File size: 45,789 Bytes
d82945e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
"""
Cogni-Engine v1 — Knowledge Graph Engine
In-memory graph structure with nodes, edges, traversal, similarity search.
This is the core data structure that represents all knowledge.
The "brain matter" — where concepts live and connect.
"""

import time
import threading
import json
from typing import List, Dict, Optional, Set, Tuple, Any
from collections import defaultdict

import numpy as np

import config
import utils
from memory import Memory


# ═══════════════════════════════════════════════════════════
# DATA STRUCTURES
# ═══════════════════════════════════════════════════════════

class Node:
    """A single knowledge node in the graph."""

    __slots__ = [
        'id', 'type', 'content', 'vector', 'weight',
        'connections', 'source', 'created_at', 'updated_at',
        '_dirty'
    ]

    def __init__(
        self,
        node_id: str,
        node_type: str,
        content: str,
        vector: np.ndarray = None,
        weight: float = 1.0,
        connections: int = 0,
        source: str = "data",
        created_at: str = "",
        updated_at: str = ""
    ):
        self.id = node_id
        self.type = node_type
        self.content = content
        self.vector = vector if vector is not None else np.zeros(config.VECTOR_DIM, dtype=np.float32)
        self.weight = weight
        self.connections = connections
        self.source = source
        self.created_at = created_at or utils.timestamp_now()
        self.updated_at = updated_at or utils.timestamp_now()
        self._dirty = False

    def to_dict(self) -> dict:
        """Serialize to dict for DB storage."""
        return {
            "id": self.id,
            "type": self.type,
            "content": self.content,
            "vector": utils.vector_to_list(self.vector),
            "weight": round(self.weight, 6),
            "connections": self.connections,
            "source": self.source,
            "created_at": self.created_at,
            "updated_at": self.updated_at
        }

    @staticmethod
    def from_dict(data: dict) -> 'Node':
        """Deserialize from dict."""
        vector = None
        if data.get("vector"):
            vector = utils.list_to_vector(data["vector"])
        return Node(
            node_id=data["id"],
            node_type=data.get("type", "fact"),
            content=data.get("content", ""),
            vector=vector,
            weight=float(data.get("weight", 1.0)),
            connections=int(data.get("connections", 0)),
            source=data.get("source", "data"),
            created_at=data.get("created_at", ""),
            updated_at=data.get("updated_at", "")
        )

    def mark_dirty(self):
        """Mark this node as needing DB sync."""
        self._dirty = True
        self.updated_at = utils.timestamp_now()


class Edge:
    """A directed relationship between two nodes."""

    __slots__ = [
        'id', 'from_node', 'to_node', 'relation', 'weight',
        'confidence', 'source', 'used_count', 'created_at',
        '_dirty'
    ]

    def __init__(
        self,
        edge_id: str,
        from_node: str,
        to_node: str,
        relation: str = "related_to",
        weight: float = 1.0,
        confidence: float = 1.0,
        source: str = "data",
        used_count: int = 0,
        created_at: str = ""
    ):
        self.id = edge_id
        self.from_node = from_node
        self.to_node = to_node
        self.relation = relation
        self.weight = weight
        self.confidence = confidence
        self.source = source
        self.used_count = used_count
        self.created_at = created_at or utils.timestamp_now()
        self._dirty = False

    def to_dict(self) -> dict:
        """Serialize to dict for DB storage."""
        return {
            "id": self.id,
            "from_node": self.from_node,
            "to_node": self.to_node,
            "relation": self.relation,
            "weight": round(self.weight, 6),
            "confidence": round(self.confidence, 6),
            "source": self.source,
            "used_count": self.used_count,
            "created_at": self.created_at
        }

    @staticmethod
    def from_dict(data: dict) -> 'Edge':
        """Deserialize from dict."""
        return Edge(
            edge_id=data["id"],
            from_node=data["from_node"],
            to_node=data["to_node"],
            relation=data.get("relation", "related_to"),
            weight=float(data.get("weight", 1.0)),
            confidence=float(data.get("confidence", 1.0)),
            source=data.get("source", "data"),
            used_count=int(data.get("used_count", 0)),
            created_at=data.get("created_at", "")
        )

    def mark_dirty(self):
        """Mark edge as needing DB sync."""
        self._dirty = True


class ReasoningChain:
    """A discovered path of reasoning through the graph."""

