memd 1.9.0

Local memory CLI for AI agents with hybrid retrieval
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
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
//! `memd consolidate` — LLM-backed memory consolidation (Phase 2).
//!
//! Builds a working region from chunks written (and retrieved) since
//! the last run, asks the configured [`Consolidator`] to rewrite them
//! into a smaller deduplicated set of `kind:consolidated` lessons,
//! and persists those lessons. Project-scoped runs soft-tombstone
//! superseded sources; tenant-wide runs retain project sources and
//! record non-destructive `derives_from` lineage. Nothing is deleted.

use std::path::{Path, PathBuf};
use std::time::{SystemTime, UNIX_EPOCH};

use serde::Deserialize;
use serde_json::{json, Value};

use super::paths::absolutize_project_dir;
use crate::consolidate::journal::LineageRelation;
use crate::consolidate::prompt::{
    build_consolidation_prompt, parse_consolidation_response, RegionChunk,
};
use crate::consolidate::select::select_consolidator;
use crate::consolidate::service::execute_consolidation_with_identity;
use crate::consolidate::Consolidator;
use crate::error::{MemdError, Result};
use crate::store::Store;
use crate::types::lifecycle::VisibilityPolicy;
use crate::types::{ChunkId, ChunkStatus, MemoryChunk, ProjectId, TenantId};

/// Invoke `close_range(CLOSE_RANGE_CLOEXEC)` without depending on the glibc
/// wrapper, which is absent from older glibc releases even when the kernel
/// supports the syscall.
#[cfg(all(target_os = "linux", target_env = "gnu"))]
fn close_range_cloexec() -> libc::c_long {
    unsafe {
        libc::syscall(
            libc::SYS_close_range,
            3 as libc::c_long,
            libc::c_uint::MAX as libc::c_long,
            libc::CLOSE_RANGE_CLOEXEC as libc::c_long,
        )
    }
}

/// Minimum region size below which consolidation is skipped unless
/// `--force` is passed. Also the dirty-chunk threshold at which
/// `memd session-start` spawns a background consolidation.
pub(super) const MIN_REGION: usize = 10;

/// Hard ceiling on chunks scanned while building the region.
const SCAN_LIMIT: usize = 500;

/// Options for the `consolidate` subcommand.
#[derive(Debug, Clone)]
pub(super) struct ConsolidateOptions {
    pub(super) tenant_id: Option<String>,
    pub(super) project_id: Option<String>,
    pub(super) project_dir: PathBuf,
    pub(super) max_region: usize,
    pub(super) dry_run: bool,
    pub(super) background: bool,
    pub(super) force: bool,
    pub(super) promote: bool,
    pub(super) legacy_immediate: bool,
}

#[derive(Debug, Clone)]
pub(super) struct ConsolidateReviewOptions {
    pub(super) run_id: Option<String>,
    pub(super) tenant_id: Option<String>,
    pub(super) project_id: Option<String>,
    pub(super) list: bool,
    pub(super) limit: usize,
    pub(super) accept: bool,
    pub(super) reject: bool,
}

/// Entry point: resolves the consolidator backend from the
/// environment, then runs the consolidation.
pub(super) async fn run_consolidate<S: Store>(
    store: &S,
    options: ConsolidateOptions,
) -> Result<Value> {
    if options.legacy_immediate {
        eprintln!(
            "warning: --legacy-immediate is deprecated and will be removed in the next release; use --promote"
        );
    }
    if options.background {
        return spawn_background(store.as_persistent().map(|s| s.data_dir()), &options);
    }
    // This process may be the detached child, holding the spawn locks through
    // descriptors whose FD_CLOEXEC the parent deliberately cleared. Consolidation
    // then execs a consolidator backend, and Rust's Command does not close
    // descriptors, so without this the backend inherits the locks and keeps them
    // after this process exits. Seal them: they stay open here, holding the
    // flocks, and no longer survive an exec.
    seal_inherited_descriptors();
    let consolidator = select_consolidator()?;
    consolidate_core(store, options, consolidator.as_ref()).await
}

/// Mark every inherited descriptor close-on-exec, keeping it open in this
/// process. Marking is not closing, so descriptors this process still needs are
/// unaffected; only what an `exec` would hand to a child changes.
#[cfg(unix)]
fn seal_inherited_descriptors() {
    // CLOSE_RANGE_CLOEXEC arrived in Linux 5.11. On 5.9 and 5.10 close_range
    // exists but rejects the flag with EINVAL, so both errors fall through to
    // the /proc walk rather than leaving descriptors unsealed.
    #[cfg(all(target_os = "linux", target_env = "gnu"))]
    if close_range_cloexec() == 0 {
        return;
    }
    let Ok(entries) = std::fs::read_dir("/proc/self/fd") else {
        return;
    };
    for entry in entries.flatten() {
        let Ok(fd) = entry.file_name().to_string_lossy().parse::<i32>() else {
            continue;
        };
        if fd < 3 {
            continue;
        }
        unsafe {
            let flags = libc::fcntl(fd, libc::F_GETFD);
            if flags >= 0 {
                libc::fcntl(fd, libc::F_SETFD, flags | libc::FD_CLOEXEC);
            }
        }
    }
}

#[cfg(not(unix))]
fn seal_inherited_descriptors() {}

pub(super) async fn run_consolidate_review<S: Store>(
    store: &S,
    options: ConsolidateReviewOptions,
) -> Result<Value> {
    let persistent = store.as_persistent().ok_or_else(|| {
        MemdError::ValidationError("consolidate-review requires a persistent store".to_string())
    })?;
    let tenant_id = options
        .tenant_id
        .as_deref()
        .map(TenantId::new)
        .transpose()?;
    ProjectId::validate_opt(options.project_id.as_deref())?;

