tools/spectral/.gitignore

@@ -0,0 +1,2 @@
+__pycache__
+output/*

tools/spectral/test_spectral.py

@@ -0,0 +1,727 @@
+#!/usr/bin/env python3
+"""Comprehensive tests for the spectral analysis tool."""
+
+from __future__ import annotations
+
+import copy
+import json
+import math
+import os
+import random
+import tempfile
+import unittest
+
+import numpy as np
+
+from spectral_analysis import (
+    AnalysisResult,
+    ComplexityMetrics,
+    DependencyGraph,
+    Edge,
+    ModuleCouplingResult,
+    Node,
+    SpectralResults,
+    build_adjacency,
+    build_laplacian,
+    compute_complexity_metrics,
+    compute_module_coupling,
+    compute_spectral,
+    compute_spectral_entropy,
+    count_connected_components,
+    generate_report,
+    get_node_ordering,
+    metrics_to_dict,
+    parse_dot,
+    run_analysis,
+    symmetrize,
+)
+
+
+# ─── Helpers ──────────────────────────────────────────────────────────────────
+
+def _make_graph(
+    names: list[str],
+    modules: list[str],
+    edge_pairs: list[tuple[str, str]],
+    module_order: list[str] | None = None,
+) -> DependencyGraph:
+    """Build a DependencyGraph from names, module assignments, and edges."""
+    assert len(names) == len(modules)
+    graph = DependencyGraph()
+    seen_modules: list[str] = []
+    for name, mod in zip(names, modules):
+        graph.nodes.append(Node(name=name, module=mod))
+        graph.node_to_module[name] = mod
+        if mod not in seen_modules:
+            seen_modules.append(mod)
+    if module_order is not None:
+        graph.modules = module_order
+    else:
+        graph.modules = seen_modules
+    for src, tgt in edge_pairs:
+        src_mod = graph.node_to_module.get(src, "")
+        tgt_mod = graph.node_to_module.get(tgt, "")
+        cross = src_mod != tgt_mod
+        graph.edges.append(Edge(
+            source=src, target=tgt, label="dep",
+            edge_type="field", cross_module=cross,
+        ))
+    return graph
+
+
+# ─── DOT Parser Tests ─────────────────────────────────────────────────────────
+
+class TestDotParser(unittest.TestCase):
+    def test_minimal_dot(self):
+        dot = '''digraph test {
+    subgraph cluster_mod1 {
+        label="mod1";
+        A [label="A", fillcolor="#fff"];
+    }
+    A -> A [label="self", style=dashed, color="#666", penwidth=1];
+}'''
+        g = parse_dot(dot)
+        self.assertEqual(len(g.nodes), 1)
+        self.assertEqual(g.nodes[0].name, "A")
+        self.assertEqual(g.nodes[0].module, "mod1")
+        self.assertEqual(len(g.edges), 1)
+
+    def test_two_module_dot(self):
+        dot = '''digraph test {
+    subgraph cluster_alpha {
+        label="alpha";
+        X [label="X"];
+        Y [label="Y"];
+    }
+    subgraph cluster_beta {
+        label="beta";
+        Z [label="Z"];
+    }
+    X -> Y [label="dep", style=dashed, color="#666", penwidth=1];
+    X -> Z [label="dep", style=solid, color="#00f", penwidth=1.5];
+}'''
+        g = parse_dot(dot)
+        self.assertEqual(len(g.nodes), 3)
+        self.assertEqual(len(g.modules), 2)
+        self.assertEqual(g.modules, ["alpha", "beta"])
+        self.assertEqual(g.node_to_module["X"], "alpha")
+        self.assertEqual(g.node_to_module["Z"], "beta")
+
+        # Edge classification
+        intra = [e for e in g.edges if not e.cross_module]
+        cross = [e for e in g.edges if e.cross_module]
+        self.assertEqual(len(intra), 1)
+        self.assertEqual(len(cross), 1)
+
+    def test_trait_impl_classification(self):
+        dot = '''digraph test {
+    subgraph cluster_m {
+        label="m";
+        A [label="A"];
+        B [label="B"];
+    }
+    A -> B [label="impl", style=dotted, color="#666", penwidth=1];
+}'''
+        g = parse_dot(dot)
+        self.assertEqual(g.edges[0].edge_type, "trait_impl")
+
+    def test_real_deps_dot(self):
+        """Parse the real deps.dot and verify expected counts."""
