videre-core 0.38.0

Shared SQLite, caching, and search helpers for the videre media library CLI
Documentation
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
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
use super::{
    FeatureVector, CLUSTER_QUALITY_FEATURE_NAMES, FEATURE_SCHEMA_VERSION, MEMBERSHIP_FEATURE_NAMES,
};
use rusqlite::{params, Connection, OptionalExtension, Row};
use serde::{Deserialize, Serialize};
use std::collections::{BTreeMap, BTreeSet};
use std::fmt;

#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum LearningAction {
    LabelCluster,
    CreatePerson,
    AssignFace,
    AssignCluster,
    RemoveFaceFromCluster,
    RemoveFaceFromPerson,
    DissolveCluster,
    QuestionYes,
    QuestionNo,
}

impl LearningAction {
    fn as_str(self) -> &'static str {
        match self {
            Self::LabelCluster => "label_cluster",
            Self::CreatePerson => "create_person",
            Self::AssignFace => "assign_face",
            Self::AssignCluster => "assign_cluster",
            Self::RemoveFaceFromCluster => "remove_face_from_cluster",
            Self::RemoveFaceFromPerson => "remove_face_from_person",
            Self::DissolveCluster => "dissolve_cluster",
            Self::QuestionYes => "question_yes",
            Self::QuestionNo => "question_no",
        }
    }

    fn parse(value: &str) -> Result<Self, LearningEventError> {
        match value {
            "label_cluster" => Ok(Self::LabelCluster),
            "create_person" => Ok(Self::CreatePerson),
            "assign_face" => Ok(Self::AssignFace),
            "assign_cluster" => Ok(Self::AssignCluster),
            "remove_face_from_cluster" => Ok(Self::RemoveFaceFromCluster),
            "remove_face_from_person" => Ok(Self::RemoveFaceFromPerson),
            "dissolve_cluster" => Ok(Self::DissolveCluster),
            "question_yes" => Ok(Self::QuestionYes),
            "question_no" => Ok(Self::QuestionNo),
            other => Err(LearningEventError::InvalidStoredValue(format!(
                "unknown learning action {other}"
            ))),
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum LearningDecisionKind {
    Membership,
    ClusterQuality,
}

impl LearningDecisionKind {
    fn as_str(self) -> &'static str {
        match self {
            Self::Membership => "membership",
            Self::ClusterQuality => "cluster_quality",
        }
    }

    fn parse(value: &str) -> Result<Self, LearningEventError> {
        match value {
            "membership" => Ok(Self::Membership),
            "cluster_quality" => Ok(Self::ClusterQuality),
            other => Err(LearningEventError::InvalidStoredValue(format!(
                "unknown learning decision kind {other}"
            ))),
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum LearningOutcome {
    Positive,
    Negative,
}

impl LearningOutcome {
    fn as_str(self) -> &'static str {
        match self {
            Self::Positive => "positive",
            Self::Negative => "negative",
        }
    }

    fn parse(value: &str) -> Result<Self, LearningEventError> {
        match value {
            "positive" => Ok(Self::Positive),
            "negative" => Ok(Self::Negative),
            other => Err(LearningEventError::InvalidStoredValue(format!(
                "unknown learning outcome {other}"
            ))),
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum EventFaceRole {
    Subject,
    ClusterMember,
    TargetSupport,
}

impl EventFaceRole {
    fn as_str(self) -> &'static str {
        match self {
            Self::Subject => "subject",
            Self::ClusterMember => "cluster_member",
            Self::TargetSupport => "target_support",
        }
    }

    fn parse(value: &str) -> Result<Self, LearningEventError> {
        match value {
            "subject" => Ok(Self::Subject),
            "cluster_member" => Ok(Self::ClusterMember),
            "target_support" => Ok(Self::TargetSupport),
            other => Err(LearningEventError::InvalidStoredValue(format!(
                "unknown event face role {other}"
            ))),
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum InvalidationReason {
    PersonRemoved,
    MissingRequiredContext,
}

impl InvalidationReason {
    fn parse(value: &str) -> Result<Self, LearningEventError> {
        match value {
            "person_removed" => Ok(Self::PersonRemoved),
            "missing_required_context" => Ok(Self::MissingRequiredContext),
            other => Err(LearningEventError::InvalidStoredValue(format!(
                "unknown invalidation reason {other}"
            ))),
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum LearningStatus {
    Current,
    Stale,
    Training,
    Failed,
}

impl LearningStatus {
    fn parse(value: &str) -> Result<Self, LearningEventError> {
        match value {
            "current" => Ok(Self::Current),
            "stale" => Ok(Self::Stale),
            "training" => Ok(Self::Training),
            "failed" => Ok(Self::Failed),
            other => Err(LearningEventError::InvalidStoredValue(format!(
                "unknown learning status {other}"
            ))),
        }
    }
}

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct EventFaceRef {
    pub face_id: i64,
    pub role: EventFaceRole,
    pub ordinal: u32,
}

#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
pub struct NewLearningEvent {
    pub action: LearningAction,
    pub decision_kind: LearningDecisionKind,
    pub outcome: LearningOutcome,
    pub embedding_model_id: String,
    pub active_profile_id: Option<i64>,
    pub target_identity: Option<String>,
    pub features: FeatureVector,
    pub support_count: u32,
    pub scorer_confidence: Option<f64>,
    pub faces: Vec<EventFaceRef>,
}

#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
pub struct StoredLearningEvent {
    pub id: i64,
    pub action: LearningAction,
    pub decision_kind: LearningDecisionKind,
    pub outcome: LearningOutcome,
    pub embedding_model_id: String,
    pub active_profile_id: Option<i64>,
    pub target_identity: Option<String>,
    pub features: FeatureVector,
    pub support_count: u32,
    pub scorer_confidence: Option<f64>,
    pub eligible: bool,
    pub invalidation_reason: Option<InvalidationReason>,
    pub created_at: String,
    pub faces: Vec<EventFaceRef>,
}

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct LearningState {
    pub generation: u64,
    pub trained_generation: u64,
    pub status: LearningStatus,
    pub training_generation: Option<u64>,
    pub last_profile_id: Option<i64>,
    pub last_error: Option<String>,
}