    __slots__ = [
        'id', 'path', 'conclusion', 'confidence',
        'used_count', 'created_at'
    ]

    def __init__(
        self,
        chain_id: str,
        path: list,
        conclusion: str = "",
        confidence: float = 0.5,
        used_count: int = 0,
        created_at: str = ""
    ):
        self.id = chain_id
        self.path = path  # [node_id, edge_id, node_id, edge_id, ...]
        self.conclusion = conclusion
        self.confidence = confidence
        self.used_count = used_count
        self.created_at = created_at or utils.timestamp_now()

    def to_dict(self) -> dict:
        return {
            "id": self.id,
            "path": self.path,
            "conclusion": self.conclusion,
            "confidence": round(self.confidence, 6),
            "used_count": self.used_count,
            "created_at": self.created_at
        }

    @staticmethod
    def from_dict(data: dict) -> 'ReasoningChain':
        return ReasoningChain(
            chain_id=data["id"],
            path=data.get("path", []),
            conclusion=data.get("conclusion", ""),
            confidence=float(data.get("confidence", 0.5)),
            used_count=int(data.get("used_count", 0)),
            created_at=data.get("created_at", "")
        )


# ═══════════════════════════════════════════════════════════
# KNOWLEDGE GRAPH
# ═══════════════════════════════════════════════════════════

class KnowledgeGraph:
    """
    In-memory knowledge graph with persistence via Memory.

    Structure:
    - nodes: dict of Node objects indexed by id
    - edges: dict of Edge objects indexed by id
    - adjacency_out: node_id → list of edge_ids (outgoing)
    - adjacency_in: node_id → list of edge_ids (incoming)
    - vector_index: numpy matrix of all node vectors for fast search
    - chains: dict of ReasoningChain objects

    Thread-safe via read-write lock:
    - Multiple readers allowed simultaneously
    - Writers get exclusive access
    """

    def __init__(self, memory: Memory):
        self.memory = memory

        # Core data
        self.nodes: Dict[str, Node] = {}
        self.edges: Dict[str, Edge] = {}
        self.chains: Dict[str, ReasoningChain] = {}

        # Adjacency indexes
        self._adj_out: Dict[str, List[str]] = defaultdict(list)  # node → [edge_ids outgoing]
        self._adj_in: Dict[str, List[str]] = defaultdict(list)   # node → [edge_ids incoming]

        # Vector index for fast similarity search
        self._vector_matrix: Optional[np.ndarray] = None
        self._vector_node_ids: List[str] = []
        self._vector_index_dirty = True

        # Thread safety
        self._lock = threading.RLock()

        # Stats
        self._stats = {
            "total_nodes": 0,
            "total_edges": 0,
            "total_chains": 0,
            "inferred_nodes": 0,
            "inferred_edges": 0,
            "max_abstraction_depth": 0,
            "avg_connections": 0.0,
            "avg_confidence": 0.0
        }

    # ───────────────────────────────────────────────────
    # INITIALIZATION
    # ───────────────────────────────────────────────────

    def load_from_memory(self) -> bool:
        """
        Load entire graph from TiDB via Memory.
        Called once at startup.
        """
        state = self.memory.load_full_state()

        if not state.get("loaded", False) and not state["nodes"]:
            print("[GRAPH] No existing state found. Starting fresh.")
            self._rebuild_stats()
            return True

        with self._lock:
            # Load nodes
            for node_data in state["nodes"]:
                node = Node.from_dict(node_data)
                self.nodes[node.id] = node

            # Load edges
            for edge_data in state["edges"]:
                edge = Edge.from_dict(edge_data)
                self.edges[edge.id] = edge
                self._adj_out[edge.from_node].append(edge.id)
                self._adj_in[edge.to_node].append(edge.id)

            # Load chains
            for chain_data in state["chains"]:
                chain = ReasoningChain.from_dict(chain_data)
                self.chains[chain.id] = chain

            # Rebuild vector index
            self._rebuild_vector_index()
            self._rebuild_stats()

        print(f"[GRAPH] Loaded: {len(self.nodes)} nodes, "
              f"{len(self.edges)} edges, {len(self.chains)} chains")
        return True

    # ───────────────────────────────────────────────────
    # NODE OPERATIONS
    # ───────────────────────────────────────────────────

    def add_node(
        self,
        content: str,
        node_type: str = "fact",
        source: str = "data",
        weight: float = None,
        vector: np.ndarray = None,
        node_id: str = None,
        tags: List[str] = None
    ) -> Optional[Node]:
        """
        Add a new node to the graph.
        If node with same id exists, update it instead.
        Returns the node, or None if invalid.
        """
        if not content or not content.strip():
            return None

        content = content.strip()

        if node_id is None:
            node_id = config.generate_node_id(content, node_type)

        # Generate vector if not provided
        if vector is None:
            vector = utils.text_to_vector_tfidf(content)

        # Register content with TF-IDF corpus
        tokens = utils.tokenize(content, remove_stopwords=True)
        utils.tfidf.add_document(tokens)

        if weight is None:
            weight = (config.DATA_KNOWLEDGE_CONFIDENCE
                      if source == "data"
                      else config.USER_KNOWLEDGE_CONFIDENCE)

        with self._lock:
            if node_id in self.nodes:
                # Update existing node
                existing = self.nodes[node_id]
                # Reinforce weight if seen again
                existing.weight = min(
                    existing.weight * config.WEIGHT_REINFORCE,
                    config.WEIGHT_MAX
                )
                existing.mark_dirty()
                self.memory.save_node(existing.to_dict())
                return existing