+        dot_path = os.path.join(os.path.dirname(__file__), "..", "..", "deps.dot")
+        if not os.path.exists(dot_path):
+            self.skipTest("deps.dot not found")
+        with open(dot_path) as f:
+            dot = f.read()
+        g = parse_dot(dot)
+        self.assertEqual(len(g.nodes), 36, f"Expected 36 nodes, got {len(g.nodes)}")
+        self.assertEqual(len(g.edges), 89, f"Expected 89 edges, got {len(g.edges)}")
+        self.assertEqual(len(g.modules), 8, f"Expected 8 modules, got {len(g.modules)}")
+
+    def test_empty_dot(self):
+        dot = "digraph empty {}"
+        g = parse_dot(dot)
+        self.assertEqual(len(g.nodes), 0)
+        self.assertEqual(len(g.edges), 0)
+
+
+# ─── Matrix Construction Tests ────────────────────────────────────────────────
+
+class TestMatrixConstruction(unittest.TestCase):
+    def test_two_node_adjacency(self):
+        g = _make_graph(["A", "B"], ["m", "m"], [("A", "B")])
+        order = get_node_ordering(g)
+        A = build_adjacency(g, order)
+        self.assertEqual(A.shape, (2, 2))
+        idx_a = order.index("A")
+        idx_b = order.index("B")
+        self.assertEqual(A[idx_a, idx_b], 1.0)
+        self.assertEqual(A[idx_b, idx_a], 0.0)
+
+    def test_symmetrize_directed(self):
+        A = np.array([[0, 1, 0],
+                      [0, 0, 1],
+                      [0, 0, 0]], dtype=float)
+        S = symmetrize(A)
+        expected = np.array([[0, 1, 0],
+                             [1, 0, 1],
+                             [0, 1, 0]], dtype=float)
+        np.testing.assert_array_equal(S, expected)
+
+    def test_symmetrize_idempotent(self):
+        """Symmetrizing an already-symmetric matrix should not change it."""
+        A = np.array([[0, 1, 1],
+                      [1, 0, 1],
+                      [1, 1, 0]], dtype=float)
+        S = symmetrize(A)
+        np.testing.assert_array_equal(S, A)
+
+    def test_laplacian_p3(self):
+        """Path graph P3: A-B-C."""
+        A_sym = np.array([[0, 1, 0],
+                          [1, 0, 1],
+                          [0, 1, 0]], dtype=float)
+        L = build_laplacian(A_sym)
+        expected = np.array([[1, -1, 0],
+                             [-1, 2, -1],
+                             [0, -1, 1]], dtype=float)
+        np.testing.assert_array_equal(L, expected)
+
+    def test_laplacian_k3(self):
+        """Complete graph K3."""
+        A_sym = np.array([[0, 1, 1],
+                          [1, 0, 1],
+                          [1, 1, 0]], dtype=float)
+        L = build_laplacian(A_sym)
+        expected = np.array([[2, -1, -1],
+                             [-1, 2, -1],
+                             [-1, -1, 2]], dtype=float)
+        np.testing.assert_array_equal(L, expected)
+
+
+# ─── Spectral Analysis Tests ─────────────────────────────────────────────────
+
+class TestSpectralAnalysis(unittest.TestCase):
+    def test_p3_eigenvalues(self):
+        """Path P3 should have eigenvalues {0, 1, 3}."""
+        g = _make_graph(["A", "B", "C"], ["m", "m", "m"],
+                        [("A", "B"), ("B", "C")])
+        s = compute_spectral(g)
+        np.testing.assert_allclose(sorted(s.eigenvalues), [0, 1, 3], atol=1e-10)
+
+    def test_k4_eigenvalues(self):
+        """Complete K4 should have eigenvalues {0, 4, 4, 4}."""
+        names = ["A", "B", "C", "D"]
+        edges = [(a, b) for a in names for b in names if a != b]
+        g = _make_graph(names, ["m"] * 4, edges)
+        s = compute_spectral(g)
+        np.testing.assert_allclose(sorted(s.eigenvalues), [0, 4, 4, 4], atol=1e-10)
+
+    def test_star_s4_fiedler(self):
+        """Star graph S4 (center + 3 leaves): lambda_2 = 1."""
+        g = _make_graph(
+            ["C", "L1", "L2", "L3"], ["m"] * 4,
+            [("C", "L1"), ("C", "L2"), ("C", "L3")],
+        )
+        s = compute_spectral(g)
+        self.assertAlmostEqual(s.fiedler_value, 1.0, places=10)
+
+    def test_disconnected_graph(self):
+        """Disconnected graph should have lambda_2 = 0."""
+        g = _make_graph(
+            ["A", "B", "C", "D"], ["m1", "m1", "m2", "m2"],
+            [("A", "B"), ("C", "D")],
+            module_order=["m1", "m2"],
+        )
+        s = compute_spectral(g)
+        self.assertAlmostEqual(s.fiedler_value, 0.0, places=10)
+
+    def test_barbell_fiedler_separation(self):
+        """Barbell graph: two K3 cliques connected by a bridge.
+
+        Fiedler vector should separate the two cliques (different signs).
+        """
+        # Clique 1: A, B, C fully connected
+        # Clique 2: D, E, F fully connected
+        # Bridge: C-D
+        names = ["A", "B", "C", "D", "E", "F"]
+        edges = [
+            ("A", "B"), ("A", "C"), ("B", "C"),
+            ("D", "E"), ("D", "F"), ("E", "F"),
+            ("C", "D"),
+        ]
+        g = _make_graph(names, ["m1", "m1", "m1", "m2", "m2", "m2"], edges,
+                        module_order=["m1", "m2"])
+        s = compute_spectral(g)
+
+        # Clique 1 nodes should have same sign, clique 2 opposite
+        order = s.node_names
+        fv = s.fiedler_vector
+        idx = {name: i for i, name in enumerate(order)}
+
+        clique1_signs = [np.sign(fv[idx[n]]) for n in ["A", "B", "C"]]
+        clique2_signs = [np.sign(fv[idx[n]]) for n in ["D", "E", "F"]]
+
+        # All in clique 1 should have the same sign
+        self.assertTrue(all(s == clique1_signs[0] for s in clique1_signs),
+                        f"Clique 1 signs should be uniform: {clique1_signs}")
+        # All in clique 2 should have the same sign
+        self.assertTrue(all(s == clique2_signs[0] for s in clique2_signs),
+                        f"Clique 2 signs should be uniform: {clique2_signs}")
+        # The two cliques should have opposite signs
+        self.assertNotEqual(clique1_signs[0], clique2_signs[0],
+                            "Cliques should have opposite Fiedler signs")
+
+    def test_single_node(self):
+        g = _make_graph(["A"], ["m"], [])
+        s = compute_spectral(g)
+        self.assertEqual(s.fiedler_value, 0.0)
+        self.assertEqual(len(s.eigenvalues), 1)
+
+    def test_empty_graph(self):
+        g = DependencyGraph()
+        s = compute_spectral(g)