#[derive(Clone, Debug, PartialEq, Eq)]
pub struct LearningBatchReceipt {
    pub generation: u64,
    pub event_ids: Vec<i64>,
}

#[derive(Debug)]
pub enum LearningEventError {
    InvalidEvent(String),
    InvalidStoredValue(String),
    TransactionRequired,
    StateConflict(String),
    Sql(rusqlite::Error),
}

impl fmt::Display for LearningEventError {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::InvalidEvent(reason) => write!(f, "invalid learning event: {reason}"),
            Self::InvalidStoredValue(reason) => {
                write!(f, "invalid stored learning value: {reason}")
            }
            Self::TransactionRequired => {
                write!(f, "learning event batches require an existing transaction")
            }
            Self::StateConflict(reason) => write!(f, "learning state conflict: {reason}"),
            Self::Sql(error) => write!(f, "learning database error: {error}"),
        }
    }
}

impl std::error::Error for LearningEventError {}

impl From<rusqlite::Error> for LearningEventError {
    fn from(value: rusqlite::Error) -> Self {
        Self::Sql(value)
    }
}

pub fn ensure_learning_tables(conn: &Connection) -> rusqlite::Result<()> {
    conn.execute_batch(
        "CREATE TABLE IF NOT EXISTS face_learning_events (
            id INTEGER PRIMARY KEY,
            action_kind TEXT NOT NULL,
            decision_kind TEXT NOT NULL,
            outcome TEXT NOT NULL,
            embedding_model_id TEXT NOT NULL,
            feature_schema_version INTEGER NOT NULL,
            active_profile_id INTEGER,
            target_identity TEXT,
            feature_snapshot_json TEXT NOT NULL,
            support_count INTEGER NOT NULL CHECK(support_count >= 0),
            scorer_confidence REAL,
            eligible INTEGER NOT NULL DEFAULT 1 CHECK(eligible IN (0, 1)),
            invalidation_reason TEXT,
            created_at TEXT NOT NULL DEFAULT (datetime('now')),
            CHECK((eligible = 1 AND invalidation_reason IS NULL) OR
                  (eligible = 0 AND invalidation_reason IS NOT NULL))
        );
        CREATE TABLE IF NOT EXISTS face_learning_event_faces (
            event_id INTEGER NOT NULL REFERENCES face_learning_events(id),
            face_id INTEGER NOT NULL,
            role TEXT NOT NULL,
            ordinal INTEGER NOT NULL CHECK(ordinal >= 0),
            PRIMARY KEY(event_id, role, ordinal),
            UNIQUE(event_id, face_id)
        );
        CREATE INDEX IF NOT EXISTS face_learning_event_faces_face
            ON face_learning_event_faces(face_id);
        CREATE TABLE IF NOT EXISTS face_learning_state (
            id INTEGER PRIMARY KEY CHECK(id = 1),
            generation INTEGER NOT NULL DEFAULT 0 CHECK(generation >= 0),
            trained_generation INTEGER NOT NULL DEFAULT 0 CHECK(trained_generation >= 0),
            status TEXT NOT NULL,
            training_generation INTEGER,
            last_profile_id INTEGER,
            last_error TEXT
        );
        INSERT OR IGNORE INTO face_learning_state
            (id, generation, trained_generation, status)
            VALUES (1, 0, 0, 'current');
        CREATE TRIGGER IF NOT EXISTS face_learning_events_immutable_update
        BEFORE UPDATE ON face_learning_events
        WHEN NEW.id IS NOT OLD.id
          OR NEW.action_kind IS NOT OLD.action_kind
          OR NEW.decision_kind IS NOT OLD.decision_kind
          OR NEW.outcome IS NOT OLD.outcome
          OR NEW.embedding_model_id IS NOT OLD.embedding_model_id
          OR NEW.feature_schema_version IS NOT OLD.feature_schema_version
          OR NEW.active_profile_id IS NOT OLD.active_profile_id
          OR NEW.target_identity IS NOT OLD.target_identity
          OR NEW.feature_snapshot_json IS NOT OLD.feature_snapshot_json
          OR NEW.support_count IS NOT OLD.support_count
          OR NEW.scorer_confidence IS NOT OLD.scorer_confidence
          OR NEW.created_at IS NOT OLD.created_at
        BEGIN SELECT RAISE(ABORT, 'face learning event content is immutable'); END;
        CREATE TRIGGER IF NOT EXISTS face_learning_events_immutable_delete
        BEFORE DELETE ON face_learning_events
        BEGIN SELECT RAISE(ABORT, 'face learning events are immutable'); END;
        CREATE TRIGGER IF NOT EXISTS face_learning_event_faces_immutable_update
        BEFORE UPDATE ON face_learning_event_faces
        BEGIN SELECT RAISE(ABORT, 'face learning provenance is immutable'); END;
        CREATE TRIGGER IF NOT EXISTS face_learning_event_faces_immutable_delete
        BEFORE DELETE ON face_learning_event_faces
        BEGIN SELECT RAISE(ABORT, 'face learning provenance is immutable'); END;",
    )
}

