diff --git a/results_analysis/src/results_analysis/core.py b/results_analysis/src/results_analysis/core.py
new file mode 100644
index 0000000..42e439b
--- /dev/null
+++ b/results_analysis/src/results_analysis/core.py
@@ -0,0 +1,76 @@
+from __future__ import annotations
+from dataclasses import dataclass, asdict
+from typing import List, Dict
+import pandas as pd
+import logging
+
+
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.INFO)
+
+
+class ResultsAnalysisError(Exception):
+    """Custom exception for results analysis errors."""
+
+
+@dataclass
+class ConfusionMatrix:
+    """Repräsentation einer Confusion-Matrix zum Vergleich zweier Policy-Versionen."""
+    TP: int
+    TN: int
+    FP: int
+    FN: int
+
+    def to_json(self) -> Dict[str, int]:
+        """Serialisiert die Matrix in ein JSON-kompatibles Format."""
+        return asdict(self)
+
+
+def _validate_results(results: List[Dict]) -> None:
+    if not isinstance(results, list):
+        raise ResultsAnalysisError("Results must be a list of dicts.")
+    required_keys = {"run_id", "decision"}
+    for item in results:
+        if not isinstance(item, dict):
+            raise ResultsAnalysisError("Each result entry must be a dictionary.")
+        if not required_keys.issubset(item.keys()):
+            raise ResultsAnalysisError(f"Missing required keys in result: {item}")
+
+
+def create_confusion_matrix(previous_results: List[Dict], new_results: List[Dict]) -> Dict[str, int]:
+    """Erzeugt eine Confusion-Matrix durch Vergleich von früheren und neuen Policy-Ergebnissen."""
+    _validate_results(previous_results)
+    _validate_results(new_results)
+
+    prev_df = pd.DataFrame(previous_results)
+    new_df = pd.DataFrame(new_results)
+
+    if prev_df.empty or new_df.empty:
+        raise ResultsAnalysisError("Input results must not be empty.")
+
+    merged = pd.merge(prev_df, new_df, on="run_id", suffixes=("_prev", "_new"))
+
+    TP = int(((merged.decision_prev == "PASS") & (merged.decision_new == "PASS")).sum())
+    TN = int(((merged.decision_prev != "PASS") & (merged.decision_new != "PASS")).sum())
+    FP = int(((merged.decision_prev != "PASS") & (merged.decision_new == "PASS")).sum())
+    FN = int(((merged.decision_prev == "PASS") & (merged.decision_new != "PASS")).sum())
+
+    cm = ConfusionMatrix(TP=TP, TN=TN, FP=FP, FN=FN)
+    logger.info("Confusion matrix created: %s", cm)
+    return cm.to_json()
+
+
+def get_deltas(previous_results: List[Dict], new_results: List[Dict]) -> List[Dict]:
+    """Ermittelt alle Fälle, in denen sich das Ergebnisverhalten zwischen Versionen geändert hat."""
+    _validate_results(previous_results)
+    _validate_results(new_results)
+
+    prev_df = pd.DataFrame(previous_results)
+    new_df = pd.DataFrame(new_results)
+
+    merged = pd.merge(prev_df, new_df, on="run_id", suffixes=("_prev", "_new"))
+    changed = merged[merged.decision_prev != merged.decision_new]
+
+    deltas = changed[["run_id", "decision_prev", "decision_new"]].to_dict(orient="records")
+    logger.info("Detected %d changed results.", len(deltas))
+    return deltas