (build_graph, pipeline) if mutate_parent else (pipeline, build_graph)
+    successors = {node: set(protected[node]) for node in protected}
+    predecessors = {node: set(protected.reverse[node]) for node in protected.reverse}
+
+    match mutation:
+        case "add":
+            changed["package"].add("publish")
+        case "remove":
+            changed["compile"].discard("package")
+        case "reverse-remove":
+            changed.reverse["compile"].discard("checkout")
+        case "discard-node":
+            changed.discard_node("compile")
+
+    assert {node: set(protected[node]) for node in protected} == successors
+    assert {node: set(protected.reverse[node]) for node in protected.reverse} == predecessors
+    for graph in (changed, protected):
+        forward_edges = {(source, target) for source in graph for target in graph[source]}
+        reverse_edges = {(source, target) for target in graph.reverse for source in graph.reverse[target]}
+        assert forward_edges == reverse_edges
diff --git a/tests/test_performance.py b/tests/test_performance.py
new file mode 100644
index 0000000..c616bec
--- /dev/null
+++ b/tests/test_performance.py
@@ -0,0 +1,334 @@
+"""Performance test scaffolding for generating large DAGs."""
+
+from __future__ import annotations
+
+import random
+import time
+from collections.abc import Callable
+
+import pytest
+
+from daglib.dag import DAG
+
+
+class DAGGenerator:
+    """Utility class for procedurally generating test DAGs."""
+
+    @staticmethod
+    def chain(n: int) -> DAG[int]:
+        """
+        Generate a simple chain: 0 -> 1 -> 2 -> ... -> n-1
+
+        Time complexity: O(n)
+        Edge count: n-1
+        """
+        g = DAG[int]()
+        for i in range(n - 1):
+            g.add_edge(i, i + 1)
+        return g
+
+    @staticmethod
+    def tree(depth: int, branching_factor: int = 2) -> DAG[int]:
+        """
+        Generate a balanced tree with given depth and branching factor.
+
+        Args:
+            depth: Depth of the tree (root is depth 0)
+            branching_factor: Number of children per node
+
+        Time complexity: O(b^d) where b=branching_factor, d=depth
+        Node count: (b^(d+1) - 1) / (b - 1)
+        """
+        g = DAG[int]()
+        node_id = 0
+        queue = [(node_id, 0)]  # (node_id, current_depth)
+        node_id += 1
+
+        while queue:
+            parent, curr_depth = queue.pop(0)
+            if curr_depth < depth:
+                for _ in range(branching_factor):
+                    child = node_id
+                    g.add_edge(parent, child)
+                    queue.append((child, curr_depth + 1))
+                    node_id += 1
+
+        return g
+
+    @staticmethod
+    def diamond(layers: int, nodes_per_layer: int = 10) -> DAG[int]:
+        """
+        Generate a diamond-shaped DAG with multiple layers.
+        Each node in layer i connects to all nodes in layer i+1.
+
+        Args:
+            layers: Number of layers
+            nodes_per_layer: Number of nodes in each middle layer
+
+        Edge count: O(layers * nodes_per_layer^2)
+        """
+        g = DAG[int]()
+        node_id = 0
+
+        # Create layers
+        layer_nodes = []
+        for layer in range(layers):
+            if layer == 0 or layer == layers - 1:
+                # Source and sink are single nodes
+                layer_nodes.append([node_id])
+                node_id += 1
+            else:
+                # Middle layers have multiple nodes
+                current_layer = list(range(node_id, node_id + nodes_per_layer))
+                layer_nodes.append(current_layer)
+                node_id += nodes_per_layer
+
+        # Connect layers
+        for i in range(len(layer_nodes) - 1):
+            for u in layer_nodes[i]:
+                for v in layer_nodes[i + 1]:
+                    g.add_edge(u, v)
+
+        return g
+
+    @staticmethod
+    def random_dag(n_nodes: int, edge_probability: float = 0.1, seed: int | None = None) -> DAG[int]:
+        """
+        Generate a random DAG by considering all possible edges i->j where i < j.