fn validate_new_event(event: &NewLearningEvent) -> Result<(), LearningEventError> {
    if event.embedding_model_id.trim().is_empty() {
        return Err(LearningEventError::InvalidEvent(
            "embedding model id is empty".to_owned(),
        ));
    }
    event
        .features
        .validate()
        .map_err(|error| LearningEventError::InvalidEvent(error.to_string()))?;
    let feature_names: Vec<_> = event.features.values.keys().map(String::as_str).collect();
    let expected_names = match event.decision_kind {
        LearningDecisionKind::Membership => MEMBERSHIP_FEATURE_NAMES,
        LearningDecisionKind::ClusterQuality => CLUSTER_QUALITY_FEATURE_NAMES,
    };
    if feature_names != expected_names {
        return Err(LearningEventError::InvalidEvent(
            "feature schema does not match decision kind".to_owned(),
        ));
    }
    if event
        .scorer_confidence
        .is_some_and(|value| !value.is_finite() || !(0.0..=1.0).contains(&value))
    {
        return Err(LearningEventError::InvalidEvent(
            "scorer confidence is not a probability".to_owned(),
        ));
    }
    if let Some(identity) = &event.target_identity {
        if crate::person::normalize(identity).as_deref() != Some(identity.as_str()) {
            return Err(LearningEventError::InvalidEvent(
                "target identity is not normalized".to_owned(),
            ));
        }
    }
    if event.faces.is_empty() {
        return Err(LearningEventError::InvalidEvent(
            "event provenance is empty".to_owned(),
        ));
    }
    let mut face_ids = BTreeSet::new();
    let mut ordinals: BTreeMap<EventFaceRole, BTreeSet<u32>> = BTreeMap::new();
    for face in &event.faces {
        if face.face_id <= 0 {
            return Err(LearningEventError::InvalidEvent(format!(
                "face id {} is not positive",
                face.face_id
            )));
        }
        if !face_ids.insert(face.face_id) {
            return Err(LearningEventError::InvalidEvent(format!(
                "face {} appears more than once",
                face.face_id
            )));
        }
        if !ordinals.entry(face.role).or_default().insert(face.ordinal) {
            return Err(LearningEventError::InvalidEvent(format!(
                "duplicate {:?} ordinal {}",
                face.role, face.ordinal
            )));
        }
    }
    let subject_count = ordinals
        .get(&EventFaceRole::Subject)
        .map_or(0, BTreeSet::len);
    let support_count = ordinals
        .get(&EventFaceRole::TargetSupport)
        .map_or(0, BTreeSet::len);
    let cluster_count = ordinals
        .get(&EventFaceRole::ClusterMember)
        .map_or(0, BTreeSet::len);
    let provenance_matches = match event.decision_kind {
        LearningDecisionKind::Membership => {
            subject_count > 0
                && support_count > 0
                && cluster_count == 0
                && event.support_count as usize == support_count
        }
        LearningDecisionKind::ClusterQuality => {
            subject_count == 0
                && support_count == 0
                && cluster_count >= 2
                && event.support_count as usize == cluster_count
        }
    };
    if !provenance_matches {
        return Err(LearningEventError::InvalidEvent(
            "provenance does not match decision kind or support count".to_owned(),
        ));
    }
    for (role, values) in ordinals {
        if values.iter().copied().ne(0..values.len() as u32) {
            return Err(LearningEventError::InvalidEvent(format!(
                "{:?} ordinals are not contiguous",
                role
            )));
        }
    }
    Ok(())
}

pub fn append_event_batch_in_transaction(
    conn: &Connection,
    events: &[NewLearningEvent],
) -> Result<LearningBatchReceipt, LearningEventError> {
    if conn.is_autocommit() {
        return Err(LearningEventError::TransactionRequired);
    }
    if events.is_empty() {
        return Err(LearningEventError::InvalidEvent(
            "event batch is empty".to_owned(),
        ));
    }
    for event in events {
        validate_new_event(event)?;
    }

    let mut event_ids = Vec::with_capacity(events.len());
    for event in events {
        let feature_json = event
            .features
            .to_canonical_json()
            .map_err(|error| LearningEventError::InvalidEvent(error.to_string()))?;
        conn.execute(
            "INSERT INTO face_learning_events (
                action_kind, decision_kind, outcome, embedding_model_id,
                feature_schema_version, active_profile_id, target_identity,
                feature_snapshot_json, support_count, scorer_confidence,
                eligible, invalidation_reason
             ) VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, 1, NULL)",
            params![
                event.action.as_str(),
                event.decision_kind.as_str(),
                event.outcome.as_str(),
                event.embedding_model_id,
                event.features.schema_version,
                event.active_profile_id,
                event.target_identity,
                feature_json,
                event.support_count,
                event.scorer_confidence,
            ],
        )?;
        let event_id = conn.last_insert_rowid();
        let mut faces = event.faces.clone();
        faces.sort_by_key(|face| (face.role, face.ordinal, face.face_id));
        for face in faces {
            conn.execute(
                "INSERT INTO face_learning_event_faces (event_id, face_id, role, ordinal)
                 VALUES (?1, ?2, ?3, ?4)",
                params![event_id, face.face_id, face.role.as_str(), face.ordinal],
            )?;
        }
        event_ids.push(event_id);
    }
    conn.execute(
        "UPDATE face_learning_state
         SET generation = generation + 1,
             status = CASE WHEN status = 'training' THEN 'training' ELSE 'stale' END,
             last_error = CASE WHEN status = 'training' THEN last_error ELSE NULL END
         WHERE id = 1",
        [],
    )?;
    let generation = nonnegative_u64(
        conn.query_row(
            "SELECT generation FROM face_learning_state WHERE id = 1",
            [],
            |row| row.get(0),
        )?,
        "generation",
    )?;
    Ok(LearningBatchReceipt {
        generation,
        event_ids,
    })
}

/// Make all evidence tied to a removed identity ineligible and advance the
/// learning generation once. The caller owns the surrounding transaction and
/// calls this only when the visible person deletion changed face state.
/// Person-removal lifecycle: invalidate the identity's eligible evidence and
/// supersede its pending questions, advancing the generation exactly once
/// when either changed. Assumes the caller's transaction.
pub fn invalidate_identity_for_removal_in_transaction(
    conn: &Connection,
    identity: &str,
) -> Result<u64, LearningEventError> {
    let normalized = crate::person::normalize(identity).ok_or_else(|| {
        LearningEventError::InvalidEvent("identity to invalidate is empty".to_owned())
    })?;
    super::ensure_question_tables(conn)?;
    let invalidated = conn.execute(
        "UPDATE face_learning_events
         SET eligible = 0, invalidation_reason = 'person_removed'
         WHERE target_identity = ?1 AND eligible = 1",
        [&normalized],
    )?;
    let superseded = conn.execute(
        "UPDATE face_learning_questions
         SET status = 'superseded', decided_at = datetime('now')
         WHERE target_identity = ?1 AND status = 'pending'",
        [&normalized],
    )?;
    if invalidated > 0 || superseded > 0 {
        conn.execute(
            "UPDATE face_learning_state
             SET generation = generation + 1,
                 status = CASE WHEN status = 'training' THEN 'training' ELSE 'stale' END,
                 last_error = CASE WHEN status = 'training' THEN last_error ELSE NULL END
             WHERE id = 1",
            [],
        )?;
    }
    nonnegative_u64(
        conn.query_row(
            "SELECT generation FROM face_learning_state WHERE id = 1",
            [],
            |row| row.get(0),
        )?,
        "generation",
    )
}

pub fn invalidate_identity_in_transaction(
    conn: &Connection,
    identity: &str,
) -> Result<u64, LearningEventError> {
    if conn.is_autocommit() {
        return Err(LearningEventError::TransactionRequired);
    }
    let identity = crate::person::normalize(identity).ok_or_else(|| {
        LearningEventError::InvalidEvent("identity to invalidate is empty".to_owned())
    })?;
    let changed = conn.execute(
        "UPDATE face_learning_events
         SET eligible = 0, invalidation_reason = 'person_removed'
         WHERE target_identity = ?1 AND eligible = 1",
        [&identity],
    )?;
    if changed > 0 {
        conn.execute(
            "UPDATE face_learning_state
             SET generation = generation + 1,
                 status = CASE WHEN status = 'training' THEN 'training' ELSE 'stale' END,
                 last_error = CASE WHEN status = 'training' THEN last_error ELSE NULL END
             WHERE id = 1",
            [],
        )?;
    }
    nonnegative_u64(
        conn.query_row(
            "SELECT generation FROM face_learning_state WHERE id = 1",
            [],
            |row| row.get(0),
        )?,
        "generation",
    )
}

struct RawLearningEvent {
    id: i64,
    action: String,
    decision_kind: String,
    outcome: String,
    embedding_model_id: String,
    feature_schema_version: i64,
    active_profile_id: Option<i64>,
    target_identity: Option<String>,
    feature_json: String,
    support_count: i64,
    scorer_confidence: Option<f64>,
    eligible: i64,
    invalidation_reason: Option<String>,
    created_at: String,
}

fn raw_event(row: &Row<'_>) -> rusqlite::Result<RawLearningEvent> {
    Ok(RawLearningEvent {
        id: row.get(0)?,
        action: row.get(1)?,
        decision_kind: row.get(2)?,
        outcome: row.get(3)?,
        embedding_model_id: row.get(4)?,
        feature_schema_version: row.get(5)?,
        active_profile_id: row.get(6)?,
        target_identity: row.get(7)?,
        feature_json: row.get(8)?,
        support_count: row.get(9)?,
        scorer_confidence: row.get(10)?,
        eligible: row.get(11)?,
        invalidation_reason: row.get(12)?,
        created_at: row.get(13)?,
    })
}

const EVENT_COLUMNS: &str = "id, action_kind, decision_kind, outcome, embedding_model_id,
     feature_schema_version, active_profile_id, target_identity,
     feature_snapshot_json, support_count, scorer_confidence, eligible,
     invalidation_reason, created_at";

fn nonnegative_u64(value: i64, name: &str) -> Result<u64, LearningEventError> {
    u64::try_from(value)
        .map_err(|_| LearningEventError::InvalidStoredValue(format!("{name} is negative: {value}")))
}

fn nonnegative_u32(value: i64, name: &str) -> Result<u32, LearningEventError> {
    u32::try_from(value).map_err(|_| {
        LearningEventError::InvalidStoredValue(format!("{name} is out of range: {value}"))
    })
}

fn load_event_faces(
    conn: &Connection,
    event_id: i64,
) -> Result<Vec<EventFaceRef>, LearningEventError> {
    let mut statement = conn.prepare(
        "SELECT face_id, role, ordinal
         FROM face_learning_event_faces
         WHERE event_id = ?1
         ORDER BY CASE role
             WHEN 'subject' THEN 0
             WHEN 'cluster_member' THEN 1
             WHEN 'target_support' THEN 2
             ELSE 3 END,
             ordinal, face_id",
    )?;
    let rows = statement
        .query_map([event_id], |row| {
            Ok((
                row.get::<_, i64>(0)?,
                row.get::<_, String>(1)?,
                row.get::<_, i64>(2)?,
            ))
        })?
        .collect::<Result<Vec<_>, _>>()?;
    rows.into_iter()
        .map(|(face_id, role, ordinal)| {
            Ok(EventFaceRef {
                face_id,
                role: EventFaceRole::parse(&role)?,
                ordinal: nonnegative_u32(ordinal, "face ordinal")?,
            })
        })
        .collect()
}

fn parse_stored_features(
    json: &str,
    schema_version: i64,
) -> Result<FeatureVector, LearningEventError> {
    let schema_version = nonnegative_u32(schema_version, "feature schema version")?;
    let features: FeatureVector = serde_json::from_str(json).map_err(|error| {
        LearningEventError::InvalidStoredValue(format!("malformed feature JSON: {error}"))
    })?;
    if features.schema_version != schema_version {
        return Err(LearningEventError::InvalidStoredValue(format!(
            "feature JSON schema {} does not match column {schema_version}",
            features.schema_version
        )));
    }
    if features.values.is_empty() {
        return Err(LearningEventError::InvalidStoredValue(
            "feature vector is empty".to_owned(),
        ));
    }
    if let Some((name, _)) = features.values.iter().find(|(_, value)| !value.is_finite()) {
        return Err(LearningEventError::InvalidStoredValue(format!(
            "feature {name} is not finite"
        )));
    }
    if schema_version == FEATURE_SCHEMA_VERSION {
        features.validate().map_err(|error| {
            LearningEventError::InvalidStoredValue(format!("invalid feature vector: {error}"))
        })?;
    }
    Ok(features)
}