    if options.list {
        if options.run_id.is_some() || options.accept || options.reject {
            return Err(MemdError::ValidationError(
                "consolidate-review --list does not accept a run id or decision".to_string(),
            ));
        }
        let mut staged = Vec::new();
        for run in persistent.metadata().list_staged_consolidation_runs(
            tenant_id.as_ref(),
            options.project_id.as_deref(),
            options.limit.clamp(1, 1_000),
        )? {
            let candidate_count = persistent
                .metadata()
                .get_consolidation_entries(&run.run_id)?
                .len();
            let source_count = persistent
                .metadata()
                .get_memory_lineage(&run.run_id)?
                .into_iter()
                .map(|edge| edge.source_chunk_id)
                .collect::<std::collections::BTreeSet<_>>()
                .len();
            staged.push(json!({
                "run_id": run.run_id.to_string(),
                "tenant_id": run.tenant_id.to_string(),
                "project_id": run.project_id,
                "state": run.state.as_str(),
                "candidate_count": candidate_count,
                "source_count": source_count,
                "consolidator": {
                    "adapter": run.consolidator,
                    "command": run.consolidator_command,
                    "model": run.consolidator_model,
                    "version": run.consolidator_version,
                },
                "created_at_ms": run.created_at_ms,
                "updated_at_ms": run.updated_at_ms,
            }));
        }
        let count = staged.len();
        return Ok(json!({
            "staged_runs": staged,
            "count": count,
        }));
    }
    if options.accept && options.reject {
        return Err(MemdError::ValidationError(
            "consolidate-review requires exactly one of --accept or --reject".to_string(),
        ));
    }
    let run_id = options.run_id.as_deref().ok_or_else(|| {
        MemdError::ValidationError(
            "consolidate-review requires a run id unless --list is used".to_string(),
        )
    })?;
    let run_id = crate::consolidate::journal::ConsolidationRunId::parse(run_id)?;
    let run = persistent
        .metadata()
        .get_consolidation_run(&run_id)?
        .ok_or_else(|| {
            unknown_review_run(&run_id, tenant_id.as_ref(), options.project_id.as_deref())
        })?;
    if tenant_id
        .as_ref()
        .is_some_and(|tenant| tenant != &run.tenant_id)
        || options
            .project_id
            .as_deref()
            .is_some_and(|project| run.project_id.as_deref() != Some(project))
    {
        return Err(unknown_review_run(
            &run_id,
            tenant_id.as_ref(),
            options.project_id.as_deref(),
        ));
    }

    if !options.accept && !options.reject {
        let entries = persistent.metadata().get_consolidation_entries(&run_id)?;
        let lineage = persistent.metadata().get_memory_lineage(&run_id)?;
        let mut candidates = Vec::with_capacity(entries.len());
        for entry in entries {
            let payload =
                review_payload(store, &run.tenant_id, entry.candidate_chunk_id.as_ref()).await?;
            candidates.push(json!({
                "entry_index": entry.entry_index,
                "state": entry.state.as_str(),
                "source_set_hash": entry.source_set_hash,
                "validation_error": entry.validation_error,
                "created_at_ms": entry.created_at_ms,
                "updated_at_ms": entry.updated_at_ms,
                "candidate": payload,
            }));
        }
        let source_ids = lineage
            .iter()
            .map(|edge| edge.source_chunk_id.clone())
            .collect::<std::collections::BTreeSet<_>>();
        let mut sources = Vec::with_capacity(source_ids.len());
        for source_id in source_ids {
            sources.push(review_payload(store, &run.tenant_id, Some(&source_id)).await?);
        }
        let lineage = lineage
            .into_iter()
            .map(|edge| {
                json!({
                    "source_chunk_id": edge.source_chunk_id.to_string(),
                    "result_chunk_id": edge.result_chunk_id.to_string(),
                    "relation": edge.relation.as_str(),
                    "tenant_id": edge.tenant_id.to_string(),
                    "project_id": edge.project_id,
                    "created_at_ms": edge.created_at_ms,
                })
            })
            .collect::<Vec<_>>();
        return Ok(json!({
            "run": {
                "run_id": run.run_id.to_string(),
                "tenant_id": run.tenant_id.to_string(),
                "project_id": run.project_id,
                "state": run.state.as_str(),
                "input_hash": run.input_hash,
                "prompt_hash": run.prompt_hash,
                "response_hash": run.response_hash,
                "audit_artifact_path": run.audit_artifact_path,
                "validation_result": run.validation_result,
                "error": run.error,
                "promotion_requested": run.promotion_requested,
                "consolidator": {
                    "adapter": run.consolidator,
                    "command": run.consolidator_command,
                    "model": run.consolidator_model,
                    "version": run.consolidator_version,
                },
                "created_at_ms": run.created_at_ms,
                "updated_at_ms": run.updated_at_ms,
            },
            "candidates": candidates,
            "sources": sources,
            "lineage": lineage,
        }));
    }

    let decision = if options.accept {
        crate::consolidate::service::ConsolidationReviewDecision::Accept
    } else {
        crate::consolidate::service::ConsolidationReviewDecision::Reject
    };
    let execution =
        crate::consolidate::service::review_consolidation_run(persistent, &run_id, decision)
            .await?;
    Ok(json!({
        "run_id": execution.run_id.to_string(),
        "state": execution.state.as_str(),
        "candidate_chunk_ids": execution.candidate_chunk_ids.iter().map(ToString::to_string).collect::<Vec<_>>(),
        "source_count": execution.source_count,
    }))
}

fn unknown_review_run(
    run_id: &crate::consolidate::journal::ConsolidationRunId,
    tenant_id: Option<&TenantId>,
    project_id: Option<&str>,
) -> MemdError {
    if tenant_id.is_some() || project_id.is_some() {
        MemdError::ValidationError(format!(
            "consolidation run {run_id} not found in requested scope"
        ))
    } else {
        MemdError::ValidationError(format!("unknown consolidation run {run_id}"))
    }
}

async fn review_payload<S: Store>(
    store: &S,
    tenant_id: &TenantId,
    chunk_id: Option<&ChunkId>,
) -> Result<Value> {
    let Some(chunk_id) = chunk_id else {
        return Ok(json!({
            "chunk_id": null,
            "payload_missing": true,
            "payload": null,
        }));
    };
    let payload = store.get(tenant_id, chunk_id).await?;
    Ok(json!({
        "chunk_id": chunk_id.to_string(),
        "payload_missing": payload.is_none(),
        "payload": payload,
    }))
}

/// Core consolidation flow, parameterised over the [`Consolidator`] so
/// tests can inject a deterministic backend.
async fn consolidate_core<S: Store>(
    store: &S,
    options: ConsolidateOptions,
    consolidator: &dyn Consolidator,
) -> Result<Value> {
    let project_dir = absolutize_project_dir(&options.project_dir)?;
    let (tenant_id, project_id) = resolve_scope(
        &project_dir,
        options.tenant_id.clone(),
        options.project_id.clone(),
    )?;
    let tenant = TenantId::new(&tenant_id)?;
    let max_region = options.max_region.clamp(MIN_REGION, SCAN_LIMIT);