            # Create new node
            node = Node(
                node_id=node_id,
                node_type=node_type,
                content=content,
                vector=vector,
                weight=weight,
                connections=0,
                source=source
            )

            # Safety check
            if len(self.nodes) >= config.MAX_GRAPH_MEMORY_NODES:
                print(f"[GRAPH] Node limit reached ({config.MAX_GRAPH_MEMORY_NODES}). Skipping.")
                return None

            self.nodes[node_id] = node
            self._vector_index_dirty = True

            # Buffer for DB write
            node._dirty = True
            self.memory.save_node(node.to_dict())

            # Create edges from tags
            if tags:
                for tag in tags:
                    tag_id = config.generate_node_id(tag, "concept")
                    if tag_id not in self.nodes:
                        self.add_node(
                            content=tag,
                            node_type="concept",
                            source=source,
                            weight=weight * 0.8
                        )
                    self.add_edge(
                        from_id=node_id,
                        to_id=tag_id,
                        relation="related_to",
                        source=source,
                        confidence=weight * 0.7
                    )

        return node

    def get_node(self, node_id: str) -> Optional[Node]:
        """Get a node by id."""
        return self.nodes.get(node_id)

    def get_node_by_content(self, content: str, node_type: str = "") -> Optional[Node]:
        """Find node by exact content match."""
        node_id = config.generate_node_id(content.strip(), node_type)
        return self.nodes.get(node_id)

    def remove_node(self, node_id: str) -> bool:
        """Remove a node and all its edges."""
        with self._lock:
            if node_id not in self.nodes:
                return False

            # Remove connected edges
            edge_ids_to_remove = []
            edge_ids_to_remove.extend(self._adj_out.get(node_id, []))
            edge_ids_to_remove.extend(self._adj_in.get(node_id, []))

            for edge_id in set(edge_ids_to_remove):
                self._remove_edge_internal(edge_id)

            # Remove adjacency entries
            self._adj_out.pop(node_id, None)
            self._adj_in.pop(node_id, None)

            # Remove node
            del self.nodes[node_id]
            self._vector_index_dirty = True

            # Buffer for DB delete
            self.memory.delete_node(node_id)

        return True

    def update_node_weight(self, node_id: str, new_weight: float):
        """Update a node's weight."""
        with self._lock:
            node = self.nodes.get(node_id)
            if node:
                node.weight = utils.clamp(new_weight, config.WEIGHT_MIN, config.WEIGHT_MAX)
                node.mark_dirty()
                self.memory.save_node(node.to_dict())

    def get_nodes_by_type(self, node_type: str) -> List[Node]:
        """Get all nodes of a specific type."""
        return [n for n in self.nodes.values() if n.type == node_type]

    def get_nodes_by_source(self, source: str) -> List[Node]:
        """Get all nodes from a specific source."""
        return [n for n in self.nodes.values() if n.source == source]

    def get_weakest_nodes(self, limit: int = 50) -> List[Node]:
        """Get nodes with lowest weight (candidates for pruning)."""
        sorted_nodes = sorted(self.nodes.values(), key=lambda n: n.weight)
        return sorted_nodes[:limit]

    def get_least_connected_nodes(self, limit: int = 50) -> List[Node]:
        """Get nodes with fewest connections (candidates for connecting)."""
        sorted_nodes = sorted(self.nodes.values(), key=lambda n: n.connections)
        return sorted_nodes[:limit]

    # ───────────────────────────────────────────────────
    # EDGE OPERATIONS
    # ───────────────────────────────────────────────────

    def add_edge(
        self,
        from_id: str,
        to_id: str,
        relation: str = "related_to",
        weight: float = 1.0,
        confidence: float = 1.0,
        source: str = "data",
        edge_id: str = None
    ) -> Optional[Edge]:
        """
        Add a directed edge between two nodes.
        If edge exists, reinforce it.
        """
        if from_id == to_id:
            return None  # No self-loops

        if from_id not in self.nodes or to_id not in self.nodes:
            return None  # Both nodes must exist

        if edge_id is None:
            edge_id = config.generate_edge_id(from_id, to_id, relation)

        with self._lock:
            if edge_id in self.edges:
                # Reinforce existing edge
                existing = self.edges[edge_id]
                existing.weight = min(
                    existing.weight * config.WEIGHT_REINFORCE,
                    config.WEIGHT_MAX
                )
                existing.confidence = min(
                    (existing.confidence + confidence) / 2.0 * 1.05,
                    1.0
                )
                existing.mark_dirty()
                self.memory.save_edge(existing.to_dict())
                return existing

