diff --git a/test_swarmmetrics.py b/test_swarmmetrics.py
index a45d04f..66acf49 100644
--- a/test_swarmmetrics.py
+++ b/test_swarmmetrics.py
@@ -4,7 +4,7 @@ import sys
 sys.path.insert(0, ".")
 from swarmmetrics import (
     _gini, _half_life_weight, score_reciprocity, score_channels,
-    score_echo, detect_shadows, analyze
+    score_echo, detect_shadows, analyze, shuffle_test
 )
 
 now = time.time()
@@ -218,6 +218,40 @@ def test_f33_gibberish_high_score():
     # The formula chain is an anti-spam filter, not a quality metric.
 
 
+def test_shuffle_test_causal_vs_correlation():
+    """Shuffle test (Anagnostopoulos et al. 2008) separates causal echo
+    from correlation. Genuine temporal diffusion should produce z-scores
+    significantly above shuffled baseline.
+    Cited: ColonistOne (Colony, 2026-07-28)."""
+    import random
+    random.seed(42)  # reproducibility
+
+    msgs = []
+    # Clear causal chain: alice introduces "reciprocity" at t=1,
+    # bob uses it at t=2 (after exposure), charlie uses it at t=3
+    msgs.append({"from_id": "alice", "to_id": "bob",
+                 "timestamp": now - 5*day, "concepts": ["reciprocity"]})
+    msgs.append({"from_id": "bob", "to_id": "charlie",
+                 "timestamp": now - 3*day, "concepts": ["reciprocity", "decay"]})
+    msgs.append({"from_id": "charlie", "to_id": "dave",
+                 "timestamp": now - 1*day, "concepts": ["reciprocity"]})
+    # Add some noise
+    msgs.append({"from_id": "dave", "to_id": "alice",
+                 "timestamp": now - 0.5*day, "concepts": ["noise"]})
+
+    result = shuffle_test(msgs, n_shuffles=50)
+
+    # Alice should have significant echo (she introduced "reciprocity"
+    # which spread temporally). Shuffling breaks the temporal order,
+    # so her observed echo should be higher than shuffled mean.
+    assert result["observed"].get("alice", 0) > 0, \
+        "Alice should have positive observed echo"
+    # The z-score should be positive (observed > shuffled mean)
+    z_alice = result["z_scores"].get("alice", 0)
+    assert z_alice > 0, \
+        f"Alice's echo should exceed shuffled baseline: z={z_alice}"
+
+
 def test_f33_adversarial_synonym_padding():
     """F3.3-aware adversary: one unique concept per message inflates score.
     Test for AX-7 (Colony): 'Do your tests include an adversary who has read F3.3?'