    // Reconcile stale journaled work before selecting a new region or paying
    // for another model call. The recovery service has an in-flight grace
    // period, so this does not claim a run that is still writing candidates.
    if !options.dry_run {
        if let Some(persistent) = store.as_persistent() {
            crate::consolidate::service::recover_consolidation_runs(persistent, 100).await?;
        }
    }

    let state_path = consolidate_state_path(&project_dir);
    let since_ms = read_last_consolidation_ms(&state_path);

    let region = collect_region(
        store,
        &tenant,
        project_id.as_deref(),
        since_ms,
        &project_dir,
        max_region,
    )
    .await?;

    if region.len() < MIN_REGION && !options.force {
        return Ok(json!({
            "skipped": "below_threshold",
            "region_size": region.len(),
            "min_region": MIN_REGION,
        }));
    }

    let prompt = build_consolidation_prompt(&region);
    if options.dry_run {
        return Ok(json!({
            "dry_run": true,
            "region_size": region.len(),
            "prompt": prompt,
        }));
    }

    let identity = consolidator.identity().await?;
    let raw = consolidator.consolidate(&prompt).await?;
    let entries = parse_consolidation_response(&raw, &region)?;

    let inherited_ctx = most_common_ctx_tags(&region, 3);
    let persistent = store.as_persistent().ok_or_else(|| {
        MemdError::ValidationError("consolidate requires a persistent store".to_string())
    })?;
    let promotion_requested = options.promote || options.legacy_immediate;
    let execution = execute_consolidation_with_identity(
        persistent,
        &tenant,
        project_id.as_deref(),
        &entries,
        if project_id.is_some() {
            LineageRelation::Supersedes
        } else {
            LineageRelation::DerivesFrom
        },
        &identity,
        &inherited_ctx,
        &prompt,
        &raw,
        promotion_requested,
    )
    .await?;
    if !matches!(
        execution.state,
        crate::consolidate::journal::ConsolidationState::Validated
            | crate::consolidate::journal::ConsolidationState::Committed
    ) {
        return Err(MemdError::StorageError(format!(
            "consolidation run {} stopped before validation in state {}",
            execution.run_id, execution.state
        )));
    }
    let written = execution
        .candidate_chunk_ids
        .iter()
        .map(ToString::to_string)
        .collect::<Vec<_>>();
    let committed = execution.state == crate::consolidate::journal::ConsolidationState::Committed;
    let tombstoned = if committed && project_id.is_some() {
        execution.source_count
    } else {
        0
    };
    let now = now_ms();

    write_last_consolidation_ms(&state_path, now)?;

    // Tenant-wide runs persist lessons with no project_id; flag that
    // project-scoped searches will not surface them.
    let tenant_wide_write = committed && project_id.is_none() && !written.is_empty();
    let mut summary = json!({
        "tenant_id": tenant_id,
        "project_id": project_id,
        "region_size": region.len(),
        "consolidated": if committed { written.len() } else { 0 },
        "staged": if committed { 0 } else { written.len() },
        "tombstoned": tombstoned,
        "state": execution.state.as_str(),
        "promotion_requested": promotion_requested,
        "consolidator": identity,
        "candidate_chunk_ids": written,
        "new_chunk_ids": if committed { execution.candidate_chunk_ids.iter().map(ToString::to_string).collect::<Vec<_>>() } else { Vec::new() },
        "run_id": execution.run_id.to_string(),
        "reused_existing_run": execution.reused_existing_run,
    });
    if tenant_wide_write {
        summary["warning"] = json!(
            "consolidated lessons written without project_id; project-scoped searches \
             will not see them directly (they surface via tenant-wide search and \
             memory-md Machine-Wide Fact Library); project-scoped sources were retained"
        );
    }
    Ok(summary)
}

/// Count the chunks that would form the consolidation region for the
/// given scope. Used by `memd session-start` to decide whether a
/// background consolidation is worth spawning.
pub(super) async fn dirty_region_size<S: Store>(
    store: &S,
    project_dir: &Path,
    tenant_arg: Option<String>,
    project_arg: Option<String>,
) -> Result<usize> {
    let (tenant_id, project_id) = resolve_scope(project_dir, tenant_arg, project_arg)?;
    let tenant = TenantId::new(&tenant_id)?;
    let since_ms = read_last_consolidation_ms(&consolidate_state_path(project_dir));
    let region = collect_region(
        store,
        &tenant,
        project_id.as_deref(),
        since_ms,
        project_dir,
        SCAN_LIMIT,
    )
    .await?;
    Ok(region.len())
}

/// Resolve `(tenant_id, project_id)` from explicit args, falling back
/// to `.memd/project_scope.json` then `.memd/config.json`. Mirrors
/// `scope::resolve_required`: an explicit `--tenant-id` suppresses
/// scope-file inheritance entirely, so a tenant-wide run is never
/// silently narrowed to the cwd scope file's project_id.
pub(super) fn resolve_scope(
    project_dir: &Path,
    tenant_arg: Option<String>,
    project_arg: Option<String>,
) -> Result<(String, Option<String>)> {
    if let Some(tenant_id) = tenant_arg {
        return Ok((tenant_id, project_arg));
    }
    let scope = read_scope_file(project_dir);
    let tenant_id = scope
        .as_ref()
        .and_then(|s| s.tenant_id.clone())
        .ok_or_else(|| {
            MemdError::ValidationError(
                "consolidate requires --tenant-id or .memd/project_scope.json".to_string(),
            )
        })?;
    let project_id = project_arg.or_else(|| scope.and_then(|s| s.project_id));
    Ok((tenant_id, project_id))
}

#[derive(Debug, Deserialize)]
struct ScopeFile {
    #[serde(default)]
    tenant_id: Option<String>,
    #[serde(default)]
    project_id: Option<String>,
}

/// Read `{tenant_id, project_id}` from `.memd/project_scope.json` or
/// `.memd/config.json`, ignoring any other fields.
fn read_scope_file(project_dir: &Path) -> Option<ScopeFile> {
    for name in [".memd/project_scope.json", ".memd/config.json"] {
        let path = project_dir.join(name);
        let Ok(text) = std::fs::read_to_string(&path) else {
            continue;
        };
        if let Ok(scope) = serde_json::from_str::<ScopeFile>(&text) {
            return Some(scope);
        }
    }
    None
}

fn consolidate_state_path(project_dir: &Path) -> PathBuf {
    project_dir.join(".memd/data/consolidate.state.json")
}

#[derive(Debug, Default, Deserialize, serde::Serialize)]
struct ConsolidateState {
    #[serde(default)]
    last_consolidation_ms: i64,
}

fn read_last_consolidation_ms(state_path: &Path) -> i64 {
    std::fs::read_to_string(state_path)
        .ok()
        .and_then(|text| serde_json::from_str::<ConsolidateState>(&text).ok())
        .map(|state| state.last_consolidation_ms)
        .unwrap_or(0)
}

fn write_last_consolidation_ms(state_path: &Path, ms: i64) -> Result<()> {
    if let Some(parent) = state_path.parent() {
        std::fs::create_dir_all(parent)?;
    }
    let state = ConsolidateState {
        last_consolidation_ms: ms,
    };
    std::fs::write(state_path, serde_json::to_string_pretty(&state)? + "\n")?;
    Ok(())
}

