diff --git a/ai-engine/services/betting_brain.py b/ai-engine/services/betting_brain.py
index df187ff..0981a1e 100644
--- a/ai-engine/services/betting_brain.py
+++ b/ai-engine/services/betting_brain.py
@@ -263,7 +263,14 @@ class BettingBrain:
         "OE": -12.0,
     }
 
-    def judge(self, package: Dict[str, Any]) -> Dict[str, Any]:
+    def judge(
+        self,
+        package: Dict[str, Any],
+        ms_real_odds: Optional[Dict[str, float]] = None,
+    ) -> Dict[str, Any]:
+        # V35c: real bookmaker MS odds (from odds_data) for reference rows —
+        # the brain must never display synthetic 1/p "fair odds" as offered.
+        self._ms_real_odds = ms_real_odds if isinstance(ms_real_odds, dict) else {}
         v27_engine = package.get("v27_engine")
         if not isinstance(v27_engine, dict):
             return package
@@ -916,9 +923,13 @@ class BettingBrain:
             prob = self._safe_float(probs.get(pick), 0.0)
             if prob is None or prob <= 0.0:
                 continue
-            implied_odd = round(1.0 / prob, 2) if prob > 0.01 else 0.0
-            ref_odd = existing_odds_by_pick.get(pick) or implied_odd
-            rows[key] = {
+            # V35c: only REAL bookmaker odds may be displayed. The old fallback
+            # showed synthetic fair-odds (1/prob) as "Oran" — users could read
+            # it as an offered price (e.g. X shown at 4.53 while the bulletin
+            # offered ~3.58). No real price → odds 0.0 and the FE renders "-".
+            real = self._safe_float(getattr(self, "_ms_real_odds", {}).get(pick), 0.0) or 0.0
+            ref_odd = existing_odds_by_pick.get(pick) or (real if real > 1.01 else 0.0)
+            row = {
                 "market": "MS",
                 "pick": pick,
                 "probability": round(prob, 4),
@@ -932,6 +943,12 @@ class BettingBrain:
                 "bet_grade": "PASS",
                 "decision_reasons": ["underdog_reference_for_completeness"],
             }
+            if ref_odd > 1.01:
+                # honest economics vs the real price (vig shows as it truly is)
+                row["implied_prob"] = round(1.0 / ref_odd, 4)
+                row["ev_edge"] = round(prob * ref_odd - 1.0, 4)
+                row["edge"] = row["ev_edge"]
+            rows[key] = row
 
     @staticmethod
     def _merge_row(existing: Optional[Dict[str, Any]], incoming: Dict[str, Any]) -> Dict[str, Any]:
diff --git a/ai-engine/services/orchestrator/upper_brain.py b/ai-engine/services/orchestrator/upper_brain.py
index 84ae345..3473fed 100644
--- a/ai-engine/services/orchestrator/upper_brain.py
+++ b/ai-engine/services/orchestrator/upper_brain.py
@@ -60,8 +60,23 @@ from models.calibration import get_calibrator
 
 
 class UpperBrainMixin:
-    def _apply_upper_brain_guards(self, package: Dict[str, Any]) -> Dict[str, Any]:
-        return BettingBrain().judge(package)
+    def _apply_upper_brain_guards(
+        self, package: Dict[str, Any], data: Any = None
+    ) -> Dict[str, Any]:
+        # V35c: hand the brain the REAL bookmaker MS odds so its reference rows
+        # can never display synthetic 1/p prices as if they were offered.
+        ms_real_odds = None
+        if data is not None:
+            try:
+                odds = getattr(data, "odds_data", None) or {}
+                ms_real_odds = {
+                    "1": self._real_market_odds(odds, "ms_h"),
+                    "X": self._real_market_odds(odds, "ms_d"),
+                    "2": self._real_market_odds(odds, "ms_a"),
+                }
+            except Exception:
+                ms_real_odds = None
+        return BettingBrain().judge(package, ms_real_odds=ms_real_odds)
 
         v27_engine = package.get("v27_engine")
         if not isinstance(v27_engine, dict) or not v27_engine.get("triple_value"):
diff --git a/ai-engine/services/single_match_orchestrator.py b/ai-engine/services/single_match_orchestrator.py
index f2c0b44..79d3a0c 100755
--- a/ai-engine/services/single_match_orchestrator.py
+++ b/ai-engine/services/single_match_orchestrator.py
@@ -668,7 +668,28 @@ class SingleMatchOrchestrator(
                 base_package.setdefault("analysis_details", {})
                 base_package["analysis_details"]["v27_loaded"] = False
 
-        base_package = self._apply_upper_brain_guards(base_package)
+        base_package = self._apply_upper_brain_guards(base_package, data)
+
+        # V35c: the brain rebuilt main/value/supporting/bet_summary AFTER the
+        # market anchor ran inside _build_prediction_package — re-stamp the
+        # calibrated display fields (Güven/CI/Model%/edge) so they stay
+        # consistent, BEFORE the commentary reads the package.
+        self._apply_anchor_to_picks(
+            base_package.get("market_board") or {},
+            base_package.get("main_pick"),
+            base_package.get("value_pick"),
+            base_package.get("aggressive_pick"),
+            base_package.get("supporting_picks"),
+            base_package.get("bet_summary"),
+        )
+        _mp = base_package.get("main_pick")
+        _advice = base_package.get("bet_advice")
+        if isinstance(_mp, dict) and isinstance(_advice, dict) and _mp.get("confidence_band"):
+            _advice["confidence_band"] = _mp["confidence_band"]
+        # no fabricated value bets: a value pick must carry measured positive edge
+        _vp = base_package.get("value_pick")
+        if isinstance(_vp, dict) and float(_vp.get("ev_edge", 0.0) or 0.0) <= 0.0:
+            base_package["value_pick"] = None
 
         # ── Match Commentary: human-readable summary ──────────────
         try: