diff --git a/src/logic_network_generator.py b/src/logic_network_generator.py
index 0285445..035b1be 100755
--- a/src/logic_network_generator.py
+++ b/src/logic_network_generator.py
@@ -1588,6 +1588,7 @@ def create_pathway_logic_network(
     decomposed_uid_mapping: pd.DataFrame,
     reaction_connections: pd.DataFrame,
     best_matches: Any,
+    diagram_bridge_pairs: Optional[Set[Tuple[str, str]]] = None,
 ) -> PathwayResult:
     """Create a pathway logic network from decomposed UID mappings and reaction connections.
 
@@ -1810,6 +1811,40 @@ def create_pathway_logic_network(
                 "edge_reaction_id": reaction_stid,
             })
 
+    # Phase 3b (optional): additive diagram bridges. For each diagram-drawn
+    # (producer, consumer) reaction pair, connect their shared entity by ADDING
+    # an edge from the producer's output-copy to the consumer's input-copy
+    # instead of MERGING the two copies (Phase 2). This preserves every node and
+    # edge — nothing is unified or dropped — while still giving the
+    # producer -> entity_out -> entity_in -> consumer path. Pairs already merged
+    # via precedingEvent (same canonical uuid) are skipped. Layout-filtered pairs
+    # only (cofactor hubs already excluded by the caller). See #39.
+    if diagram_bridge_pairs:
+        n_bridges = 0
+        for a_rid, b_rid in diagram_bridge_pairs:
+            for p_vr in reactome_to_vr.get(a_rid, []):
+                p_outputs = set(vr_entities.get(p_vr, ([], [], {}, {}))[1])
+                for f_vr in reactome_to_vr.get(b_rid, []):
+                    f_inputs = set(vr_entities.get(f_vr, ([], [], {}, {}))[0])
+                    for eid in p_outputs & f_inputs:
+                        src = entity_uuid_registry.get((eid, p_vr, "output"))
+                        tgt = entity_uuid_registry.get((eid, f_vr, "input"))
+                        # Skip if missing or already the same node (already
+                        # connected via a precedingEvent merge).
+                        if not src or not tgt or src == tgt:
+                            continue
+                        pathway_logic_network_data.append({
+                            "source_id": src,
+                            "target_id": tgt,
+                            "pos_neg": "pos",
+                            "and_or": None,
+                            "edge_type": "diagram_bridge",
+                            "stoichiometry": 1,
+                            "edge_reaction_id": None,
+                        })
+                        n_bridges += 1
+        logger.info(f"Diagram bridges: +{n_bridges} additive edges (no merge)")
+
     # Log UUID registry statistics
     unique_uuids = set(entity_uuid_registry.values())
     unique_entities = set(key[0] for key in entity_uuid_registry.keys())
diff --git a/src/pathway_generator.py b/src/pathway_generator.py
index adcd1bd..6510da8 100755
--- a/src/pathway_generator.py
+++ b/src/pathway_generator.py
@@ -122,17 +122,34 @@ def generate_pathway_file(
                 logger.warning(f"Could not cache decomposition results: {e}")
                 # Continue without caching
 
-        # Augment connectivity with curator-drawn diagram flow (product->substrate
-        # pairs the diagram links but precedingEvent may omit — esp. old pathways).
-        # Only the connectivity/merge step sees this; the matching layer above
-        # stays on pure precedingEvent. See reactome/logic-network-generator#39.
-        from src.diagram_connectivity import augment_reaction_connections
-        connectivity = augment_reaction_connections(pathway_id, reaction_connections)
+        # Add curator-drawn diagram connectivity (product->substrate pairs the
+        # diagram links but precedingEvent may omit). See reactome/logic-network-generator#39.
+        #
+        # DEFAULT = additive-bridge mode (LNG_DIAGRAM_BRIDGE=1): the diagram-drawn
+        # shared-entity connections are added as NEW bridge edges (producer's
+        # output-copy -> consumer's input-copy) — nothing is merged or lost. Raw
+        # precedingEvent connectivity still merges its shared entities (the core
+        # mechanism), but the diagram json NEVER drives a merge. Benchmark-neutral
+        # vs the legacy merge (curator 83.2%, experimental 73.6%; TP53 +2) while
+        # preserving every node and edge. Set LNG_DIAGRAM_BRIDGE=0 to fall back to
+        # the legacy merge-based augmentation; set both LNG_DIAGRAM_BRIDGE=0 and
+        # LNG_DIAGRAM_CONNECTIVITY=0 to disable diagram connectivity entirely.
+        from src.diagram_connectivity import (
+            augment_reaction_connections,
+            diagram_shared_product_pairs,
+        )
+        diagram_bridge_pairs = None
+        if os.environ.get("LNG_DIAGRAM_BRIDGE", "1") == "1":
+            connectivity = reaction_connections
+            diagram_bridge_pairs = diagram_shared_product_pairs(pathway_id)
+        else:
+            connectivity = augment_reaction_connections(pathway_id, reaction_connections)
 
         # Generate logic network
         logger.info("Creating pathway logic network...")
         result = create_pathway_logic_network(
-            decomposed_uid_mapping, connectivity, best_matches
+            decomposed_uid_mapping, connectivity, best_matches,
+            diagram_bridge_pairs=diagram_bridge_pairs,
         )
 
         # Save logic network (main output file users need)