/// Build the consolidation working region: chunks written since
/// `since_ms`, unioned with chunk ids seen in recent search logs.
/// `kind:consolidated` / `kind:superseded` chunks are excluded so the
/// region is always raw material, never prior output.
async fn collect_region<S: Store>(
    store: &S,
    tenant: &TenantId,
    project_id: Option<&str>,
    since_ms: i64,
    project_dir: &Path,
    max_region: usize,
) -> Result<Vec<RegionChunk>> {
    let mut seen: std::collections::HashSet<String> = std::collections::HashSet::new();
    let mut region: Vec<RegionChunk> = Vec::new();
    let visibility = VisibilityPolicy::default();
    let visibility_now_ms = now_ms();

    let recent = store
        .list_chunks_for_project(tenant, project_id, SCAN_LIMIT, 0)
        .await?;
    for chunk in recent {
        let Some(resolved) = store.get_with_lifecycle(tenant, &chunk.chunk_id).await? else {
            continue;
        };
        if !visibility.is_visible_at(resolved.status, &resolved.lifecycle, visibility_now_ms) {
            continue;
        }
        consider_chunk(
            resolved.chunk,
            RegionReason::NewWrite,
            since_ms,
            project_id,
            &mut seen,
            &mut region,
        );
    }

    for chunk_id in search_log_chunk_ids(project_dir, since_ms) {
        if seen.contains(&chunk_id) {
            continue;
        }
        let Ok(parsed) = ChunkId::parse(&chunk_id) else {
            continue;
        };
        // Lifecycle can change after a search log is written. Resolve its
        // overlay now so an intervening correction cannot revive a stale
        // source and fork its supersession lineage. The lookup remains
        // tenant-scoped, and `consider_chunk` re-checks `project_id`.
        if let Ok(Some(resolved)) = store.get_with_lifecycle(tenant, &parsed).await {
            if !visibility.is_visible_at(resolved.status, &resolved.lifecycle, visibility_now_ms) {
                continue;
            }
            consider_chunk(
                resolved.chunk,
                RegionReason::RecentHit,
                since_ms,
                project_id,
                &mut seen,
                &mut region,
            );
        }
    }

    // Newest first, then id ascending for deterministic ties.
    region.sort_by(|a, b| {
        b.timestamp_created
            .cmp(&a.timestamp_created)
            .then_with(|| a.chunk_id.cmp(&b.chunk_id))
    });
    region.truncate(max_region);
    Ok(region)
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum RegionReason {
    NewWrite,
    RecentHit,
}

/// Apply region-membership filters to one chunk and push it if it
/// qualifies. When `project_id` is `Some`, chunks outside that
/// project are rejected — this guards the search-log union path,
/// where chunk ids are resolved tenant-wide.
fn consider_chunk(
    chunk: MemoryChunk,
    reason: RegionReason,
    since_ms: i64,
    project_id: Option<&str>,
    seen: &mut std::collections::HashSet<String>,
    region: &mut Vec<RegionChunk>,
) {
    if let Some(project_id) = project_id {
        if chunk.project_id.as_option() != Some(project_id) {
            return;
        }
    }
    if reason == RegionReason::NewWrite && chunk.timestamp_created <= since_ms {
        return;
    }
    if matches!(
        chunk.status,
        ChunkStatus::Candidate
            | ChunkStatus::Superseded
            | ChunkStatus::Deleted
            | ChunkStatus::Expired
            | ChunkStatus::Error
    ) {
        return;
    }
    if chunk
        .tags
        .iter()
        .any(|t| t.starts_with("kind:consolidated") || t.starts_with("kind:superseded"))
    {
        return;
    }
    let id = chunk.chunk_id.to_string();
    if !seen.insert(id.clone()) {
        return;
    }
    region.push(RegionChunk {
        chunk_id: id,
        chunk_type: chunk.chunk_type.to_string(),
        tags: chunk.tags.clone(),
        timestamp_created: chunk.timestamp_created,
        text: chunk.text,
        project_id: chunk.project_id.as_option().map(str::to_string),
    });
}

/// Scan `.memd/search-logs/*.json` for chunk ids retrieved since
/// `since_ms`. The unix-ms stamp in each filename
/// (`memd_search_<stamp>.json`) bounds the scan. Best-effort: any
/// unreadable or unparseable file is silently skipped.
fn search_log_chunk_ids(project_dir: &Path, since_ms: i64) -> Vec<String> {
    let log_dir = project_dir.join(".memd/search-logs");
    let Ok(entries) = std::fs::read_dir(&log_dir) else {
        return Vec::new();
    };
    let mut ids = Vec::new();
    for entry in entries.flatten() {
        let name = entry.file_name().to_string_lossy().to_string();
        if !name.ends_with(".json") || name.ends_with("_log.jsonl") {
            continue;
        }
        let Some(stamp) = name
            .rsplit_once('_')
            .and_then(|(_, tail)| tail.strip_suffix(".json"))
            .and_then(|s| s.parse::<i64>().ok())
        else {
            continue;
        };
        if stamp <= since_ms {
            continue;
        }
        let Ok(text) = std::fs::read_to_string(entry.path()) else {
            continue;
        };
        let Ok(payload) = serde_json::from_str::<Value>(&text) else {
            continue;
        };
        if let Some(results) = payload.get("results").and_then(Value::as_array) {
            for result in results {
                if let Some(id) = result.get("chunk_id").and_then(Value::as_str) {
                    if !id.is_empty() {
                        ids.push(id.to_string());
                    }
                }
            }
        }
    }
    ids
}

/// The `ctx:*` tags most common across the region, up to `limit`.
/// Consolidated lessons inherit these so subsystem/file context is
/// not lost when raw chunks are tombstoned.
fn most_common_ctx_tags(region: &[RegionChunk], limit: usize) -> Vec<String> {
    let mut counts: std::collections::HashMap<&str, usize> = std::collections::HashMap::new();
    for chunk in region {
        for tag in &chunk.tags {
            if tag.starts_with("ctx:") {
                *counts.entry(tag.as_str()).or_insert(0) += 1;
            }
        }
    }
    let mut ranked: Vec<(&str, usize)> = counts.into_iter().collect();
    // Frequency desc, then tag asc for deterministic ties.
    ranked.sort_by(|a, b| b.1.cmp(&a.1).then_with(|| a.0.cmp(b.0)));
    ranked
        .into_iter()
        .take(limit)
        .map(|(tag, _)| tag.to_string())
        .collect()
}