+        self.assertEqual(s.fiedler_value, 0.0)
+        self.assertEqual(len(s.eigenvalues), 0)
+
+
+# ─── Module Coupling Tests ────────────────────────────────────────────────────
+
+class TestModuleCoupling(unittest.TestCase):
+    def test_directed_counts(self):
+        g = _make_graph(
+            ["A", "B", "C"], ["m1", "m1", "m2"],
+            [("A", "C"), ("B", "C"), ("C", "A")],
+            module_order=["m1", "m2"],
+        )
+        c = compute_module_coupling(g)
+        # m1->m2: 2 edges (A->C, B->C)
+        # m2->m1: 1 edge (C->A)
+        idx_m1 = c.module_names.index("m1")
+        idx_m2 = c.module_names.index("m2")
+        self.assertEqual(c.coupling_matrix[idx_m1, idx_m2], 2.0)
+        self.assertEqual(c.coupling_matrix[idx_m2, idx_m1], 1.0)
+
+    def test_cross_module_ratio(self):
+        g = _make_graph(
+            ["A", "B", "C", "D"], ["m1", "m1", "m2", "m2"],
+            [("A", "B"), ("A", "C"), ("C", "D")],
+            module_order=["m1", "m2"],
+        )
+        c = compute_module_coupling(g)
+        # 1 cross-module edge (A->C) out of 3 total
+        self.assertEqual(c.cross_module_edges, 1)
+        self.assertEqual(c.total_edges, 3)
+
+    def test_intra_only(self):
+        g = _make_graph(
+            ["A", "B"], ["m1", "m1"],
+            [("A", "B")],
+            module_order=["m1"],
+        )
+        c = compute_module_coupling(g)
+        self.assertEqual(c.cross_module_edges, 0)
+        self.assertEqual(c.coupling_matrix[0, 0], 1.0)
+
+
+# ─── Complexity Metrics Tests ─────────────────────────────────────────────────
+
+class TestComplexityMetrics(unittest.TestCase):
+    def test_k4_spectral_entropy(self):
+        """K4 has uniform positive eigenvalues {4,4,4} -> entropy = log2(3)."""
+        evals = np.array([0.0, 4.0, 4.0, 4.0])
+        H = compute_spectral_entropy(evals)
+        self.assertAlmostEqual(H, math.log2(3), places=10)
+
+    def test_star_entropy_less_than_complete(self):
+        """Star graph has less uniform eigenvalues than complete graph."""
+        # Star S4: eigenvalues are 0, 1, 1, 4
+        star_evals = np.array([0.0, 1.0, 1.0, 4.0])
+        k4_evals = np.array([0.0, 4.0, 4.0, 4.0])
+        H_star = compute_spectral_entropy(star_evals)
+        H_k4 = compute_spectral_entropy(k4_evals)
+        self.assertLess(H_star, H_k4)
+
+    def test_cci_in_range(self):
+        """CCI should always be in [0, 1]."""
+        for _ in range(20):
+            n = random.randint(2, 10)
+            names = [f"N{i}" for i in range(n)]
+            mods = [f"m{i % 3}" for i in range(n)]
+            edges = []
+            for _ in range(random.randint(1, n * 2)):
+                a, b = random.sample(names, 2)
+                edges.append((a, b))
+            g = _make_graph(names, mods, edges,
+                            module_order=sorted(set(mods)))
+            result = run_analysis(g)
+            self.assertGreaterEqual(result.metrics.cci, 0.0,
+                                    "CCI should be >= 0")
+            self.assertLessEqual(result.metrics.cci, 1.0,
+                                 "CCI should be <= 1")
+
+    def test_cci_increases_with_coupling(self):
+        """Adding cross-module edges should increase CCI."""