            # Safety check
            if len(self.edges) >= config.MAX_GRAPH_MEMORY_EDGES:
                print(f"[GRAPH] Edge limit reached ({config.MAX_GRAPH_MEMORY_EDGES}). Skipping.")
                return None

            edge = Edge(
                edge_id=edge_id,
                from_node=from_id,
                to_node=to_id,
                relation=relation,
                weight=weight,
                confidence=confidence,
                source=source
            )

            self.edges[edge_id] = edge
            self._adj_out[from_id].append(edge_id)
            self._adj_in[to_id].append(edge_id)

            # Update connection counts
            self.nodes[from_id].connections += 1
            self.nodes[to_id].connections += 1

            # Buffer for DB
            edge._dirty = True
            self.memory.save_edge(edge.to_dict())

        return edge

    def get_edge(self, edge_id: str) -> Optional[Edge]:
        """Get an edge by id."""
        return self.edges.get(edge_id)

    def get_edges_from(self, node_id: str) -> List[Edge]:
        """Get all outgoing edges from a node."""
        edge_ids = self._adj_out.get(node_id, [])
        return [self.edges[eid] for eid in edge_ids if eid in self.edges]

    def get_edges_to(self, node_id: str) -> List[Edge]:
        """Get all incoming edges to a node."""
        edge_ids = self._adj_in.get(node_id, [])
        return [self.edges[eid] for eid in edge_ids if eid in self.edges]

    def get_all_edges_for(self, node_id: str) -> List[Edge]:
        """Get all edges (in + out) connected to a node."""
        edges = self.get_edges_from(node_id)
        edges.extend(self.get_edges_to(node_id))
        return edges

    def get_neighbors(self, node_id: str) -> List[Tuple[Node, Edge]]:
        """Get all neighboring nodes with their connecting edges."""
        neighbors = []
        for edge in self.get_edges_from(node_id):
            target = self.nodes.get(edge.to_node)
            if target:
                neighbors.append((target, edge))
        for edge in self.get_edges_to(node_id):
            source = self.nodes.get(edge.from_node)
            if source:
                neighbors.append((source, edge))
        return neighbors

    def edge_exists(self, from_id: str, to_id: str, relation: str = None) -> bool:
        """Check if an edge exists between two nodes."""
        for edge_id in self._adj_out.get(from_id, []):
            edge = self.edges.get(edge_id)
            if edge and edge.to_node == to_id:
                if relation is None or edge.relation == relation:
                    return True
        return False

    def remove_edge(self, edge_id: str) -> bool:
        """Remove an edge."""
        with self._lock:
            return self._remove_edge_internal(edge_id)

    def _remove_edge_internal(self, edge_id: str) -> bool:
        """Internal edge removal (must be called under lock)."""
        edge = self.edges.get(edge_id)
        if not edge:
            return False

        # Remove from adjacency
        if edge_id in self._adj_out.get(edge.from_node, []):
            self._adj_out[edge.from_node].remove(edge_id)
        if edge_id in self._adj_in.get(edge.to_node, []):
            self._adj_in[edge.to_node].remove(edge_id)

        # Update connection counts
        from_node = self.nodes.get(edge.from_node)
        to_node = self.nodes.get(edge.to_node)
        if from_node:
            from_node.connections = max(0, from_node.connections - 1)
        if to_node:
            to_node.connections = max(0, to_node.connections - 1)

        # Remove edge
        del self.edges[edge_id]
        self.memory.delete_edge(edge_id)

        return True

    def reinforce_edge(self, edge_id: str, factor: float = None):
        """Increase edge weight (used when edge participates in response)."""
        if factor is None:
            factor = config.WEIGHT_REINFORCE
        with self._lock:
            edge = self.edges.get(edge_id)
            if edge:
                edge.weight = min(edge.weight * factor, config.WEIGHT_MAX)
                edge.used_count += 1
                edge.mark_dirty()
                self.memory.save_edge(edge.to_dict())

    def decay_edge(self, edge_id: str, factor: float = None):
        """Decrease edge weight (unused edge decay)."""
        if factor is None:
            factor = config.WEIGHT_DECAY_RATE
        with self._lock:
            edge = self.edges.get(edge_id)
            if edge:
                edge.weight = max(edge.weight * factor, config.WEIGHT_MIN)
                edge.mark_dirty()
                self.memory.save_edge(edge.to_dict())

    def get_weakest_edges(self, limit: int = 100, source_filter: str = "inferred") -> List[Edge]:
        """Get edges with lowest weight (candidates for pruning)."""
        filtered = [
            e for e in self.edges.values()
            if source_filter is None or e.source == source_filter
        ]
        sorted_edges = sorted(filtered, key=lambda e: e.weight)
        return sorted_edges[:limit]