/// Ceiling on background consolidations running at once across every scope.
///
/// The per-scope lock stops one project stacking spawns, but nothing bounded
/// the machine: each project that crosses [`MIN_REGION`] adds a process, and
/// they all queue on the one store writer. Four keeps several projects making
/// progress while leaving the herd bounded and small.
const SPAWN_SLOTS: usize = 4;

/// Path of the slot lock bounding total concurrent background consolidations.
fn spawn_slot_path(data_dir: &Path, slot: usize) -> PathBuf {
    data_dir.join(format!(".consolidate-slot-{slot}.lock"))
}

/// Claim any free slot, bounding concurrency across all scopes.
///
/// Slots are interchangeable, so this walks them in order and takes the first
/// that is free. `Busy` means every slot is held, which is the bound doing its
/// job rather than an error.
fn claim_spawn_slot(data_dir: &Path) -> Result<SpawnClaim> {
    for slot in 0..SPAWN_SLOTS {
        match try_claim_spawn_lock(&spawn_slot_path(data_dir, slot))? {
            SpawnClaim::Free(guard) => return Ok(SpawnClaim::Free(guard)),
            SpawnClaim::Busy => continue,
        }
    }
    Ok(SpawnClaim::Busy)
}

/// Path of the single-flight lock guarding background consolidation for
/// one scope. The scope is hashed because tenant and project ids are free
/// text and must not shape a filename.
fn spawn_lock_path(data_dir: &Path, tenant_id: &str, project_id: Option<&str>) -> PathBuf {
    use sha2::{Digest, Sha256};
    let mut hasher = Sha256::new();
    hasher.update(tenant_id.as_bytes());
    hasher.update([0u8]);
    hasher.update(project_id.unwrap_or("").as_bytes());
    let digest = format!("{:x}", hasher.finalize());
    data_dir.join(format!(".consolidate-{}.lock", &digest[..16]))
}

/// Outcome of trying to claim the single-flight lock for one scope.
enum SpawnClaim {
    /// Free to spawn. Carries the parent's descriptor when a lock was taken,
    /// and `None` on platforms without `flock`, where no guard is available
    /// and the previous unguarded behavior stands.
    Free(Option<std::fs::File>),
    /// Another background consolidation already owns this scope.
    Busy,
}

/// Try to claim the spawn lock for one scope.
///
/// The returned descriptor keeps `FD_CLOEXEC`; [`inherit_lock_fd`] is what
/// hands it to a single child. The parent closes its own copy right after
/// spawning, so the lock lives as long as the child and the kernel releases it
/// even if the child is killed.
#[cfg(unix)]
fn try_claim_spawn_lock(path: &Path) -> Result<SpawnClaim> {
    use std::os::fd::{AsRawFd, FromRawFd, OwnedFd};

    let opened = std::fs::OpenOptions::new()
        .read(true)
        .write(true)
        .create(true)
        .truncate(false)
        .open(path)?;
    // Keep the lock off fds 0-2. If stdin/stdout/stderr were closed, open()
    // hands back one of them and the child's Stdio::null() setup would later
    // overwrite it, silently dropping the guard and restoring the herd.
    let file = if opened.as_raw_fd() < 3 {
        let raw = unsafe { libc::fcntl(opened.as_raw_fd(), libc::F_DUPFD_CLOEXEC, 3) };
        if raw < 0 {
            return Err(MemdError::IoError(std::io::Error::last_os_error()));
        }
        drop(opened);
        std::fs::File::from(unsafe { OwnedFd::from_raw_fd(raw) })
    } else {
        opened
    };
    let fd = file.as_raw_fd();
    if unsafe { libc::flock(fd, libc::LOCK_EX | libc::LOCK_NB) } != 0 {
        // EWOULDBLOCK and EAGAIN are the same value on Linux; compare rather
        // than match so the duplicate arm does not read as unreachable.
        let err = std::io::Error::last_os_error();
        if err.raw_os_error() == Some(libc::EWOULDBLOCK) || err.raw_os_error() == Some(libc::EAGAIN)
        {
            return Ok(SpawnClaim::Busy);
        }
        return Err(MemdError::IoError(err));
    }
    Ok(SpawnClaim::Free(Some(file)))
}

/// Hand the claimed lock to one specific child.
///
/// `FD_CLOEXEC` is cleared after fork and before exec, so the descriptor
/// reaches this child and nothing else. Clearing it on the parent's own
/// descriptor instead would leak the lock into every process spawned by any
/// thread while the claim is open, and those processes would hold the scope
/// blocked for as long as they live.
/// Every descriptor is handed over by one closure. Registering a second
/// `pre_exec` would run `close_range` again and re-flag the descriptors an
/// earlier closure had just cleared, silently dropping those locks at `exec`.
#[cfg(unix)]
fn inherit_lock_fds(command: &mut std::process::Command, locks: &[&std::fs::File]) {
    use std::os::fd::{AsRawFd, RawFd};
    use std::os::unix::process::CommandExt;

    // Copied into the closure by value: allocating after `fork` is not
    // async-signal-safe, so the descriptors travel in a fixed array.
    let mut fds: [RawFd; SPAWN_SLOTS + 1] = [-1; SPAWN_SLOTS + 1];
    let count = locks.len().min(fds.len());
    for (slot, file) in fds.iter_mut().zip(locks.iter().take(count)) {
        *slot = file.as_raw_fd();
    }
    if count == 0 {
        return;
    }
    unsafe {
        command.pre_exec(move || {
            // `fork` copies the whole descriptor table, so this child starts out
            // holding every descriptor the parent had, including the store's
            // SQLite and Tantivy locks. Only `FD_CLOEXEC` drops them at `exec`,
            // and this code cannot guarantee every descriptor in the process
            // carries it. Mark the whole range first, then clear the flag on the
            // one descriptor the child is meant to keep, so inheritance is
            // stated here rather than inferred from every other open file.
            //
            // CLOSE_RANGE_CLOEXEC sets the flag instead of closing, so a
            // descriptor std still needs for stdio setup is not pulled out from
            // under it. Kernels without the call (pre-5.9) fall back to the
            // previous behavior.
            //
            // Use the raw syscall on linux-gnu so linking does not require a
            // recent glibc close_range wrapper. Other targets retain the prior
            // behavior and rely on std's own FD_CLOEXEC discipline for other
            // files.
            #[cfg(all(target_os = "linux", target_env = "gnu"))]
            if close_range_cloexec() != 0 {
                let err = std::io::Error::last_os_error();
                // CLOSE_RANGE_CLOEXEC arrived in Linux 5.11. On 5.9 and 5.10 the
                // syscall exists but rejects the flag with EINVAL, so treating
                // only ENOSYS as tolerable made every spawn fail on those
                // kernels. Neither is fatal: std already marks its own
                // descriptors close-on-exec, and the parent seals the rest.
                let tolerable =
                    matches!(err.raw_os_error(), Some(libc::ENOSYS) | Some(libc::EINVAL));
                if !tolerable {
                    return Err(err);
                }
            }
            // fcntl is async-signal-safe, which pre_exec requires.
            for fd in fds.iter().take(count) {
                if libc::fcntl(*fd, libc::F_SETFD, 0) != 0 {
                    return Err(std::io::Error::last_os_error());
                }
            }
            Ok(())
        });
    }
}