+        # Base graph: two modules, minimal coupling
+        g1 = _make_graph(
+            ["A", "B", "C", "D"], ["m1", "m1", "m2", "m2"],
+            [("A", "B"), ("C", "D"), ("A", "C")],
+            module_order=["m1", "m2"],
+        )
+        # More coupling
+        g2 = _make_graph(
+            ["A", "B", "C", "D"], ["m1", "m1", "m2", "m2"],
+            [("A", "B"), ("C", "D"), ("A", "C"), ("A", "D"),
+             ("B", "C"), ("B", "D"), ("C", "A"), ("D", "B")],
+            module_order=["m1", "m2"],
+        )
+        r1 = run_analysis(g1)
+        r2 = run_analysis(g2)
+        self.assertLess(r1.metrics.cci, r2.metrics.cci)
+
+    def test_connected_components(self):
+        A_sym = np.array([
+            [0, 1, 0, 0],
+            [1, 0, 0, 0],
+            [0, 0, 0, 1],
+            [0, 0, 1, 0],
+        ], dtype=float)
+        self.assertEqual(count_connected_components(A_sym), 2)
+
+    def test_single_component(self):
+        A_sym = np.array([
+            [0, 1, 1],
+            [1, 0, 1],
+            [1, 1, 0],
+        ], dtype=float)
+        self.assertEqual(count_connected_components(A_sym), 1)
+
+
+# ─── Complexity Ladder ────────────────────────────────────────────────────────
+
+class TestComplexityLadder(unittest.TestCase):
+    """Verify CCI correctly orders synthetic codebases of increasing complexity."""
+
+    def _rung1_linear_chain(self) -> DependencyGraph:
+        """5 nodes in a single module, linear chain A->B->C->D->E."""
+        return _make_graph(
+            ["A", "B", "C", "D", "E"],
+            ["m1"] * 5,
+            [("A", "B"), ("B", "C"), ("C", "D"), ("D", "E")],
+            module_order=["m1"],
+        )
+
+    def _rung2_clean_tree(self) -> DependencyGraph:
+        """6 nodes across 2 modules, tree with mostly intra-module edges."""
+        return _make_graph(
+            ["R", "A", "B", "C", "D", "E"],
+            ["core", "core", "core", "util", "util", "util"],
+            [
+                ("R", "A"), ("A", "B"), ("R", "C"),  # intra core
+                ("D", "E"),  # intra util
+                ("R", "D"), ("C", "E"),  # 2 cross edges
+            ],
+            module_order=["core", "util"],
+        )
+
+    def _rung3_layered_dag(self) -> DependencyGraph:
+        """8 nodes across 3 modules in a layered architecture."""
+        return _make_graph(
+            ["C1", "C2", "S1", "S2", "S3", "D1", "D2", "D3"],
+            ["ctrl", "ctrl", "svc", "svc", "svc", "data", "data", "data"],
+            [
+                ("C1", "C2"),  # intra ctrl
+                ("S1", "S2"), ("S2", "S3"),  # intra svc
+                ("D1", "D2"), ("D2", "D3"),  # intra data
+                ("C1", "S1"), ("C1", "S2"), ("C2", "S3"),  # ctrl->svc
+                ("S1", "D1"), ("S2", "D2"), ("S3", "D3"),  # svc->data
+            ],
+            module_order=["ctrl", "svc", "data"],
+        )
+
+    def _rung4_diamond_cross(self) -> DependencyGraph:
+        """10 nodes across 5 modules with diamond patterns and cross-coupling."""
+        return _make_graph(
+            ["A1", "A2", "B1", "B2", "C1", "C2", "D1", "D2", "E1", "E2"],
+            ["ma", "ma", "mb", "mb", "mc", "mc", "md", "md", "me", "me"],
+            [
+                ("A1", "A2"), ("B1", "B2"), ("C1", "C2"),  # intra
+                ("D1", "D2"), ("E1", "E2"),  # intra
+                # Diamonds across modules
+                ("A1", "B1"), ("A1", "C1"), ("B1", "D1"), ("C1", "D1"),
+                ("A2", "B2"), ("A2", "C2"), ("B2", "D2"), ("C2", "D2"),
+                # Extra cross-coupling
+                ("D1", "E1"), ("D2", "E2"), ("B1", "E1"),
+            ],
+            module_order=["ma", "mb", "mc", "md", "me"],
+        )
+
+    def _rung5_hub_backlinks(self) -> DependencyGraph:
+        """10 nodes across 5 modules, hub-dominated with back-edges."""