    # ───────────────────────────────────────────────────
    # VECTOR INDEX & SIMILARITY SEARCH
    # ───────────────────────────────────────────────────

    def _rebuild_vector_index(self):
        """Rebuild the vector matrix for fast batch similarity search."""
        with self._lock:
            if not self.nodes:
                self._vector_matrix = np.zeros((0, config.VECTOR_DIM), dtype=np.float32)
                self._vector_node_ids = []
                self._vector_index_dirty = False
                return

            node_ids = []
            vectors = []
            for nid, node in self.nodes.items():
                if node.vector is not None and len(node.vector) == config.VECTOR_DIM:
                    node_ids.append(nid)
                    vectors.append(node.vector)

            if vectors:
                self._vector_matrix = np.array(vectors, dtype=np.float32)
            else:
                self._vector_matrix = np.zeros((0, config.VECTOR_DIM), dtype=np.float32)
            self._vector_node_ids = node_ids
            self._vector_index_dirty = False

    def _ensure_vector_index(self):
        """Rebuild vector index if dirty."""
        if self._vector_index_dirty:
            self._rebuild_vector_index()

    def find_similar_nodes(
        self,
        query_vector: np.ndarray,
        top_k: int = None,
        min_similarity: float = 0.0,
        exclude_ids: Set[str] = None,
        type_filter: str = None
    ) -> List[Tuple[Node, float]]:
        """
        Find nodes most similar to query vector.
        Returns list of (node, similarity_score) sorted by similarity desc.
        """
        if top_k is None:
            top_k = config.MAX_NODES_PER_SEARCH

        self._ensure_vector_index()

        if self._vector_matrix.shape[0] == 0:
            return []

        # Batch cosine similarity
        similarities = utils.batch_cosine_similarity(query_vector, self._vector_matrix)

        # Apply filters and sort
        results = []
        for i, sim in enumerate(similarities):
            sim_val = float(sim)
            if sim_val < min_similarity:
                continue
            node_id = self._vector_node_ids[i]
            if exclude_ids and node_id in exclude_ids:
                continue
            node = self.nodes.get(node_id)
            if not node:
                continue
            if type_filter and node.type != type_filter:
                continue
            results.append((node, sim_val))

        # Sort by similarity descending
        results.sort(key=lambda x: x[1], reverse=True)

        return results[:top_k]

    def find_similar_to_text(
        self,
        text: str,
        top_k: int = None,
        min_similarity: float = 0.0,
        exclude_ids: Set[str] = None,
        type_filter: str = None
    ) -> List[Tuple[Node, float]]:
        """
        Find nodes most similar to a text query.
        Convenience wrapper around find_similar_nodes.
        """
        query_vector = utils.text_to_vector_tfidf(text)
        return self.find_similar_nodes(
            query_vector, top_k, min_similarity,
            exclude_ids, type_filter
        )

    def find_similar_to_node(
        self,
        node_id: str,
        top_k: int = None,
        min_similarity: float = None
    ) -> List[Tuple[Node, float]]:
        """Find nodes most similar to an existing node."""
        node = self.nodes.get(node_id)
        if not node:
            return []
        if min_similarity is None:
            min_similarity = config.SIMILARITY_THRESHOLD
        return self.find_similar_nodes(
            node.vector, top_k, min_similarity,
            exclude_ids={node_id}
        )

    # ───────────────────────────────────────────────────
    # GRAPH TRAVERSAL
    # ───────────────────────────────────────────────────

    def traverse_bfs(
        self,
        start_ids: List[str],
        max_depth: int = None,
        max_nodes: int = 100
    ) -> Dict[str, Tuple[int, List[str]]]:
        """
        Breadth-first traversal from starting nodes.
        Returns: {node_id: (depth, [path_from_start])}
        """
        if max_depth is None:
            max_depth = config.MAX_TRAVERSAL_DEPTH

        visited = {}  # node_id → (depth, path)
        queue = []

        for sid in start_ids:
            if sid in self.nodes:
                visited[sid] = (0, [sid])
                queue.append((sid, 0, [sid]))

        while queue and len(visited) < max_nodes:
            current_id, depth, path = queue.pop(0)
            if depth >= max_depth:
                continue

            for neighbor, edge in self.get_neighbors(current_id):
                if neighbor.id not in visited:
                    new_path = path + [edge.id, neighbor.id]
                    visited[neighbor.id] = (depth + 1, new_path)
                    queue.append((neighbor.id, depth + 1, new_path))