+
+        Args:
+            n_nodes: Number of nodes
+            edge_probability: Probability of creating each possible edge
+            seed: Random seed for reproducibility
+
+        Expected edge count: n*(n-1)/2 * edge_probability
+        """
+        if seed is not None:
+            random.seed(seed)
+
+        g = DAG[int]()
+        for i in range(n_nodes):
+            for j in range(i + 1, n_nodes):
+                if random.random() < edge_probability:
+                    g.add_edge(i, j)
+
+        return g
+
+    @staticmethod
+    def grid_dag(rows: int, cols: int) -> DAG[tuple[int, int]]:
+        """
+        Generate a grid DAG where edges flow right and down.
+        Node (i,j) connects to (i+1,j) and (i,j+1).
+
+        Args:
+            rows: Number of rows
+            cols: Number of columns
+
+        Node count: rows * cols
+        Edge count: ~2 * rows * cols
+        """
+        g = DAG[tuple[int, int]]()
+
+        for i in range(rows):
+            for j in range(cols):
+                # Add edge going right
+                if j < cols - 1:
+                    g.add_edge((i, j), (i, j + 1))
+                # Add edge going down
+                if i < rows - 1:
+                    g.add_edge((i, j), (i + 1, j))
+
+        return g
+
+    @staticmethod
+    def layered_random(
+        n_layers: int, nodes_per_layer: int, connection_probability: float = 0.3, seed: int | None = None
+    ) -> DAG[int]:
+        """
+        Generate a layered DAG where nodes only connect to the next layer.
+
+        Args:
+            n_layers: Number of layers
+            nodes_per_layer: Number of nodes in each layer
+            connection_probability: Probability of edge between adjacent layers
+            seed: Random seed
+
+        Guarantees: No cycles (edges only go forward)
+        """
+        if seed is not None:
+            random.seed(seed)
+
+        g = DAG[int]()
+        node_id = 0
+        layers = []
+
+        # Create layers
+        for _ in range(n_layers):
+            layer = list(range(node_id, node_id + nodes_per_layer))
+            layers.append(layer)
+            node_id += nodes_per_layer
+
+        # Connect adjacent layers
+        for i in range(len(layers) - 1):
+            for u in layers[i]:
+                for v in layers[i + 1]:
+                    if random.random() < connection_probability:
+                        g.add_edge(u, v)
+
+        return g
+
+    @staticmethod
+    def complete_bipartite(n_left: int, n_right: int) -> DAG[int]:
+        """
+        Generate a complete bipartite DAG.
+        All nodes in left partition connect to all nodes in right partition.
+
+        Args:
+            n_left: Number of nodes in left partition
+            n_right: Number of nodes in right partition
+
+        Edge count: n_left * n_right
+        """
+        g = DAG[int]()
+
+        for i in range(n_left):
+            for j in range(n_left, n_left + n_right):
+                g.add_edge(i, j)
+
+        return g
+
+
+# Performance benchmarking utilities
+
+
+def benchmark_operation(operation: Callable[[], None], n_runs: int = 3) -> dict[str, float]:
+    """
+    Benchmark an operation by running it multiple times.
+
+    Returns:
+        Dictionary with min, max, avg, and median times in seconds
+    """
+    times = []
+    for _ in range(n_runs):
+        start = time.perf_counter()
+        operation()
+        end = time.perf_counter()
+        times.append(end - start)
+
+    times.sort()
+    return {
+        "min": min(times),
+        "max": max(times),
+        "avg": sum(times) / len(times),
+        "median": times[len(times) // 2],
+    }
+
+
+# Example performance tests
+
+
+class TestPerformanceChain:
+    """Performance tests for chain DAGs."""
+
+    @pytest.mark.parametrize("n", [100, 1000, 10_000])
+    def test_chain_creation(self, n: int) -> None:
+        """Test creation time for chain DAGs."""
+        result = benchmark_operation(lambda: DAGGenerator.chain(n))
+        print(f"\nChain({n}) creation: {result['avg']:.4f}s")
+
+    @pytest.mark.parametrize("n", [100, 1000, 5_000])
+    def test_chain_traversal(self, n: int) -> None:
+        """Test traversal time for chain DAGs."""