fn parse_raw_event(
    conn: &Connection,
    raw: RawLearningEvent,
) -> Result<StoredLearningEvent, LearningEventError> {
    if raw.embedding_model_id.trim().is_empty() {
        return Err(LearningEventError::InvalidStoredValue(
            "embedding model id is empty".to_owned(),
        ));
    }
    let eligible = match raw.eligible {
        0 => false,
        1 => true,
        other => {
            return Err(LearningEventError::InvalidStoredValue(format!(
                "eligible is not boolean: {other}"
            )))
        }
    };
    let invalidation_reason = raw
        .invalidation_reason
        .as_deref()
        .map(InvalidationReason::parse)
        .transpose()?;
    if eligible != invalidation_reason.is_none() {
        return Err(LearningEventError::InvalidStoredValue(
            "eligibility and invalidation reason disagree".to_owned(),
        ));
    }
    if raw
        .scorer_confidence
        .is_some_and(|value| !value.is_finite() || !(0.0..=1.0).contains(&value))
    {
        return Err(LearningEventError::InvalidStoredValue(
            "scorer confidence is not a probability".to_owned(),
        ));
    }
    let target_identity = raw.target_identity;
    if let Some(identity) = &target_identity {
        if crate::person::normalize(identity).as_deref() != Some(identity.as_str()) {
            return Err(LearningEventError::InvalidStoredValue(
                "target identity is not normalized".to_owned(),
            ));
        }
    }
    let features = parse_stored_features(&raw.feature_json, raw.feature_schema_version)?;
    let faces = load_event_faces(conn, raw.id)?;
    Ok(StoredLearningEvent {
        id: raw.id,
        action: LearningAction::parse(&raw.action)?,
        decision_kind: LearningDecisionKind::parse(&raw.decision_kind)?,
        outcome: LearningOutcome::parse(&raw.outcome)?,
        embedding_model_id: raw.embedding_model_id,
        active_profile_id: raw.active_profile_id,
        target_identity,
        features,
        support_count: nonnegative_u32(raw.support_count, "support count")?,
        scorer_confidence: raw.scorer_confidence,
        eligible,
        invalidation_reason,
        created_at: raw.created_at,
        faces,
    })
}

pub fn list_learning_events(
    conn: &Connection,
    limit: usize,
    before_id: Option<i64>,
) -> Result<Vec<StoredLearningEvent>, LearningEventError> {
    let sql = format!(
        "SELECT {EVENT_COLUMNS} FROM face_learning_events
         WHERE (?2 IS NULL OR id < ?2)
         ORDER BY id DESC LIMIT ?1"
    );
    let mut statement = conn.prepare(&sql)?;
    let limit = i64::try_from(limit).map_err(|_| {
        LearningEventError::InvalidStoredValue("event limit is out of range".to_owned())
    })?;
    let rows = statement
        .query_map(params![limit, before_id], raw_event)?
        .collect::<Result<Vec<_>, _>>()?;
    rows.into_iter()
        .map(|raw| parse_raw_event(conn, raw))
        .collect()
}

pub fn learning_event(
    conn: &Connection,
    event_id: i64,
) -> Result<Option<StoredLearningEvent>, LearningEventError> {
    let sql = format!("SELECT {EVENT_COLUMNS} FROM face_learning_events WHERE id = ?1");
    conn.query_row(&sql, [event_id], raw_event)
        .optional()?
        .map(|raw| parse_raw_event(conn, raw))
        .transpose()
}

pub fn eligible_events_for_training(
    conn: &Connection,
    embedding_model_id: &str,
    feature_schema_version: u32,
) -> Result<Vec<StoredLearningEvent>, LearningEventError> {
    let sql = format!(
        "SELECT {EVENT_COLUMNS} FROM face_learning_events
         WHERE eligible = 1
           AND embedding_model_id = ?1
           AND feature_schema_version = ?2
         ORDER BY id"
    );
    let mut statement = conn.prepare(&sql)?;
    let rows = statement
        .query_map(
            params![embedding_model_id, feature_schema_version],
            raw_event,
        )?
        .collect::<Result<Vec<_>, _>>()?;
    rows.into_iter()
        .map(|raw| parse_raw_event(conn, raw))
        .collect()
}

fn raw_learning_state(conn: &Connection) -> Result<LearningState, LearningEventError> {
    let raw = conn.query_row(
        "SELECT generation, trained_generation, status, training_generation,
                last_profile_id, last_error
         FROM face_learning_state WHERE id = 1",
        [],
        |row| {
            Ok((
                row.get::<_, i64>(0)?,
                row.get::<_, i64>(1)?,
                row.get::<_, String>(2)?,
                row.get::<_, Option<i64>>(3)?,
                row.get::<_, Option<i64>>(4)?,
                row.get::<_, Option<String>>(5)?,
            ))
        },
    )?;
    let generation = nonnegative_u64(raw.0, "generation")?;
    let trained_generation = nonnegative_u64(raw.1, "trained generation")?;
    if trained_generation > generation {
        return Err(LearningEventError::InvalidStoredValue(
            "trained generation exceeds current generation".to_owned(),
        ));
    }
    let status = LearningStatus::parse(&raw.2)?;
    let training_generation = raw
        .3
        .map(|value| nonnegative_u64(value, "training generation"))
        .transpose()?;
    if (status == LearningStatus::Training) != training_generation.is_some() {
        return Err(LearningEventError::InvalidStoredValue(
            "training status and generation disagree".to_owned(),
        ));
    }
    if training_generation.is_some_and(|value| value > generation) {
        return Err(LearningEventError::InvalidStoredValue(
            "training generation exceeds current generation".to_owned(),
        ));
    }
    Ok(LearningState {
        generation,
        trained_generation,
        status,
        training_generation,
        last_profile_id: raw.4,
        last_error: raw.5,
    })
}

pub fn learning_state(conn: &Connection) -> Result<LearningState, LearningEventError> {
    raw_learning_state(conn)
}

pub fn mark_training_started(conn: &Connection) -> Result<LearningState, LearningEventError> {
    let changed = conn.execute(
        "UPDATE face_learning_state
         SET status = 'training', training_generation = generation, last_error = NULL
         WHERE id = 1
           AND generation > trained_generation
           AND status IN ('stale', 'failed')",
        [],
    )?;
    if changed != 1 {
        return Err(LearningEventError::StateConflict(
            "no stale generation is available to train".to_owned(),
        ));
    }
    raw_learning_state(conn)
}