        return visited

    def traverse_weighted_random(
        self,
        start_id: str,
        max_depth: int = None,
        temperature: float = 0.7
    ) -> List[Tuple[str, str]]:
        """
        Weighted random walk from a starting node.
        Edge weight determines probability of following that edge.
        Returns: [(node_id, edge_id), ...] — the path taken.
        """
        if max_depth is None:
            max_depth = config.MAX_TRAVERSAL_DEPTH

        if start_id not in self.nodes:
            return []

        path = [(start_id, "")]
        visited = {start_id}
        current = start_id

        for _ in range(max_depth):
            neighbors = self.get_neighbors(current)
            # Filter out already visited
            unvisited = [
                (node, edge) for node, edge in neighbors
                if node.id not in visited
            ]

            if not unvisited:
                break

            # Weight-based selection
            items = unvisited
            weights = [
                edge.weight * edge.confidence * node.weight
                for node, edge in items
            ]

            chosen_node, chosen_edge = utils.weighted_choice(
                items, weights, temperature
            )

            visited.add(chosen_node.id)
            path.append((chosen_node.id, chosen_edge.id))
            current = chosen_node.id

        return path

    def find_paths(
        self,
        from_id: str,
        to_id: str,
        max_depth: int = None,
        max_paths: int = 5
    ) -> List[List[str]]:
        """
        Find paths between two nodes using DFS.
        Returns list of paths, each path is [node_id, edge_id, node_id, ...].
        """
        if max_depth is None:
            max_depth = config.MAX_TRAVERSAL_DEPTH

        if from_id not in self.nodes or to_id not in self.nodes:
            return []

        all_paths = []

        def dfs(current: str, target: str, path: list, visited: set, depth: int):
            if len(all_paths) >= max_paths:
                return
            if depth > max_depth:
                return
            if current == target:
                all_paths.append(list(path))
                return

            for neighbor, edge in self.get_neighbors(current):
                if neighbor.id not in visited:
                    visited.add(neighbor.id)
                    path.extend([edge.id, neighbor.id])
                    dfs(neighbor.id, target, path, visited, depth + 1)
                    # Backtrack
                    path.pop()
                    path.pop()
                    visited.discard(neighbor.id)

        dfs(from_id, to_id, [from_id], {from_id}, 0)
        return all_paths

    # ───────────────────────────────────────────────────
    # REASONING CHAINS
    # ───────────────────────────────────────────────────

    def build_reasoning_chains(
        self,
        start_nodes: List[str],
        max_chains: int = None,
        max_depth: int = None
    ) -> List[ReasoningChain]:
        """
        Build reasoning chains from starting nodes.
        Combines BFS exploration with weighted random walks.
        Returns scored and sorted chains.
        """
        if max_chains is None:
            max_chains = config.MAX_CHAINS_PER_RESPONSE
        if max_depth is None:
            max_depth = config.MAX_TRAVERSAL_DEPTH

        chains = []

        for start_id in start_nodes:
            if start_id not in self.nodes:
                continue

            # Strategy 1: Weighted random walks (multiple)
            for _ in range(min(3, max_chains)):
                walk = self.traverse_weighted_random(start_id, max_depth)
                if len(walk) >= 2:
                    path = []
                    for node_id, edge_id in walk:
                        if edge_id:
                            path.append(edge_id)
                        path.append(node_id)

                    confidence = self._score_chain(path)
                    conclusion = self._chain_to_conclusion(path)

                    chain = ReasoningChain(
                        chain_id=config.generate_chain_id(path),
                        path=path,
                        conclusion=conclusion,
                        confidence=confidence
                    )
                    chains.append(chain)

            # Strategy 2: Follow high-weight edges
            high_weight_path = self._follow_strongest_path(start_id, max_depth)
            if len(high_weight_path) >= 3:
                confidence = self._score_chain(high_weight_path)
                conclusion = self._chain_to_conclusion(high_weight_path)

                chain = ReasoningChain(
                    chain_id=config.generate_chain_id(high_weight_path),
                    path=high_weight_path,
                    conclusion=conclusion,
                    confidence=confidence
                )
                chains.append(chain)

        # Deduplicate by chain id
        seen = set()
        unique_chains = []
        for c in chains:
            if c.id not in seen:
                seen.add(c.id)
                unique_chains.append(c)

        # Sort by confidence descending
        unique_chains.sort(key=lambda c: c.confidence, reverse=True)
        return unique_chains[:max_chains]

    def _follow_strongest_path(self, start_id: str, max_depth: int) -> list:
        """Follow the highest-weight edges from a starting node."""
        path = [start_id]
        visited = {start_id}
        current = start_id

        for _ in range(max_depth):
            edges = self.get_edges_from(current)
            # Filter unvisited
            candidates = [
                e for e in edges
                if e.to_node not in visited and e.to_node in self.nodes
            ]
            if not candidates:
                break

