diff --git a/llm-api/plan_router.py b/llm-api/plan_router.py
index 339da1d..24dc743 100644
--- a/llm-api/plan_router.py
+++ b/llm-api/plan_router.py
@@ -1,18 +1,9 @@
 # -*- coding: utf-8 -*-
 """
-plan_router.py – v0.12.6 (WP-15)
+plan_router.py – v0.13.0 (WP-15)
 
-Minimal-CRUD für Plan-Templates & Pläne (POST/GET) + Idempotenz via Fingerprint
-+ List-/Filter-Endpoints – robust & dokumentiert.
-
-Änderungen ggü. v0.12.4:
-- FIX: POST /plan materialisiert jetzt `plan_section_names` (dedupliziert, getrimmt, sortiert)
-- FIX: GET /plans filtert ausschließlich über `plan_section_names` (kein verschachteltes `sections.name` mehr)
-- Doku: Swagger-Descriptions/Beispiele für alle neuen Query-Parameter
-- Beibehalt: optionale Strict-Validierung von Exercises via ENV `PLAN_STRICT_EXERCISES`
-- Beibehalt: Referenz-Validierung `template_id` → 422, wenn unbekannt
-
-Voraussetzung: KEYWORD-Index für `plans.plan_section_names` durch bootstrap-script (v1.2.x)
+Minimal-CRUD + List/Filter für Templates & Pläne.
+Fix: Zeitfenster-Filter per Qdrant-Range über `created_at_ts` (FLOAT).
 """
 from fastapi import APIRouter, HTTPException, Query
 from pydantic import BaseModel, Field
@@ -24,7 +15,10 @@ import json
 import os
 
 from clients import model, qdrant
-from qdrant_client.models import PointStruct, Filter, FieldCondition, MatchValue, VectorParams, Distance
+from qdrant_client.models import (
+    PointStruct, Filter, FieldCondition, MatchValue,
+    VectorParams, Distance, Range
+)
 
 router = APIRouter(tags=["plans"])
 
@@ -105,7 +99,6 @@ class PlanList(BaseModel):
 # -----------------
 
 def _ensure_collection(name: str):
-    """Legt Collection an, wenn sie fehlt (analog exercise_router)."""
     if not qdrant.collection_exists(name):
         qdrant.recreate_collection(
             collection_name=name,
@@ -114,7 +107,6 @@ def _ensure_collection(name: str):
 
 
 def _norm_list(xs: List[str]) -> List[str]:
-    """Trimmen, casefolded deduplizieren, stabil sortieren."""
     seen, out = set(), []
     for x in xs or []:
         s = str(x).strip()
@@ -144,7 +136,6 @@ def _embed(text: str):
 
 
 def _fingerprint_for_plan(p: Plan) -> str:
-    """sha256(title, total_minutes, sections.items.exercise_external_id, sections.items.duration)"""
     core = {
         "title": p.title,
         "total_minutes": int(p.total_minutes),
@@ -169,7 +160,6 @@ def _get_by_field(collection: str, key: str, value: Any) -> Optional[Dict[str, A
 
 
 def _as_model(model_cls, payload: Dict[str, Any]):
-    """Filtert unbekannte Payload-Felder heraus (Pydantic v1/v2 kompatibel)."""
     fields = getattr(model_cls, "model_fields", None) or getattr(model_cls, "__fields__", {})
     allowed = set(fields.keys())
     data = {k: payload[k] for k in payload.keys() if k in allowed}
@@ -337,8 +327,26 @@ def create_plan(p: Plan):
     # Normalisieren + Materialisierung
     p.goals = _norm_list(p.goals)
     payload = p.model_dump()
-    if isinstance(payload.get("created_at"), datetime):
-        payload["created_at"] = payload["created_at"].astimezone(timezone.utc).isoformat()
+
+    # created_at → ISO + numerischer Zeitstempel (FLOAT)
+    dt = payload.get("created_at")
+    if isinstance(dt, datetime):
+        dt = dt.astimezone(timezone.utc).isoformat()
+    elif isinstance(dt, str):
+        # sicherheitshalber nach UTC normalisieren
+        try:
+            _ = datetime.fromisoformat(dt.replace("Z", "+00:00"))
+        except Exception:
+            dt = datetime.now(timezone.utc).isoformat()
+    else:
+        dt = datetime.now(timezone.utc).isoformat()
+    payload["created_at"] = dt
+    try:
+        ts = datetime.fromisoformat(dt.replace("Z", "+00:00")).timestamp()
+    except Exception:
+        ts = datetime.now(timezone.utc).timestamp()
+    payload["created_at_ts"] = float(ts)
+
     # Materialisierte Section-Namen für robuste Filter/Indizes
     try:
         payload["plan_section_names"] = _norm_list([
@@ -382,7 +390,7 @@ def get_plan(plan_id: str):
         "**Filter** (exakte Matches, KEYWORD-Felder):\n"
         "- created_by, discipline, age_group, target_group, goal\n"
         "- section: Section-Name (nutzt materialisiertes `plan_section_names`)\n"
-        "- created_from / created_to: ISO-8601 Zeitfenster (lokal ausgewertet).\n\n"
+        "- created_from / created_to: ISO-8601 Zeitfenster → serverseitiger Range-Filter über `created_at_ts` (FLOAT).\n\n"
         "**Pagination:** limit/offset. Feld `count` entspricht der Anzahl zurückgegebener Items (keine Gesamtsumme)."
     ),
 )
@@ -413,12 +421,24 @@ def list_plans(
     if section:
         must.append(FieldCondition(key="plan_section_names", match=MatchValue(value=section)))
 
+    # Range-Filter über numerisches Feld (FLOAT)
+    range_args: Dict[str, float] = {}
+    try:
+        if created_from:
+            range_args["gte"] = float(datetime.fromisoformat(created_from.replace("Z", "+00:00")).timestamp())
+        if created_to:
+            range_args["lte"] = float(datetime.fromisoformat(created_to.replace("Z", "+00:00")).timestamp())
+    except Exception:
+        range_args = {}
+    if range_args:
+        must.append(FieldCondition(key="created_at_ts", range=Range(**range_args)))
+
     flt = Filter(must=must or None) if must else None
 
     fetch_n = max(offset + limit, 1)
     pts, _ = qdrant.scroll(collection_name=PLAN_COLLECTION, scroll_filter=flt, limit=fetch_n, with_payload=True)
 
-    # optionales Zeitfenster lokal anwenden
+    # Fallback: lokaler Zeitfilter (für Alt-Daten ohne created_at_ts)
     def _in_window(py: Dict[str, Any]) -> bool:
         if not (created_from or created_to):
             return True
@@ -456,4 +476,4 @@ def list_plans(
 
     sliced = payloads[offset:offset+limit]
     items = [_as_model(Plan, x) for x in sliced]
-    return PlanList(items=items, limit=limit, offset=offset, count=len(items))
+    return PlanList(items=items, limit=limit, offset=offset, count=len(items))
\ No newline at end of file