+        return _make_graph(
+            ["Hub", "A1", "A2", "B1", "B2", "C1", "C2", "D1", "D2", "D3"],
+            ["core", "sa", "sa", "sb", "sb", "sc", "sc", "sd", "sd", "sd"],
+            [
+                ("A1", "A2"), ("B1", "B2"), ("C1", "C2"),  # intra
+                ("D1", "D2"), ("D2", "D3"),  # intra
+                # Hub connections (cross-module)
+                ("Hub", "A1"), ("Hub", "B1"), ("Hub", "C1"), ("Hub", "D1"),
+                ("A1", "Hub"), ("B1", "Hub"), ("C1", "Hub"),
+                # Additional cross-module
+                ("A1", "B1"), ("B1", "C1"), ("C1", "D1"),
+                ("A2", "B2"), ("B2", "C2"), ("C2", "D2"),
+                ("A1", "D1"), ("B2", "D3"),
+            ],
+            module_order=["core", "sa", "sb", "sc", "sd"],
+        )
+
+    def _rung6_dense_mesh(self) -> DependencyGraph:
+        """10 nodes across 4 modules with heavy cross-module coupling."""
+        names = ["X1", "X2", "X3", "Y1", "Y2", "Y3", "Z1", "Z2", "W1", "W2"]
+        mods = ["mx", "mx", "mx", "my", "my", "my", "mz", "mz", "mw", "mw"]
+        # Dense cross-module edges
+        edges = [
+            # intra
+            ("X1", "X2"), ("X2", "X3"), ("Y1", "Y2"), ("Y2", "Y3"),
+            ("Z1", "Z2"), ("W1", "W2"),
+            # cross - nearly every module to every other
+            ("X1", "Y1"), ("X1", "Z1"), ("X1", "W1"),
+            ("X2", "Y2"), ("X2", "Z2"), ("X2", "W2"),
+            ("X3", "Y3"), ("X3", "Z1"),
+            ("Y1", "X1"), ("Y1", "Z1"), ("Y1", "W1"),
+            ("Y2", "X2"), ("Y2", "Z2"),
+            ("Y3", "X3"), ("Y3", "W2"),
+            ("Z1", "X1"), ("Z1", "Y1"), ("Z1", "W1"),
+            ("Z2", "X2"), ("Z2", "Y2"), ("Z2", "W2"),
+            ("W1", "X1"), ("W1", "Y1"), ("W1", "Z1"),
+            ("W2", "X2"), ("W2", "Y2"), ("W2", "Z2"),
+        ]
+        return _make_graph(names, mods, edges,
+                           module_order=["mx", "my", "mz", "mw"])
+
+    def test_complexity_ladder(self):
+        """CCI must strictly increase across the ladder rungs."""
+        ladder = [
+            self._rung1_linear_chain(),
+            self._rung2_clean_tree(),
+            self._rung3_layered_dag(),
+            self._rung4_diamond_cross(),
+            self._rung5_hub_backlinks(),
+            self._rung6_dense_mesh(),
+        ]
+        ccis = [run_analysis(g).metrics.cci for g in ladder]
+        for i in range(len(ccis) - 1):
+            self.assertLess(
+                ccis[i], ccis[i + 1],
+                f"Rung {i + 1} (CCI={ccis[i]:.4f}) should be less complex "
+                f"than rung {i + 2} (CCI={ccis[i + 1]:.4f})"
+            )
+
+
+# ─── Perturbation Tests ──────────────────────────────────────────────────────
+
+class TestPerturbation(unittest.TestCase):
+    """Test that CCI responds correctly to architectural changes on the real graph."""
+
+    def _load_real_graph(self) -> DependencyGraph:
+        dot_path = os.path.join(os.path.dirname(__file__), "..", "..", "deps.dot")
+        if not os.path.exists(dot_path):
+            self.skipTest("deps.dot not found")
+        with open(dot_path) as f:
+            return parse_dot(f.read())
+
+    def test_remove_most_coupled_module(self):
+        """Removing the runtime module should decrease CCI."""