            # Pick strongest edge
            best_edge = max(candidates, key=lambda e: e.weight * e.confidence)
            path.append(best_edge.id)
            path.append(best_edge.to_node)
            visited.add(best_edge.to_node)
            current = best_edge.to_node

        return path

    def _score_chain(self, path: list) -> float:
        """
        Score a reasoning chain.
        Considers: edge weights, confidences, chain length, node weights.
        """
        if len(path) < 3:
            return 0.0

        edge_scores = []
        node_weights = []

        for item_id in path:
            if item_id in self.edges:
                edge = self.edges[item_id]
                edge_scores.append(edge.weight * edge.confidence)
            elif item_id in self.nodes:
                node_weights.append(self.nodes[item_id].weight)

        if not edge_scores:
            return 0.0

        avg_edge_score = sum(edge_scores) / len(edge_scores)
        avg_node_weight = sum(node_weights) / len(node_weights) if node_weights else 0.5

        # Shorter chains are generally more reliable
        length_penalty = 1.0 / (1.0 + 0.1 * len(edge_scores))

        score = avg_edge_score * avg_node_weight * length_penalty
        return utils.clamp(score, 0.0, 1.0)

    def _chain_to_conclusion(self, path: list) -> str:
        """
        Generate a text conclusion from a reasoning chain path.
        Extracts content from nodes in the path.
        """
        node_contents = []
        for item_id in path:
            node = self.nodes.get(item_id)
            if node:
                node_contents.append(node.content)

        if not node_contents:
            return ""
        return " → ".join(node_contents)

    def save_chain(self, chain: ReasoningChain):
        """Save a reasoning chain."""
        with self._lock:
            self.chains[chain.id] = chain
            self.memory.save_chain(chain.to_dict())

    def reinforce_chain(self, chain_id: str):
        """Reinforce a chain that was used in a response."""
        with self._lock:
            chain = self.chains.get(chain_id)
            if chain:
                chain.used_count += 1
                chain.confidence = min(chain.confidence * 1.02, 1.0)
                self.memory.save_chain(chain.to_dict())

                # Also reinforce all edges in the chain
                for item_id in chain.path:
                    if item_id in self.edges:
                        self.reinforce_edge(item_id)

    # ───────────────────────────────────────────────────
    # MERGE & PRUNE
    # ───────────────────────────────────────────────────

    def merge_nodes(self, node_id_keep: str, node_id_remove: str) -> bool:
        """
        Merge two redundant nodes. Keep the first, remove the second.
        Redirect all edges from removed node to kept node.
        """
        with self._lock:
            keep = self.nodes.get(node_id_keep)
            remove = self.nodes.get(node_id_remove)

            if not keep or not remove:
                return False

            # Combine weights
            keep.weight = min(keep.weight + remove.weight * 0.5, config.WEIGHT_MAX)

            # Average vectors
            keep.vector = utils.normalize(
                utils.vector_add(keep.vector, remove.vector) / 2.0
            )

            # Redirect edges
            edges_to_redirect = self.get_all_edges_for(node_id_remove)
            for edge in edges_to_redirect:
                new_from = node_id_keep if edge.from_node == node_id_remove else edge.from_node
                new_to = node_id_keep if edge.to_node == node_id_remove else edge.to_node

                if new_from == new_to:
                    continue  # Would create self-loop

                # Create redirected edge if doesn't exist
                if not self.edge_exists(new_from, new_to, edge.relation):
                    self.add_edge(
                        from_id=new_from,
                        to_id=new_to,
                        relation=edge.relation,
                        weight=edge.weight,
                        confidence=edge.confidence,
                        source=edge.source
                    )

            # Remove the merged node (and its old edges)
            self.remove_node(node_id_remove)
            keep.mark_dirty()
            self.memory.save_node(keep.to_dict())

            self._vector_index_dirty = True

        return True

    def prune_weak_edges(self, threshold: float = None) -> int:
        """Remove edges below weight threshold. Returns count removed."""
        if threshold is None:
            threshold = config.PRUNE_WEIGHT_THRESHOLD

        to_remove = []
        for edge in self.edges.values():
            if edge.weight < threshold and edge.source == "inferred":
                to_remove.append(edge.id)

        with self._lock:
            for edge_id in to_remove:
                self._remove_edge_internal(edge_id)

        return len(to_remove)

    def prune_orphan_nodes(self) -> int:
        """Remove nodes with no connections and low weight. Returns count removed."""
        to_remove = []
        for node in self.nodes.values():
            if (node.connections == 0 and
                    node.weight < config.WEIGHT_MIN * 2 and
                    node.source == "inferred"):
                to_remove.append(node.id)

        with self._lock:
            for node_id in to_remove:
                if node_id in self.nodes:
                    del self.nodes[node_id]
                    self.memory.delete_node(node_id)