pub fn mark_training_failed(
    conn: &Connection,
    generation: u64,
    error: &str,
) -> Result<LearningState, LearningEventError> {
    if error.trim().is_empty() {
        return Err(LearningEventError::StateConflict(
            "training failure message is empty".to_owned(),
        ));
    }
    let generation = i64::try_from(generation).map_err(|_| {
        LearningEventError::StateConflict("training generation is out of range".to_owned())
    })?;
    let changed = conn.execute(
        "UPDATE face_learning_state
         SET status = 'failed', training_generation = NULL, last_error = ?1
         WHERE id = 1 AND status = 'training' AND training_generation = ?2",
        params![error, generation],
    )?;
    if changed != 1 {
        return Err(LearningEventError::StateConflict(format!(
            "generation {generation} is not training"
        )));
    }
    raw_learning_state(conn)
}

pub fn mark_generation_trained(
    conn: &Connection,
    generation: u64,
    profile_id: Option<i64>,
) -> Result<LearningState, LearningEventError> {
    let generation = i64::try_from(generation).map_err(|_| {
        LearningEventError::StateConflict("trained generation is out of range".to_owned())
    })?;
    let changed = conn.execute(
        "UPDATE face_learning_state
         SET trained_generation = ?1,
             status = CASE WHEN generation = ?1 THEN 'current' ELSE 'stale' END,
             training_generation = NULL,
             last_profile_id = ?2,
             last_error = NULL
         WHERE id = 1 AND status = 'training' AND training_generation = ?1",
        params![generation, profile_id],
    )?;
    if changed != 1 {
        return Err(LearningEventError::StateConflict(format!(
            "generation {generation} is not training"
        )));
    }
    raw_learning_state(conn)
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::face_learning::{
        extract_cluster_quality_features, extract_membership_features, DecisionStage,
        FaceObservation, FEATURE_SCHEMA_VERSION,
    };
    use rusqlite::{params, Connection};

    fn observation(face_id: i64, embedding: [f32; 2]) -> FaceObservation {
        FaceObservation {
            face_id,
            embedding: embedding.to_vec(),
            bbox_min_side: Some(80.0),
            blur: Some(160.0),
            det_score: Some(0.95),
            landmark_residual: Some(2.5),
            photo_hash: format!("photo-{face_id}"),
        }
    }

    fn event(action: LearningAction) -> NewLearningEvent {
        let subject = observation(10, [1.0, 0.0]);
        let target = observation(20, [0.8, 0.6]);
        NewLearningEvent {
            action,
            decision_kind: LearningDecisionKind::Membership,
            outcome: LearningOutcome::Positive,
            embedding_model_id: "buffalo_l/w600k_r50.onnx".to_owned(),
            active_profile_id: Some(7),
            target_identity: Some("alice".to_owned()),
            features: extract_membership_features(
                &[subject],
                &[target],
                DecisionStage::GallerySingleton,
            )
            .unwrap(),
            support_count: 1,
            scorer_confidence: Some(0.82),
            faces: vec![
                EventFaceRef {
                    face_id: 20,
                    role: EventFaceRole::TargetSupport,
                    ordinal: 0,
                },
                EventFaceRef {
                    face_id: 10,
                    role: EventFaceRole::Subject,
                    ordinal: 0,
                },
            ],
        }
    }

    fn append_committed(conn: &Connection, events: &[NewLearningEvent]) -> LearningBatchReceipt {
        conn.execute_batch("BEGIN IMMEDIATE").unwrap();
        let receipt = append_event_batch_in_transaction(conn, events).unwrap();
        conn.execute_batch("COMMIT").unwrap();
        receipt
    }

    #[test]
    fn event_round_trip_preserves_every_field_and_orders_provenance() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        let mut new = event(LearningAction::AssignCluster);
        new.decision_kind = LearningDecisionKind::ClusterQuality;
        new.outcome = LearningOutcome::Negative;
        new.features = extract_cluster_quality_features(
            &[
                observation(11, [1.0, 0.0]),
                observation(12, [0.8, 0.6]),
                observation(21, [0.6, 0.8]),
                observation(22, [0.0, 1.0]),
            ],
            DecisionStage::GalleryCluster,
        )
        .unwrap();
        new.faces = vec![
            EventFaceRef {
                face_id: 22,
                role: EventFaceRole::ClusterMember,
                ordinal: 3,
            },
            EventFaceRef {
                face_id: 11,
                role: EventFaceRole::ClusterMember,
                ordinal: 0,
            },
            EventFaceRef {
                face_id: 21,
                role: EventFaceRole::ClusterMember,
                ordinal: 2,
            },
            EventFaceRef {
                face_id: 12,
                role: EventFaceRole::ClusterMember,
                ordinal: 1,
            },
        ];
        new.support_count = 4;

        let receipt = append_committed(&conn, &[new.clone()]);
        assert_eq!(receipt.generation, 1);
        assert_eq!(receipt.event_ids.len(), 1);
        let stored = learning_event(&conn, receipt.event_ids[0])
            .unwrap()
            .unwrap();
        assert_eq!(stored.id, receipt.event_ids[0]);
        assert_eq!(stored.action, new.action);
        assert_eq!(stored.decision_kind, new.decision_kind);
        assert_eq!(stored.outcome, new.outcome);
        assert_eq!(stored.embedding_model_id, new.embedding_model_id);
        assert_eq!(stored.active_profile_id, new.active_profile_id);
        assert_eq!(stored.target_identity, new.target_identity);
        assert_eq!(stored.features, new.features);
        assert_eq!(stored.support_count, new.support_count);
        assert_eq!(stored.scorer_confidence, new.scorer_confidence);
        assert!(stored.eligible);
        assert_eq!(stored.invalidation_reason, None);
        assert!(!stored.created_at.is_empty());
        assert_eq!(
            stored.faces,
            vec![
                EventFaceRef {
                    face_id: 11,
                    role: EventFaceRole::ClusterMember,
                    ordinal: 0,
                },
                EventFaceRef {
                    face_id: 12,
                    role: EventFaceRole::ClusterMember,
                    ordinal: 1,
                },
                EventFaceRef {
                    face_id: 21,
                    role: EventFaceRole::ClusterMember,
                    ordinal: 2,
                },
                EventFaceRef {
                    face_id: 22,
                    role: EventFaceRole::ClusterMember,
                    ordinal: 3,
                },
            ]
        );
        assert_eq!(list_learning_events(&conn, 10, None).unwrap(), vec![stored]);
    }

