diff --git a/max_outlier_analysis_script/tests/test_core.py b/max_outlier_analysis_script/tests/test_core.py
new file mode 100644
index 0000000..9e188ed
--- /dev/null
+++ b/max_outlier_analysis_script/tests/test_core.py
@@ -0,0 +1,77 @@
+import pytest
+import pandas as pd
+import math
+from pathlib import Path
+
+from src.max_outlier_analysis_script import core
+
+SAMPLE_CSV_PATH = Path(__file__).parent / 'data' / 'test_outlier_sample.csv'
+
+@pytest.fixture
+def sample_data():
+    """Lädt CSV-Testdaten in ein passendes Python-Objekt (Liste von Dicts)."""
+    if not SAMPLE_CSV_PATH.exists():
+        # Fallback: künstliche Testdaten, falls Datei fehlt
+        return [
+            {
+                'corr_id': 'c1', 'stratum': 'near-expiry-unpinned', 'job_parallelism': 5,
+                'expires_at_dist_hours': 0.2, 'retry_total_overhead_ms': 12.5, 'latency_max': 350.0
+            },
+            {
+                'corr_id': 'c2', 'stratum': 'normal', 'job_parallelism': 4,
+                'expires_at_dist_hours': 3.1, 'retry_total_overhead_ms': 6.2, 'latency_max': 180.0
+            },
+            {
+                'corr_id': 'c3', 'stratum': 'near-expiry-unpinned', 'job_parallelism': 6,
+                'expires_at_dist_hours': 0.1, 'retry_total_overhead_ms': 15.8, 'latency_max': 420.0
+            },
+        ]
+    df = pd.read_csv(SAMPLE_CSV_PATH)
+    return df.to_dict(orient='records')
+
+
+def test_analyze_returns_expected_keys(sample_data):
+    """Verifiziert, dass alle erwarteten Kennzahlen im Ergebnis enthalten sind."""
+    result = core.analyze_max_outliers(sample_data)
+
+    assert isinstance(result, dict)
+    expected_keys = {'max_above_p99_count', 'near_expiry_cluster_percentage', 'retry_overhead_variance'}
+    assert expected_keys.issubset(result.keys())
+
+
+def test_result_value_types(sample_data):
+    """Stellt sicher, dass die Rückgabewerte vom korrekten Typ sind."""
+    result = core.analyze_max_outliers(sample_data)
+
+    assert isinstance(result['max_above_p99_count'], int)
+    assert isinstance(result['near_expiry_cluster_percentage'], float)
+    assert isinstance(result['retry_overhead_variance'], float)
+
+
+def test_statistical_consistency(sample_data):
+    """Überprüft einfache statistische Plausibilität der berechneten Werte."""
+    result = core.analyze_max_outliers(sample_data)
+
+    # Max über p99 kann nicht negativ sein
+    assert result['max_above_p99_count'] >= 0
+    # Prozentsatz muss im Bereich 0-100 liegen
+    assert 0.0 <= result['near_expiry_cluster_percentage'] <= 100.0
+    # Varianz >= 0
+    assert result['retry_overhead_variance'] >= 0.0
+
+
+def test_error_handling_on_empty_input():
+    """Leere Eingaben sollten einen kontrollierten Fehler oder leeren Output liefern."""
+    with pytest.raises((ValueError, AssertionError, ZeroDivisionError, KeyError)):
+        _ = core.analyze_max_outliers([])
+
+
+def test_invariance_same_input():
+    """Gleiche Eingaben müssen deterministische Ergebnisse liefern (idempotent)."""
+    d = [
+        {'corr_id': 'x1', 'stratum': 's1', 'job_parallelism': 2, 'expires_at_dist_hours': 1.0, 'retry_total_overhead_ms': 5.0, 'latency_max': 120.0},
+        {'corr_id': 'x2', 'stratum': 's1', 'job_parallelism': 3, 'expires_at_dist_hours': 2.0, 'retry_total_overhead_ms': 8.0, 'latency_max': 150.0},
+    ]
+    res1 = core.analyze_max_outliers(d)
+    res2 = core.analyze_max_outliers(d)
+    assert res1 == res2
\ No newline at end of file