/// Without `flock` there is no guard, so spawning proceeds as it did before.
#[cfg(not(unix))]
fn try_claim_spawn_lock(_path: &Path) -> Result<SpawnClaim> {
    Ok(SpawnClaim::Free(None))
}

/// Re-exec `memd consolidate` as a detached background child and
/// return immediately. The child runs the same scope without
/// `--background`.
///
/// `lock_root` enables single-flight: without it every `memd session-start`
/// stacks another detached child, and because those children cannot finish
/// while they queue behind each other the dirty region never shrinks below
/// [`MIN_REGION`], so each new session spawns one more. The guard has to sit
/// here in the parent rather than in the child's CLI handler, because the
/// child opens the store (contacting the warm worker) before any handler
/// runs, which is the contention being avoided.
fn spawn_background(lock_root: Option<&Path>, options: &ConsolidateOptions) -> Result<Value> {
    let mut lock = None;
    let mut slot = None;
    let mut options = options.clone();
    if let Some(data_dir) = lock_root {
        let (tenant_id, project_id) = resolve_scope(
            &options.project_dir,
            options.tenant_id.clone(),
            options.project_id.clone(),
        )?;
        let path = spawn_lock_path(data_dir, &tenant_id, project_id.as_deref());
        match try_claim_spawn_lock(&path)? {
            SpawnClaim::Free(guard) => lock = guard,
            SpawnClaim::Busy => return Ok(json!({ "skipped": "already_running" })),
        }
        // The scope lock bounds one project; this bounds the machine. Claimed
        // second so a contended scope reports itself as such rather than
        // consuming a slot it will not use.
        match claim_spawn_slot(data_dir)? {
            SpawnClaim::Free(guard) => slot = guard,
            SpawnClaim::Busy => {
                return Ok(json!({
                    "skipped": "consolidation_slots_busy",
                    "slots": SPAWN_SLOTS,
                }))
            }
        }
        // Pin the child to the scope the lock was keyed on. Left implicit, the
        // child re-resolves the scope file after the spawn, and a scope edit in
        // that window would have it consolidate one scope while holding
        // another's lock, letting a second child claim the free one.
        options.tenant_id = Some(tenant_id);
        options.project_id = project_id;
    }
    let held: Vec<&std::fs::File> = [lock.as_ref(), slot.as_ref()]
        .into_iter()
        .flatten()
        .collect();
    let result = spawn_background_inner(&options, &held);
    drop(held);
    // Explicit: closes only the parent's descriptors, after the child has been
    // spawned and inherited its own copies, which is what keeps the flocks held.
    drop(lock);
    drop(slot);
    result
}

fn spawn_background_inner(
    options: &ConsolidateOptions,
    #[cfg_attr(not(unix), allow(unused_variables))] locks: &[&std::fs::File],
) -> Result<Value> {
    let exe = std::env::current_exe()
        .map_err(|e| MemdError::ProtocolError(format!("cannot resolve current executable: {e}")))?;
    let mut command = std::process::Command::new(exe);
    command.arg("consolidate");
    if let Some(tenant) = &options.tenant_id {
        command.arg("--tenant-id").arg(tenant);
    }
    if let Some(project) = &options.project_id {
        command.arg("--project-id").arg(project);
    }
    command
        .arg("--project-dir")
        .arg(&options.project_dir)
        .arg("--max-region")
        .arg(options.max_region.to_string());
    if options.force {
        command.arg("--force");
    }
    if options.promote {
        command.arg("--promote");
    }
    if options.legacy_immediate {
        command.arg("--legacy-immediate");
    }
    command
        .stdin(std::process::Stdio::null())
        .stdout(std::process::Stdio::null())
        .stderr(std::process::Stdio::null());
    #[cfg(unix)]
    {
        use std::os::unix::process::CommandExt;
        command.process_group(0);
        inherit_lock_fds(&mut command, locks);
    }
    let child = command.spawn().map_err(|e| {
        MemdError::ProtocolError(format!("failed to spawn background consolidate: {e}"))
    })?;
    Ok(json!({ "spawned_background": true, "pid": child.id() }))
}

fn now_ms() -> i64 {
    SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .map(|d| d.as_millis() as i64)
        .unwrap_or(0)
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::consolidate::MockEnvConsolidator;
    use crate::mcp::handlers::{handle_memory_search, SearchParams};
    use crate::store::metadata::MetadataStore;
    use crate::store::persistent::{PersistentStore, PersistentStoreConfig};
    use crate::store::Store;
    use crate::types::lifecycle::LifecycleDelta;
    use crate::types::{ChunkStatus, ChunkType, ProjectId};
    use tempfile::tempdir;

    /// Serialises tests that set the process-global `MOCK_RESPONSE_ENV`.
    static MOCK_ENV_LOCK: tokio::sync::Mutex<()> = tokio::sync::Mutex::const_new(());

    #[test]
    fn spawn_lock_is_scope_keyed() {
        let dir = tempdir().unwrap();
        let a = spawn_lock_path(dir.path(), "default", Some("virosync"));
        assert_eq!(a, spawn_lock_path(dir.path(), "default", Some("virosync")));
        assert_ne!(a, spawn_lock_path(dir.path(), "default", Some("memd")));
        assert_ne!(a, spawn_lock_path(dir.path(), "other", Some("virosync")));
        // A free-text scope must not shape the filename.
        let hostile = spawn_lock_path(dir.path(), "../../etc", Some("a/b"));
        assert_eq!(hostile.parent(), Some(dir.path()));
    }