+        g = DAGGenerator.chain(n)
+        result = benchmark_operation(lambda: g.topo_sort())
+        print(f"\nChain({n}) topo_sort: {result['avg']:.4f}s")
+
+
+class TestPerformanceTree:
+    """Performance tests for tree DAGs."""
+
+    @pytest.mark.parametrize("depth,branching", [(5, 2), (4, 3), (3, 5)])
+    def test_tree_creation(self, depth: int, branching: int) -> None:
+        """Test creation time for tree DAGs."""
+        result = benchmark_operation(lambda: DAGGenerator.tree(depth, branching))
+        g = DAGGenerator.tree(depth, branching)
+        print(f"\nTree(depth={depth}, branch={branching}, nodes={len(g._succ)}) creation: {result['avg']:.4f}s")
+
+
+class TestPerformanceDiamond:
+    """Performance tests for diamond DAGs."""
+
+    @pytest.mark.parametrize("layers,nodes", [(5, 10), (5, 50), (10, 20)])
+    def test_diamond_creation(self, layers: int, nodes: int) -> None:
+        """Test creation time for diamond DAGs."""
+        result = benchmark_operation(lambda: DAGGenerator.diamond(layers, nodes))
+        g = DAGGenerator.diamond(layers, nodes)
+        print(
+            f"\nDiamond(layers={layers}, nodes_per_layer={nodes}, total_edges={len(g)}) creation: {result['avg']:.4f}s"
+        )
+
+
+class TestPerformanceRandom:
+    """Performance tests for random DAGs."""
+
+    @pytest.mark.parametrize("n,prob", [(100, 0.1), (500, 0.05), (1000, 0.01)])
+    def test_random_creation(self, n: int, prob: float) -> None:
+        """Test creation time for random DAGs."""
+        result = benchmark_operation(lambda: DAGGenerator.random_dag(n, prob, seed=42))
+        g = DAGGenerator.random_dag(n, prob, seed=42)
+        print(f"\nRandomDAG(n={n}, p={prob}, edges={len(g)}) creation: {result['avg']:.4f}s")
+
+
+# Usage examples in docstring
+
+"""
+Example usage in your own tests:
+
+```python
+from tests.test_performance import DAGGenerator, benchmark_operation
+
+# Generate a large chain for testing
+g = DAGGenerator.chain(10000)
+
+# Generate a random DAG
+g = DAGGenerator.random_dag(1000, edge_probability=0.05, seed=42)
+
+# Generate a layered DAG (useful for realistic pipeline tests)
+g = DAGGenerator.layered_random(10, 100, connection_probability=0.2)
+
+# Benchmark a custom operation
+def my_operation():
+    g = DAGGenerator.tree(5, 3)
+    return g.topo_sort()
+
+stats = benchmark_operation(my_operation, n_runs=10)
+print(f"Operation took {stats['avg']:.4f}s on average")
+
+# Generate different graph types for stress testing
+graphs = {
+    "small_chain": DAGGenerator.chain(100),
+    "large_chain": DAGGenerator.chain(10000),
+    "balanced_tree": DAGGenerator.tree(depth=7, branching_factor=2),
+    "wide_tree": DAGGenerator.tree(depth=3, branching_factor=10),
+    "dense_diamond": DAGGenerator.diamond(5, 100),
+    "sparse_random": DAGGenerator.random_dag(1000, 0.01),
+    "dense_random": DAGGenerator.random_dag(100, 0.3),
+    "grid_10x10": DAGGenerator.grid_dag(10, 10),
+    "grid_100x100": DAGGenerator.grid_dag(100, 100),
+}
+```
+
+To run performance tests:
+```bash
+pytest tests/test_performance.py -v -s
+```
+
+To run specific test patterns:
+```bash
+# Run only chain tests
+pytest tests/test_performance.py::TestPerformanceChain -v -s
+
+# Run with specific parameters
+pytest tests/test_performance.py::TestPerformanceTree::test_tree_creation -v -s
+```
+"""