+        g = self._load_real_graph()
+        original_cci = run_analysis(g).metrics.cci
+
+        # Remove runtime nodes and their edges
+        g2 = DependencyGraph()
+        g2.modules = [m for m in g.modules if m != "runtime"]
+        for node in g.nodes:
+            if node.module != "runtime":
+                g2.nodes.append(node)
+                g2.node_to_module[node.name] = node.module
+        runtime_nodes = {n.name for n in g.nodes if n.module == "runtime"}
+        for edge in g.edges:
+            if edge.source not in runtime_nodes and edge.target not in runtime_nodes:
+                src_mod = g2.node_to_module.get(edge.source, "")
+                tgt_mod = g2.node_to_module.get(edge.target, "")
+                g2.edges.append(Edge(
+                    source=edge.source, target=edge.target, label=edge.label,
+                    edge_type=edge.edge_type,
+                    cross_module=src_mod != tgt_mod,
+                ))
+
+        reduced_cci = run_analysis(g2).metrics.cci
+        self.assertLess(reduced_cci, original_cci,
+                        f"Removing runtime should decrease CCI: "
+                        f"{reduced_cci:.4f} vs {original_cci:.4f}")
+
+    def test_add_random_cross_edges(self):
+        """Adding 10 random cross-module edges should increase CCI."""
+        g = self._load_real_graph()
+        original_cci = run_analysis(g).metrics.cci
+
+        g2 = copy.deepcopy(g)
+        random.seed(42)
+        node_names = [n.name for n in g2.nodes]
+        added = 0
+        attempts = 0
+        while added < 10 and attempts < 100:
+            src, tgt = random.sample(node_names, 2)
+            src_mod = g2.node_to_module[src]
+            tgt_mod = g2.node_to_module[tgt]
+            if src_mod != tgt_mod:
+                g2.edges.append(Edge(
+                    source=src, target=tgt, label="added",
+                    edge_type="field", cross_module=True,
+                ))
+                added += 1
+            attempts += 1
+
+        augmented_cci = run_analysis(g2).metrics.cci
+        self.assertGreater(augmented_cci, original_cci,
+                           f"Adding cross-module edges should increase CCI: "
+                           f"{augmented_cci:.4f} vs {original_cci:.4f}")
+
+    def test_merge_modules_decreases_cci(self):
+        """Merging two small modules into one should decrease CCI.
+
+        Merging error + config into a single module reduces cross-module
+        edges (their mutual and outward coupling consolidates), lowering CCI.
+        """
+        g = self._load_real_graph()
+        original_cci = run_analysis(g).metrics.cci
+
+        # Merge error and config into "error_config"
+        merge_set = {"error", "config"}
+        merged_name = "error_config"
+
+        g2 = DependencyGraph()
+        g2.modules = [merged_name if m in merge_set else m
+                       for m in g.modules if m not in merge_set]
+        if merged_name not in g2.modules:
+            g2.modules.insert(0, merged_name)
+        # Deduplicate
+        seen = set()
+        g2.modules = [m for m in g2.modules if not (m in seen or seen.add(m))]
+
+        for node in g.nodes:
+            new_mod = merged_name if node.module in merge_set else node.module
+            g2.nodes.append(Node(name=node.name, module=new_mod))
+            g2.node_to_module[node.name] = new_mod
+
+        for edge in g.edges:
+            src_mod = g2.node_to_module.get(edge.source, "")
+            tgt_mod = g2.node_to_module.get(edge.target, "")
+            g2.edges.append(Edge(
+                source=edge.source, target=edge.target, label=edge.label,
+                edge_type=edge.edge_type,
+                cross_module=src_mod != tgt_mod,
+            ))
+
+        merged_cci = run_analysis(g2).metrics.cci
+        self.assertLess(merged_cci, original_cci,
+                        f"Merging error+config should decrease CCI: "
+                        f"{merged_cci:.4f} vs {original_cci:.4f}")
+
+
+# ─── Integration Tests ────────────────────────────────────────────────────────
+
+class TestIntegration(unittest.TestCase):
+    def test_full_pipeline_real_graph(self):
+        """Run full pipeline on real deps.dot and sanity-check outputs."""
