diff --git a/latency_calculator/src/latency_calculator/core.py b/latency_calculator/src/latency_calculator/core.py
new file mode 100644
index 0000000..e57284f
--- /dev/null
+++ b/latency_calculator/src/latency_calculator/core.py
@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass
+from datetime import datetime
+from typing import Dict, Any, Union
+import pandas as pd
+import logging
+
+logger = logging.getLogger(__name__)
+logging.basicConfig(level=logging.INFO)
+
+
+@dataclass
+class TimestampEntry:
+    t_publish: datetime
+    t_gate_read: datetime
+    t_index_visible: datetime
+
+
+@dataclass
+class LatencyStatistics:
+    p50: float
+    p95: float
+    max: float
+
+
+def _parse_timestamp(value: Union[str, int, float, datetime]) -> datetime:
+    if isinstance(value, datetime):
+        return value
+    if isinstance(value, (int, float)):
+        return datetime.fromtimestamp(value)
+    if isinstance(value, str):
+        try:
+            return datetime.fromisoformat(value)
+        except ValueError:
+            # support timestamps with Z
+            if value.endswith('Z'):
+                try:
+                    return datetime.fromisoformat(value.replace('Z', '+00:00'))
+                except ValueError as e:
+                    raise ValueError(f"Ungültiges Zeitformat: {value}") from e
+            raise ValueError(f"Ungültiges Zeitformat: {value}")
+    raise TypeError(f"Ungültiger Zeittyp: {type(value)}")
+
+
+def calculate_latency(timestamp_entry: Dict[str, Any]) -> Dict[str, float]:
+    """Berechnet p50, p95 und maximale Latenzwerte aus einer Menge von Zeitstempeln."""
+    assert isinstance(timestamp_entry, dict), "timestamp_entry muss ein Dictionary sein"
+
+    required_keys = {"t_publish", "t_gate_read", "t_index_visible"}
+    if not required_keys.issubset(timestamp_entry.keys()):
+        missing = required_keys - set(timestamp_entry.keys())
+        raise KeyError(f"Fehlende Keys in timestamp_entry: {missing}")
+
+    # Parse nested timestamp lists or single dicts
+    entries_raw = timestamp_entry.get("entries") if "entries" in timestamp_entry else [timestamp_entry]
+    if not isinstance(entries_raw, list):
+        raise ValueError("timestamp_entry['entries'] muss eine Liste sein")
+
+    logger.info(f"Verarbeite {len(entries_raw)} Zeitstempel-Einträge für Latenzberechnung")
+
+    records = []
+    for entry in entries_raw:
+        try:
+            t_publish = _parse_timestamp(entry["t_publish"])
+            t_gate_read = _parse_timestamp(entry["t_gate_read"])
+            t_index_visible = _parse_timestamp(entry["t_index_visible"])
+            
+            gate_latency = (t_gate_read - t_publish).total_seconds()
+            visible_latency = (t_index_visible - t_publish).total_seconds()
+
+            if gate_latency < 0 or visible_latency < 0:
+                logger.warning(f"Negative Latenz erkannt in {entry}")
+            
+            records.append({
+                "latency_gate": gate_latency,
+                "latency_visible": visible_latency
+            })
+        except Exception as exc:
+            logger.error(f"Fehler beim Verarbeiten des Eintrags {entry}: {exc}")
+
+    if not records:
+        raise ValueError("Keine gültigen Datensätze zur Latenzberechnung")
+
+    df = pd.DataFrame(records)
+
+    # Combine both latency types as one vector for statistics
+    combined_latencies = pd.concat([df["latency_gate"], df["latency_visible"]], ignore_index=True)
+
+    p50 = float(combined_latencies.quantile(0.50))
+    p95 = float(combined_latencies.quantile(0.95))
+    max_latency = float(combined_latencies.max())
+
+    stats = LatencyStatistics(p50=p50, p95=p95, max=max_latency)
+
+    # Validations for CI readiness
+    assert stats.p50 >= 0, "p50 darf nicht negativ sein"
+    assert stats.max >= stats.p95 >= stats.p50, "Werte müssen monoton sein"
+
+    result = {"p50": stats.p50, "p95": stats.p95, "max": stats.max}
+
+    logger.info(f"Berechnete Latenzstatistik: {json.dumps(result, indent=2)}")
+
+    return result