    #[test]
    fn a_two_event_action_advances_generation_once() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        let receipt = append_committed(
            &conn,
            &[
                event(LearningAction::LabelCluster),
                event(LearningAction::AssignCluster),
            ],
        );

        assert_eq!(receipt.generation, 1);
        assert_eq!(receipt.event_ids.len(), 2);
        assert_eq!(learning_state(&conn).unwrap().generation, 1);
        assert_eq!(learning_state(&conn).unwrap().status, LearningStatus::Stale);
    }

    #[test]
    fn decision_features_roles_and_support_count_must_agree() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();

        let mut wrong_features = event(LearningAction::AssignFace);
        wrong_features.decision_kind = LearningDecisionKind::ClusterQuality;
        let mut wrong_count = event(LearningAction::AssignFace);
        wrong_count.support_count = 2;
        let mut wrong_roles = event(LearningAction::AssignFace);
        wrong_roles.faces[0].role = EventFaceRole::ClusterMember;

        for malformed in [wrong_features, wrong_count, wrong_roles] {
            conn.execute_batch("BEGIN IMMEDIATE").unwrap();
            assert!(matches!(
                append_event_batch_in_transaction(&conn, &[malformed]),
                Err(LearningEventError::InvalidEvent(_))
            ));
            conn.execute_batch("ROLLBACK").unwrap();
        }
        assert!(list_learning_events(&conn, 10, None).unwrap().is_empty());
        assert_eq!(learning_state(&conn).unwrap().generation, 0);
    }

    #[test]
    fn event_history_uses_a_stable_id_cursor() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        for action in [
            LearningAction::CreatePerson,
            LearningAction::AssignFace,
            LearningAction::RemoveFaceFromPerson,
        ] {
            append_committed(&conn, &[event(action)]);
        }

        let first = list_learning_events(&conn, 2, None).unwrap();
        assert_eq!(
            first.iter().map(|event| event.id).collect::<Vec<_>>(),
            vec![3, 2]
        );
        append_committed(&conn, &[event(LearningAction::QuestionNo)]);
        let second = list_learning_events(&conn, 2, Some(2)).unwrap();
        assert_eq!(
            second.iter().map(|event| event.id).collect::<Vec<_>>(),
            vec![1]
        );
    }

    #[test]
    fn validation_precedes_inserts_and_child_failure_rolls_back_with_the_caller() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        conn.execute_batch("BEGIN IMMEDIATE").unwrap();
        let mut invalid = event(LearningAction::QuestionNo);
        invalid.scorer_confidence = Some(f64::NAN);
        let error = append_event_batch_in_transaction(
            &conn,
            &[event(LearningAction::QuestionYes), invalid],
        )
        .unwrap_err();
        assert!(matches!(error, LearningEventError::InvalidEvent(_)));
        assert_eq!(
            conn.query_row("SELECT COUNT(*) FROM face_learning_events", [], |row| row
                .get::<_, i64>(
                0
            ))
            .unwrap(),
            0
        );
        conn.execute_batch("ROLLBACK").unwrap();

        conn.execute_batch(
            "CREATE TRIGGER reject_learning_child
             BEFORE INSERT ON face_learning_event_faces
             WHEN NEW.face_id = 999
             BEGIN SELECT RAISE(ABORT, 'test child failure'); END;
             BEGIN IMMEDIATE;",
        )
        .unwrap();
        let mut second = event(LearningAction::RemoveFaceFromPerson);
        second.faces[0].face_id = 999;
        assert!(append_event_batch_in_transaction(
            &conn,
            &[event(LearningAction::AssignFace), second]
        )
        .is_err());
        conn.execute_batch("ROLLBACK").unwrap();
        assert_eq!(list_learning_events(&conn, 10, None).unwrap(), Vec::new());
        assert_eq!(learning_state(&conn).unwrap().generation, 0);
    }

    #[test]
    fn event_content_and_provenance_are_immutable() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        let receipt = append_committed(&conn, &[event(LearningAction::AssignFace)]);
        let id = receipt.event_ids[0];

        assert!(conn
            .execute(
                "UPDATE face_learning_events SET outcome = 'negative' WHERE id = ?1",
                [id]
            )
            .is_err());
        assert!(conn
            .execute("DELETE FROM face_learning_events WHERE id = ?1", [id])
            .is_err());
        assert!(conn
            .execute(
                "UPDATE face_learning_event_faces SET face_id = 77 WHERE event_id = ?1",
                [id]
            )
            .is_err());
        assert!(conn
            .execute(
                "DELETE FROM face_learning_event_faces WHERE event_id = ?1",
                [id]
            )
            .is_err());

        conn.execute(
            "UPDATE face_learning_events
             SET eligible = 0, invalidation_reason = 'person_removed'
             WHERE id = ?1",
            [id],
        )
        .unwrap();
        let stored = learning_event(&conn, id).unwrap().unwrap();
        assert!(!stored.eligible);
        assert_eq!(
            stored.invalidation_reason,
            Some(InvalidationReason::PersonRemoved)
        );
    }

