backend/placeholder_registry.py

@@ -260,21 +260,34 @@ class PlaceholderRegistry:
 
 def build_ai_placeholder_caption(metadata: PlaceholderMetadata, max_len: int = 400) -> str:
     """
-    Kurztext für KI-Kontext (z. B. Modifier |d): Bedeutung/Skala, ohne die Rohausgabe zu ersetzen.
-    Nutzt business_meaning / semantic_contract; bei Scores explizite 0–100-Erläuterung.
+    Text für |d und Exportfeld ai_caption: zuerst **was** der Platzhalter misst (description),
+    dann **Einordnung** (business_meaning oder gekürzter semantic_contract).
+    So ist klar, worauf sich der konkrete Wert bezieht — nicht nur eine „Meta-Bedeutung“.
     """
-    chunks: List[str] = []
+    desc = (metadata.description or "").strip()
     bm = (metadata.business_meaning or "").strip()
     sc = (metadata.semantic_contract or "").strip()
-    desc = (metadata.description or "").strip()
 
-    if bm:
-        chunks.append(bm)
-    elif sc:
-        chunks.append(sc if len(sc) <= max_len else sc[: max_len - 1] + "…")
-    elif desc:
+    chunks: List[str] = []
+    if desc:
         chunks.append(desc)
 
+    interpret = bm
+    if not interpret and sc:
+        interpret = sc if len(sc) <= max_len else sc[: max_len - 1] + "…"
+
+    if interpret:
+        blob = " ".join(chunks).lower()
+        il = interpret.lower()
+        # Keine Dublette: gleicher Text oder lange Description bereits in der Interpretation
+        redundant = il in blob or (
+            desc
+            and len(desc) >= 10
+            and desc.lower() in il
+        )
+        if not redundant:
+            chunks.append(interpret)
+
     if metadata.placeholder_type == PlaceholderType.SCORE:
         chunks.append("Skala 0–100: höher = im Modell günstiger / besser abgestimmt.")
 
@@ -282,9 +295,18 @@ def build_ai_placeholder_caption(metadata: PlaceholderMetadata, max_len: int = 4
     if unit and metadata.placeholder_type != PlaceholderType.SCORE:
         blob = " ".join(chunks).lower()
         u_low = unit.lower()
-        if u_low not in blob and u_low.replace(" ", "") not in blob.replace(" ", ""):
-            if u_low not in ("score (0-100)", "0-100", "0–100", "dimensionless"):
-                chunks.append(f"Technischer Bezug: {unit}.")
+        # Einheit oft schon in description („… in g (30d)“, „Kalorien“) — nicht doppeln
+        compact_blob = blob.replace(" ", "").replace("/", "")
+        compact_u = u_low.replace(" ", "").replace("/", "")
+        unit_redundant = compact_u in compact_blob or (
+            "g/day" in u_low and ("g/" in blob or "gramm" in blob or " protein" in blob or " fett" in blob or " kh" in blob)
+        ) or ("kcal" in u_low and ("kcal" in blob or "kalorien" in blob))
+
+        if (
+            not unit_redundant
+            and u_low not in ("score (0-100)", "0-100", "0–100", "dimensionless")
+        ):
+            chunks.append(f"Technischer Bezug: {unit}.")
 
     out = " ".join(c for c in chunks if c).strip()
     if len(out) > max_len + 120:

backend/placeholder_resolver.py

@@ -10,7 +10,7 @@ This module now focuses on FORMATTING for AI consumption.
 import json
 import re
 from datetime import datetime, timedelta
-from typing import Dict, List, Optional, Callable, Tuple
+from typing import Any, Dict, List, Optional, Callable, Tuple
 from db import get_db, get_cursor, r2d
 
 # Phase 0c: Import data layer
@@ -62,6 +62,17 @@ def _ai_caption_for_placeholder_key(key: str) -> Optional[str]:
     return None
 
 
+def format_value_with_d_modifier(value: str, catalog_row: Dict[str, Any]) -> str:
+    """
+    Entspricht der Prompt-Ersetzung bei {{key|d}}: „Wert — Kontext“.
+    Kontext: ai_caption aus dem Katalog, sonst description (wie prompt_executor).
+    """
+    cap = (catalog_row.get("ai_caption") or catalog_row.get("description") or "").strip()
+    if cap:
+        return f"{value} — {cap}"
+    return str(value)
+
+
 # ── Helper Functions ──────────────────────────────────────────────────────────
 
 def get_profile_data(profile_id: str) -> Dict:

backend/routers/prompts.py

@@ -21,6 +21,7 @@ from placeholder_resolver import (
     resolve_placeholders,
     get_unknown_placeholders,
     get_placeholder_example_values,
+    format_value_with_d_modifier,
     get_available_placeholders,
     get_placeholder_catalog
 )
@@ -457,8 +458,8 @@ def export_placeholder_values(session: dict = Depends(require_auth)):
     """
     Export all available placeholders with their current resolved values.
 