        if to_remove:
            self._vector_index_dirty = True

        return len(to_remove)

    def find_redundant_pairs(self, limit: int = 20) -> List[Tuple[str, str, float]]:
        """
        Find pairs of nodes that might be redundant (very high similarity).
        Returns [(node_id_1, node_id_2, similarity), ...]
        """
        self._ensure_vector_index()
        pairs = []

        node_list = list(self.nodes.values())
        # Sample to avoid O(n²) for large graphs
        if len(node_list) > 500:
            sample_indices = np.random.choice(len(node_list), 500, replace=False)
            node_list = [node_list[i] for i in sample_indices]

        for i in range(len(node_list)):
            for j in range(i + 1, len(node_list)):
                n1 = node_list[i]
                n2 = node_list[j]
                if n1.type != n2.type:
                    continue  # Only merge same-type nodes
                sim = utils.cosine_similarity(n1.vector, n2.vector)
                if sim >= config.MERGE_THRESHOLD:
                    pairs.append((n1.id, n2.id, sim))
                    if len(pairs) >= limit:
                        return pairs

        return pairs

    # ───────────────────────────────────────────────────
    # STATISTICS
    # ───────────────────────────────────────────────────

    def _rebuild_stats(self):
        """Rebuild graph statistics."""
        total_nodes = len(self.nodes)
        total_edges = len(self.edges)

        inferred_nodes = sum(1 for n in self.nodes.values() if n.source == "inferred")
        inferred_edges = sum(1 for e in self.edges.values() if e.source == "inferred")

        avg_connections = 0.0
        if total_nodes > 0:
            avg_connections = sum(n.connections for n in self.nodes.values()) / total_nodes

        avg_confidence = 0.0
        if total_edges > 0:
            avg_confidence = sum(e.confidence for e in self.edges.values()) / total_edges

        # Max abstraction depth
        max_depth = 0
        for node in self.nodes.values():
            if node.type == "abstraction":
                depth = self._get_abstraction_depth(node.id)
                max_depth = max(max_depth, depth)

        self._stats = {
            "total_nodes": total_nodes,
            "total_edges": total_edges,
            "total_chains": len(self.chains),
            "inferred_nodes": inferred_nodes,
            "inferred_edges": inferred_edges,
            "max_abstraction_depth": max_depth,
            "avg_connections": round(avg_connections, 2),
            "avg_confidence": round(avg_confidence, 4),
            "inference_ratio": round(
                inferred_edges / max(total_edges, 1), 4
            ),
            "avg_chain_length": round(
                sum(len(c.path) for c in self.chains.values()) / max(len(self.chains), 1), 2
            )
        }

    def _get_abstraction_depth(self, node_id: str, visited: set = None) -> int:
        """Get the abstraction depth of a node (recursive)."""
        if visited is None:
            visited = set()
        if node_id in visited:
            return 0
        visited.add(node_id)

        max_child_depth = 0
        for edge in self.get_edges_to(node_id):
            if edge.relation == "instance_of":
                child_depth = self._get_abstraction_depth(edge.from_node, visited)
                max_child_depth = max(max_child_depth, child_depth)

        return max_child_depth + 1 if max_child_depth > 0 else (
            1 if self.nodes.get(node_id, Node("", "", "")).type in ("abstraction", "meta_abstraction") else 0
        )

    def get_stats(self) -> dict:
        """Get current graph statistics."""
        self._rebuild_stats()
        return dict(self._stats)

    def get_intelligence_score(self) -> float:
        """Calculate and return intelligence score."""
        self._rebuild_stats()
        return utils.calculate_intelligence_score(self._stats)

    # ───────────────────────────────────────────────────
    # SYNC
    # ───────────────────────────────────────────────────

    def sync(self) -> Optional[dict]:
        """Flush buffered changes to DB if needed."""
        return self.memory.flush_if_needed()

    def force_sync(self) -> dict:
        """Force flush all buffered changes to DB."""
        return self.memory.flush()

    # ───────────────────────────────────────────────────
    # DEBUG / INSPECTION
    # ───────────────────────────────────────────────────

    def describe_node(self, node_id: str) -> Optional[dict]:
        """Get detailed description of a node and its connections."""
        node = self.nodes.get(node_id)
        if not node:
            return None

        neighbors = self.get_neighbors(node_id)

        return {
            "id": node.id,
            "type": node.type,
            "content": node.content,
            "weight": node.weight,
            "connections": node.connections,
            "source": node.source,
            "neighbors": [
                {
                    "node_id": n.id,
                    "content": utils.truncate_text(n.content, 80),
                    "relation": e.relation,
                    "edge_weight": e.weight,
                    "edge_confidence": e.confidence
                }
                for n, e in neighbors
            ]
        }