diff --git a/max_outlier_analysis_script/src/max_outlier_analysis_script/core.py b/max_outlier_analysis_script/src/max_outlier_analysis_script/core.py
new file mode 100644
index 0000000..c384673
--- /dev/null
+++ b/max_outlier_analysis_script/src/max_outlier_analysis_script/core.py
@@ -0,0 +1,110 @@
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+from typing import List, Dict, Any
+import pandas as pd
+import statistics
+
+__all__ = ["analyze_max_outliers"]
+
+
+# Configure logging for CI diagnostics
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.INFO)
+
+
+class DataValidationError(Exception):
+    """Raised when input data validation fails."""
+    pass
+
+
+@dataclass
+class OutlierRecord:
+    corr_id: str
+    stratum: str
+    job_parallelism: int
+    expires_at_dist_hours: float
+    retry_total_overhead_ms: float
+    latency_max: float
+
+
+@dataclass
+class AnalysisResults:
+    max_above_p99_count: int
+    near_expiry_cluster_percentage: float
+    retry_overhead_variance: float
+
+
+_DEF_NEAR_EXPIRY_KEY = "near-expiry-unpinned"
+
+
+def _validate_data(data: List[Dict[str, Any]]) -> None:
+    required_fields = {f.name for f in OutlierRecord.__dataclass_fields__.values()}
+    if not isinstance(data, list):
+        raise DataValidationError("Input data must be a list of dictionaries.")
+    for i, record in enumerate(data):
+        if not isinstance(record, dict):
+            raise DataValidationError(f"Element {i} is not a dict.")
+        missing = required_fields - record.keys()
+        if missing:
+            raise DataValidationError(f"Record {i} missing fields: {missing}")
+
+
+
+def analyze_max_outliers(data: List[Dict[str, Any]]) -> Dict[str, Any]:
+    """Analysiert Max-Outlier-Daten und berechnet Summary-Kennzahlen.
+
+    Args:
+        data: Liste von Messwert-Dictionaries aus CI-Lasttest-Exports.
+
+    Returns:
+        dict: Ergebnisse mit Kennzahlen für Max-Ausreißer.
+    """
+    logger.info("Starting max outlier analysis on %d records", len(data))
+    _validate_data(data)
+
+    df = pd.DataFrame(data)
+    if df.empty:
+        return {
+            "max_above_p99_count": 0,
+            "near_expiry_cluster_percentage": 0.0,
+            "retry_overhead_variance": 0.0,
+        }
+
+    # Compute percentiles
+    p99 = df["latency_max"].quantile(0.99)
+    above_p99 = df[df["latency_max"] > p99]
+    max_above_p99_count = int(above_p99.shape[0])
+
+    # Cluster analysis for 'near-expiry-unpinned'
+    total_count = df.shape[0]
+    near_expiry_count = df[df["stratum"] == _DEF_NEAR_EXPIRY_KEY].shape[0]
+    near_expiry_cluster_percentage = (
+        (near_expiry_count / total_count) * 100.0 if total_count > 0 else 0.0
+    )
+
+    # Retry overhead variance
+    try:
+        retry_overhead_variance = statistics.variance(df["retry_total_overhead_ms"].tolist())
+    except statistics.StatisticsError:
+        retry_overhead_variance = 0.0
+
+    result = AnalysisResults(
+        max_above_p99_count=max_above_p99_count,
+        near_expiry_cluster_percentage=near_expiry_cluster_percentage,
+        retry_overhead_variance=retry_overhead_variance,
+    )
+
+    logger.info(
+        "Analysis complete: max_above_p99_count=%d, near_expiry_cluster_percentage=%.2f, retry_overhead_variance=%.3f",
+        result.max_above_p99_count,
+        result.near_expiry_cluster_percentage,
+        result.retry_overhead_variance,
+    )
+
+    return {
+        "max_above_p99_count": result.max_above_p99_count,
+        "near_expiry_cluster_percentage": result.near_expiry_cluster_percentage,
+        "retry_overhead_variance": result.retry_overhead_variance,
+    }