-    Returns JSON export suitable for download with all placeholders
-    resolved for the current user's profile.
+    Pro Zeile: value = Rohwert wie bei {{key}}, example = Vorschau wie bei {{key|d}}
+    (Wert — ai_caption bzw. description). JSON-Download für das aktive Profil.
     """
     from datetime import datetime
     profile_id = session['profile_id']
@@ -486,11 +487,12 @@ def export_placeholder_values(session: dict = Depends(require_auth)):
         export_data['placeholders_by_category'][category] = []
         for item in items:
             key = item['key'].replace('{{', '').replace('}}', '')
+            raw_val = cleaned_values.get(key, 'nicht verfügbar')
             row = {
                 'key': item['key'],
                 'description': item['description'],
-                'value': cleaned_values.get(key, 'nicht verfügbar'),
-                'example': item.get('example'),
+                'value': raw_val,
+                'example': format_value_with_d_modifier(str(raw_val), item),
             }
             if item.get('ai_caption'):
                 row['ai_caption'] = item['ai_caption']
@@ -662,11 +664,12 @@ def export_placeholder_values_extended(
         export_data['legacy']['placeholders_by_category'][category] = []
         for item in items:
             key = item['key'].replace('{{', '').replace('}}', '')
+            raw_val = cleaned_values.get(key, 'nicht verfügbar')
             export_data['legacy']['placeholders_by_category'][category].append({
                 'key': item['key'],
                 'description': item['description'],
-                'value': cleaned_values.get(key, 'nicht verfügbar'),
-                'example': item.get('example')
+                'value': raw_val,
+                'example': format_value_with_d_modifier(str(raw_val), item),
             })
 
     # Fill metadata flat

backend/tests/test_placeholder_modifier_d.py

@@ -6,6 +6,7 @@ from placeholder_registry import (
     build_ai_placeholder_caption,
 )
 import placeholder_resolver as pr
+from placeholder_resolver import format_value_with_d_modifier
 
 
 def test_build_ai_caption_prefers_business_meaning():
@@ -22,9 +23,28 @@ def test_build_ai_caption_prefers_business_meaning():
         output_type=OutputType.NUMERIC,
     )
     cap = build_ai_placeholder_caption(m)
+    assert cap.startswith("Kurzbeschreibung")
     assert "Kernbedeutung" in cap
 
 
+def test_build_ai_caption_description_then_meaning_like_protein_avg():
+    m = PlaceholderMetadata(
+        key="protein_avg",
+        category="Ernährung",
+        description="Durchschn. Protein in g (30d)",
+        resolver_module="m",
+        resolver_function="f",
+        business_meaning="Zentraler Placeholder für Muskelerhalt.",
+        unit="g/day",
+        placeholder_type=PlaceholderType.INTERPRETED,
+        output_type=OutputType.NUMERIC,
+    )
+    cap = build_ai_placeholder_caption(m)
+    assert cap.startswith("Durchschn. Protein in g (30d)")
+    assert "Muskelerhalt" in cap
+    assert "Technischer Bezug" not in cap
+
+
 def test_build_ai_caption_score_adds_scale():
     m = PlaceholderMetadata(
         key="test_score",
@@ -54,3 +74,19 @@ def test_placeholder_token_regex_optional_modifier():
 def test_get_unknown_placeholders_strips_modifier():
     unk = pr.get_unknown_placeholders("{{not_a_real_key|d}}")
     assert set(unk) == {"not_a_real_key"}
+
+
+def test_format_value_with_d_modifier_matches_prompt_executor():
+    row = {
+        "key": "protein_avg",
+        "description": "Durchschn. Protein in g (30d)",
+        "example": "119g/Tag",
+        "ai_caption": "Durchschn. Protein in g (30d). Zentral für Muskelerhalt.",
+    }
+    out = format_value_with_d_modifier("119g/Tag", row)
+    assert out == "119g/Tag — Durchschn. Protein in g (30d). Zentral für Muskelerhalt."
+
+
+def test_format_value_with_d_modifier_falls_back_to_description():
+    row = {"description": "Nur Beschreibung", "key": "x"}
+    assert format_value_with_d_modifier("42", row) == "42 — Nur Beschreibung"