    /// The consolidate child holds the spawn locks through descriptors whose
    /// FD_CLOEXEC the parent cleared, and it then execs a consolidator backend.
    /// Sealing must leave the descriptor open here while stopping it reaching
    /// that backend, otherwise an orphaned backend keeps the slot forever.
    #[cfg(unix)]
    #[test]
    fn sealing_keeps_the_lock_but_denies_it_to_an_exec() {
        use std::os::fd::AsRawFd;

        let dir = tempdir().unwrap();
        let path = spawn_lock_path(dir.path(), "default", Some("virosync"));
        let held = match try_claim_spawn_lock(&path).unwrap() {
            SpawnClaim::Free(guard) => guard.expect("unix claim holds a descriptor"),
            SpawnClaim::Busy => panic!("first claim must succeed"),
        };
        // The parent hands the descriptor over by clearing the flag.
        let fd = held.as_raw_fd();
        unsafe { libc::fcntl(fd, libc::F_SETFD, 0) };
        assert_eq!(
            unsafe { libc::fcntl(fd, libc::F_GETFD) } & libc::FD_CLOEXEC,
            0
        );

        seal_inherited_descriptors();

        assert_ne!(
            unsafe { libc::fcntl(fd, libc::F_GETFD) } & libc::FD_CLOEXEC,
            0,
            "sealing must set close-on-exec so a backend cannot inherit the lock"
        );
        // Still open, so the flock is still held: a second claim is refused.
        assert!(matches!(
            try_claim_spawn_lock(&path).unwrap(),
            SpawnClaim::Busy
        ));
        drop(held);
    }

    /// The per-scope lock bounds one project; this bounds the machine. Claims
    /// the guards directly rather than through children, so the test adds no
    /// process to a suite whose other tests hold index locks.
    #[cfg(unix)]
    #[test]
    fn spawn_slots_bound_concurrent_consolidations() {
        let dir = tempdir().unwrap();
        let mut held = Vec::new();
        for slot in 0..SPAWN_SLOTS {
            // Any test in this binary that spawns a process forks the whole
            // descriptor table, so a slot can read as busy for the instant
            // before that child reaches its own `exec`. Retry briefly rather
            // than mistake that for the ceiling.
            let claimed = (0..200).find_map(|_| {
                match claim_spawn_slot(dir.path()).expect("claim must not error") {
                    SpawnClaim::Free(guard) => Some(guard.expect("unix claim holds a descriptor")),
                    SpawnClaim::Busy => {
                        std::thread::sleep(std::time::Duration::from_millis(10));
                        None
                    }
                }
            });
            held.push(claimed.unwrap_or_else(|| panic!("slot {slot} must become free")));
        }
        assert!(
            matches!(claim_spawn_slot(dir.path()).unwrap(), SpawnClaim::Busy),
            "a spawn beyond the ceiling must be refused"
        );
        // Releasing one consolidation lets the next project through. Same
        // transient-holder allowance as above: a concurrently forked child can
        // still hold the descriptor released here until it reaches its `exec`.
        held.pop();
        let freed = (0..200).any(|_| {
            if matches!(
                claim_spawn_slot(dir.path()).expect("claim must not error"),
                SpawnClaim::Free(_)
            ) {
                return true;
            }
            std::thread::sleep(std::time::Duration::from_millis(10));
            false
        });
        assert!(freed, "releasing a consolidation must free its slot");
    }

    /// The guard is only useful if the lock outlives the parent that took it.
    /// Hold it across a real `exec`, then prove a second claim is refused while
    /// that child lives and succeeds once it is gone.
    #[cfg(unix)]
    #[test]
    fn spawn_lock_survives_exec_and_releases_with_the_child() {
        let dir = tempdir().unwrap();
        let path = spawn_lock_path(dir.path(), "default", Some("virosync"));

        let held = match try_claim_spawn_lock(&path).unwrap() {
            SpawnClaim::Free(guard) => guard.expect("unix claim holds a descriptor"),
            SpawnClaim::Busy => panic!("first claim must succeed"),
        };
        // Hand the lock over exactly as spawn_background_inner does.
        let mut command = std::process::Command::new("sleep");
        command.arg("30");
        inherit_lock_fds(&mut command, &[&held]);
        let mut child = command.spawn().expect("spawn holder");
        // The parent's descriptor goes away exactly as it does after a real
        // background spawn; only the child's inherited copy remains.
        drop(held);

        assert!(
            matches!(try_claim_spawn_lock(&path).unwrap(), SpawnClaim::Busy),
            "second claim must be refused while the child holds the inherited lock"
        );

        child.kill().unwrap();
        child.wait().unwrap();
        // `FD_CLOEXEC` bounds `exec`, not `fork`: any of the ~1000 tests in this
        // binary that spawns a process during the claim window inherits the
        // descriptor until its own exec runs, and holds the lock for that
        // instant. Allow those transient holders to clear before concluding the
        // kernel failed to release.
        let released = (0..200).any(|_| {
            if matches!(try_claim_spawn_lock(&path).unwrap(), SpawnClaim::Free(_)) {
                return true;
            }
            std::thread::sleep(std::time::Duration::from_millis(10));
            false
        });
        assert!(
            released,
            "kernel must release the lock when the child dies, leaving no stale state"
        );
    }

    fn persistent_store(dir: &Path) -> PersistentStore {
        let cfg = PersistentStoreConfig {
            data_dir: dir.to_path_buf(),
            enable_dense_search: false,
            enable_hybrid_search: false,
            ..Default::default()
        };
        PersistentStore::open(cfg).expect("open store")
    }

    fn region_chunk(id: &str, ctx: &[&str]) -> RegionChunk {
        RegionChunk {
            chunk_id: id.to_string(),
            chunk_type: "summary".to_string(),
            tags: ctx.iter().map(|t| t.to_string()).collect(),
            timestamp_created: 1,
            text: "t".to_string(),
            project_id: None,
        }
    }

    #[test]
    fn most_common_ctx_prefers_frequent_tags() {
        let region = vec![
            region_chunk("a", &["ctx:file:x", "ctx:subsystem:s"]),
            region_chunk("b", &["ctx:file:x"]),
            region_chunk("c", &["ctx:file:y"]),
        ];
        let tags = most_common_ctx_tags(&region, 2);
        assert_eq!(tags[0], "ctx:file:x");
        assert_eq!(tags.len(), 2);
    }

    #[test]
    fn state_roundtrip() {
        let dir = tempdir().unwrap();
        let path = consolidate_state_path(dir.path());
        assert_eq!(read_last_consolidation_ms(&path), 0);
        write_last_consolidation_ms(&path, 4242).unwrap();
        assert_eq!(read_last_consolidation_ms(&path), 4242);
    }