    #[test]
    fn learning_state_handles_feedback_arriving_during_training() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        assert_eq!(
            learning_state(&conn).unwrap(),
            LearningState {
                generation: 0,
                trained_generation: 0,
                status: LearningStatus::Current,
                training_generation: None,
                last_profile_id: None,
                last_error: None,
            }
        );

        append_committed(&conn, &[event(LearningAction::CreatePerson)]);
        let state = mark_training_started(&conn).unwrap();
        assert_eq!(state.status, LearningStatus::Training);
        assert_eq!(state.training_generation, Some(1));

        append_committed(&conn, &[event(LearningAction::RemoveFaceFromCluster)]);
        let state = learning_state(&conn).unwrap();
        assert_eq!(state.generation, 2);
        assert_eq!(state.status, LearningStatus::Training);
        assert_eq!(state.training_generation, Some(1));

        let state = mark_generation_trained(&conn, 1, Some(42)).unwrap();
        assert_eq!(state.trained_generation, 1);
        assert_eq!(state.status, LearningStatus::Stale);
        assert_eq!(state.last_profile_id, Some(42));

        mark_training_started(&conn).unwrap();
        let failed = mark_training_failed(&conn, 2, "training stopped").unwrap();
        assert_eq!(failed.status, LearningStatus::Failed);
        assert_eq!(failed.last_error.as_deref(), Some("training stopped"));
        assert_eq!(failed.trained_generation, 1);
        assert_eq!(failed.last_profile_id, Some(42));

        let retried = mark_training_started(&conn).unwrap();
        assert_eq!(retried.training_generation, Some(2));
        assert_eq!(retried.last_error, None);
        let current = mark_generation_trained(&conn, 2, Some(43)).unwrap();
        assert_eq!(current.status, LearningStatus::Current);
        assert_eq!(current.trained_generation, 2);
        assert_eq!(current.last_profile_id, Some(43));
    }

    #[test]
    fn incompatible_events_are_visible_but_excluded_from_training() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        let mut wrong_model = event(LearningAction::AssignFace);
        wrong_model.embedding_model_id = "other/model.onnx".to_owned();
        let receipt = append_committed(&conn, &[event(LearningAction::CreatePerson), wrong_model]);
        let mut old_features = event(LearningAction::DissolveCluster).features;
        old_features.schema_version = FEATURE_SCHEMA_VERSION + 1;
        insert_raw_event(
            &conn,
            "dissolve_cluster",
            "buffalo_l/w600k_r50.onnx",
            FEATURE_SCHEMA_VERSION + 1,
            &serde_json::to_string(&old_features).unwrap(),
        );

        assert_eq!(list_learning_events(&conn, 10, None).unwrap().len(), 3);
        let eligible =
            eligible_events_for_training(&conn, "buffalo_l/w600k_r50.onnx", FEATURE_SCHEMA_VERSION)
                .unwrap();
        assert_eq!(eligible.len(), 1);
        assert_eq!(eligible[0].id, receipt.event_ids[0]);
    }

    fn insert_raw_event(
        conn: &Connection,
        action: &str,
        embedding_model_id: &str,
        feature_schema_version: u32,
        feature_json: &str,
    ) {
        insert_raw_event_values(
            conn,
            action,
            "membership",
            "positive",
            embedding_model_id,
            feature_schema_version,
            feature_json,
            1,
            None,
        );
    }

    #[allow(clippy::too_many_arguments)]
    fn insert_raw_event_values(
        conn: &Connection,
        action: &str,
        decision_kind: &str,
        outcome: &str,
        embedding_model_id: &str,
        feature_schema_version: u32,
        feature_json: &str,
        eligible: i64,
        invalidation_reason: Option<&str>,
    ) {
        conn.execute(
            "INSERT INTO face_learning_events (
                action_kind, decision_kind, outcome, embedding_model_id,
                feature_schema_version, active_profile_id, target_identity,
                feature_snapshot_json, support_count, scorer_confidence,
                eligible, invalidation_reason
             ) VALUES (?1, ?2, ?3, ?4, ?5, NULL, NULL,
                       ?6, 0, NULL, ?7, ?8)",
            params![
                action,
                decision_kind,
                outcome,
                embedding_model_id,
                feature_schema_version,
                feature_json,
                eligible,
                invalidation_reason,
            ],
        )
        .unwrap();
    }

    #[test]
    fn corrupt_enum_and_feature_values_fail_closed() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        let json = event(LearningAction::AssignFace)
            .features
            .to_canonical_json()
            .unwrap();
        insert_raw_event(&conn, "unknown_action", "model", 1, &json);
        assert!(matches!(
            list_learning_events(&conn, 10, None),
            Err(LearningEventError::InvalidStoredValue(_))
        ));

        for (decision_kind, outcome) in [
            ("unknown_decision", "positive"),
            ("membership", "unknown_outcome"),
        ] {
            let conn = Connection::open_in_memory().unwrap();
            ensure_learning_tables(&conn).unwrap();
            insert_raw_event_values(
                &conn,
                "assign_face",
                decision_kind,
                outcome,
                "model",
                1,
                &json,
                1,
                None,
            );
            assert!(matches!(
                list_learning_events(&conn, 10, None),
                Err(LearningEventError::InvalidStoredValue(_))
            ));
        }

        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        insert_raw_event_values(
            &conn,
            "assign_face",
            "membership",
            "positive",
            "model",
            1,
            &json,
            0,
            Some("unknown_reason"),
        );
        assert!(matches!(
            list_learning_events(&conn, 10, None),
            Err(LearningEventError::InvalidStoredValue(_))
        ));

        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        conn.execute(
            "UPDATE face_learning_state SET status = 'unknown_status' WHERE id = 1",
            [],
        )
        .unwrap();
        assert!(matches!(
            learning_state(&conn),
            Err(LearningEventError::InvalidStoredValue(_))
        ));

        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        insert_raw_event(&conn, "assign_face", "model", 1, &json);
        let id = conn.last_insert_rowid();
        conn.execute(
            "INSERT INTO face_learning_event_faces (event_id, face_id, role, ordinal)
             VALUES (?1, 1, 'unknown_role', 0)",
            [id],
        )
        .unwrap();
        assert!(matches!(
            list_learning_events(&conn, 10, None),
            Err(LearningEventError::InvalidStoredValue(_))
        ));

        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        insert_raw_event(
            &conn,
            "assign_face",
            "model",
            1,
            r#"{"schema_version":1,"values":{"bad":1e999}}"#,
        );
        assert!(matches!(
            list_learning_events(&conn, 10, None),
            Err(LearningEventError::InvalidStoredValue(_))
        ));
    }

    #[test]
    fn appending_requires_the_callers_transaction() {
        let conn = Connection::open_in_memory().unwrap();
        ensure_learning_tables(&conn).unwrap();
        assert!(matches!(
            append_event_batch_in_transaction(&conn, &[event(LearningAction::AssignFace)]),
            Err(LearningEventError::TransactionRequired)
        ));
    }
}