+        dot_path = os.path.join(os.path.dirname(__file__), "..", "..", "deps.dot")
+        if not os.path.exists(dot_path):
+            self.skipTest("deps.dot not found")
+        with open(dot_path) as f:
+            graph = parse_dot(f.read())
+
+        result = run_analysis(graph)
+
+        # Basic sanity checks
+        self.assertEqual(result.metrics.n_nodes, 36)
+        self.assertEqual(result.metrics.n_edges, 89)
+        self.assertEqual(result.metrics.n_modules, 8)
+
+        # Connected graph -> lambda_2 > 0
+        self.assertGreater(result.spectral.fiedler_value, 0,
+                           "Connected graph should have lambda_2 > 0")
+
+        # CCI should be in a reasonable range for a well-structured codebase
+        self.assertGreater(result.metrics.cci, 0.05)
+        self.assertLess(result.metrics.cci, 0.9)
+
+        # Eigenvalues should be non-negative (Laplacian property)
+        self.assertTrue(np.all(result.spectral.eigenvalues >= -1e-10),
+                        "Laplacian eigenvalues should be non-negative")
+
+        # First eigenvalue should be 0
+        self.assertAlmostEqual(result.spectral.eigenvalues[0], 0.0, places=8)
+
+    def test_report_generation(self):
+        """Verify report contains expected sections."""
+        dot_path = os.path.join(os.path.dirname(__file__), "..", "..", "deps.dot")
+        if not os.path.exists(dot_path):
+            self.skipTest("deps.dot not found")
+        with open(dot_path) as f:
+            graph = parse_dot(f.read())
+        result = run_analysis(graph)
+        report = generate_report(result)
+
+        self.assertIn("GRAPH SUMMARY", report)
+        self.assertIn("LAPLACIAN EIGENVALUE SPECTRUM", report)
+        self.assertIn("FIEDLER VECTOR", report)
+        self.assertIn("MODULE COUPLING MATRIX", report)
+        self.assertIn("CONNECTOME COMPLEXITY INDEX", report)
+
+    def test_json_output(self):
+        """Verify JSON output is well-formed and contains expected keys."""
+        g = _make_graph(
+            ["A", "B", "C"], ["m1", "m1", "m2"],
+            [("A", "B"), ("A", "C")],
+            module_order=["m1", "m2"],
+        )
+        result = run_analysis(g)
+        d = metrics_to_dict(result)
+
+        self.assertIn("graph", d)
+        self.assertIn("spectral", d)
+        self.assertIn("module_coupling", d)
+        self.assertIn("metrics", d)
+        self.assertEqual(d["graph"]["n_nodes"], 3)
+        self.assertIsInstance(d["spectral"]["eigenvalues"], list)
+        self.assertIsInstance(d["metrics"]["cci"], float)
+
+        # Should be JSON-serializable
+        json_str = json.dumps(d)
+        self.assertIsInstance(json_str, str)
+
+    def test_dashboard_generation(self):
+        """Verify dashboard PNG can be generated without errors."""
+        try:
+            import matplotlib
+        except ImportError:
+            self.skipTest("matplotlib not available")
+
+        g = _make_graph(
+            ["A", "B", "C", "D"], ["m1", "m1", "m2", "m2"],
+            [("A", "B"), ("A", "C"), ("C", "D")],
+            module_order=["m1", "m2"],
+        )
+        result = run_analysis(g)
+
+        with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as f:
+            path = f.name
+        try:
+            from spectral_analysis import generate_dashboard
+            generate_dashboard(result, path)
+            self.assertTrue(os.path.exists(path))
+            self.assertGreater(os.path.getsize(path), 1000,
+                               "Dashboard should be a non-trivial PNG")
+        finally:
+            os.unlink(path)
+
+
+if __name__ == "__main__":
+    unittest.main()