    #[tokio::test]
    async fn recent_search_hit_includes_chunk_created_before_watermark() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut old_chunk = MemoryChunk::new(
            tenant.clone(),
            "old lesson retrieved after the last consolidation",
            ChunkType::Summary,
        )
        .with_project(ProjectId::from("p"));
        old_chunk.timestamp_created = 100;
        let old_id = store.add(old_chunk).await.unwrap();

        let log_dir = dir.path().join(".memd/search-logs");
        std::fs::create_dir_all(&log_dir).unwrap();
        std::fs::write(
            log_dir.join("memd_search_2000.json"),
            serde_json::to_string(&json!({
                "results": [{"chunk_id": old_id.to_string()}]
            }))
            .unwrap(),
        )
        .unwrap();

        let region = collect_region(&store, &tenant, Some("p"), 1000, dir.path(), 50)
            .await
            .unwrap();

        assert_eq!(region.len(), 1, "a recent hit must revive an old chunk");
        assert_eq!(region[0].chunk_id, old_id.to_string());
    }

    #[tokio::test]
    async fn recent_search_hit_excludes_chunk_superseded_after_search() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut old_chunk = MemoryChunk::new(
            tenant.clone(),
            "stale lesson retrieved before a later correction",
            ChunkType::Summary,
        )
        .with_project(ProjectId::from("p"));
        old_chunk.timestamp_created = 100;
        let old_id = store.add(old_chunk).await.unwrap();

        let log_dir = dir.path().join(".memd/search-logs");
        std::fs::create_dir_all(&log_dir).unwrap();
        std::fs::write(
            log_dir.join("memd_search_2000.json"),
            serde_json::to_string(&json!({
                "results": [{"chunk_id": old_id.to_string()}]
            }))
            .unwrap(),
        )
        .unwrap();

        store
            .update_lifecycle_if_exists(
                &tenant,
                &old_id,
                &LifecycleDelta {
                    status: Some(ChunkStatus::Superseded),
                    superseded_by: Some(ChunkId::new()),
                    lifecycle_updated_at_ms: Some(2500),
                    ..Default::default()
                },
            )
            .await
            .unwrap();

        let region = collect_region(&store, &tenant, Some("p"), 1000, dir.path(), 50)
            .await
            .unwrap();

        assert!(
            region.is_empty(),
            "a recent hit superseded after search must not re-enter consolidation"
        );
    }

    #[tokio::test]
    async fn new_write_excludes_chunk_superseded_before_consolidation() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut stale_chunk = MemoryChunk::new(
            tenant.clone(),
            "new lesson corrected before consolidation",
            ChunkType::Summary,
        )
        .with_project(ProjectId::from("p"));
        stale_chunk.timestamp_created = 2000;
        let stale_id = store.add(stale_chunk).await.unwrap();
        store
            .update_lifecycle_if_exists(
                &tenant,
                &stale_id,
                &LifecycleDelta {
                    status: Some(ChunkStatus::Superseded),
                    superseded_by: Some(ChunkId::new()),
                    lifecycle_updated_at_ms: Some(2500),
                    ..Default::default()
                },
            )
            .await
            .unwrap();

        let region = collect_region(&store, &tenant, Some("p"), 1000, dir.path(), 50)
            .await
            .unwrap();

        assert!(
            region.is_empty(),
            "a new write superseded before consolidation must not enter the region"
        );
    }

    #[tokio::test]
    async fn recent_search_hits_reject_foreign_project_chunks() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let own_id = store
            .add(
                MemoryChunk::new(tenant.clone(), "own project lesson", ChunkType::Summary)
                    .with_project(ProjectId::from("p")),
            )
            .await
            .unwrap();
        let foreign_id = store
            .add(
                MemoryChunk::new(tenant.clone(), "foreign project lesson", ChunkType::Summary)
                    .with_project(ProjectId::from("q")),
            )
            .await
            .unwrap();

        let log_dir = dir.path().join(".memd/search-logs");
        std::fs::create_dir_all(&log_dir).unwrap();
        std::fs::write(
            log_dir.join("memd_search_2000.json"),
            serde_json::to_string(&json!({
                "results": [
                    {"chunk_id": own_id.to_string()},
                    {"chunk_id": foreign_id.to_string()}
                ]
            }))
            .unwrap(),
        )
        .unwrap();

        let region = collect_region(&store, &tenant, Some("p"), 0, dir.path(), 50)
            .await
            .unwrap();

        let ids = region
            .iter()
            .map(|chunk| chunk.chunk_id.as_str())
            .collect::<std::collections::HashSet<_>>();
        assert!(ids.contains(own_id.to_string().as_str()));
        assert!(!ids.contains(foreign_id.to_string().as_str()));
    }

    #[tokio::test]
    async fn region_scan_boundary_is_deterministic_for_equal_timestamps() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut expected = Vec::new();
        for i in 0..=SCAN_LIMIT {
            let mut chunk = MemoryChunk::new(
                tenant.clone(),
                format!("equal timestamp lesson {i}"),
                ChunkType::Summary,
            )
            .with_project(ProjectId::from("p"));
            chunk.timestamp_created = 2000;
            expected.push(store.add(chunk).await.unwrap().to_string());
        }
        expected.sort();
        expected.truncate(SCAN_LIMIT);

        let region = collect_region(&store, &tenant, Some("p"), 1000, dir.path(), SCAN_LIMIT)
            .await
            .unwrap();
        let actual = region
            .iter()
            .map(|chunk| chunk.chunk_id.clone())
            .collect::<Vec<_>>();

        assert_eq!(actual, expected);
    }

    #[test]
    fn unstamped_search_log_is_ignored() {
        let dir = tempdir().unwrap();
        let log_dir = dir.path().join(".memd/search-logs");
        std::fs::create_dir_all(&log_dir).unwrap();
        std::fs::write(
            log_dir.join("manual.json"),
            serde_json::to_string(&json!({
                "results": [{"chunk_id": ChunkId::new().to_string()}]
            }))
            .unwrap(),
        )
        .unwrap();

        assert!(search_log_chunk_ids(dir.path(), 1000).is_empty());
    }

    #[test]
    fn resolve_scope_prefers_explicit_args() {
        let dir = tempdir().unwrap();
        let (tenant, project) =
            resolve_scope(dir.path(), Some("t".to_string()), Some("p".to_string())).unwrap();
        assert_eq!(tenant, "t");
        assert_eq!(project.as_deref(), Some("p"));
    }

    #[test]
    fn resolve_scope_reads_config_json() {
        let dir = tempdir().unwrap();
        std::fs::create_dir_all(dir.path().join(".memd")).unwrap();
        std::fs::write(
            dir.path().join(".memd/config.json"),
            r#"{"tenant_id":"ct","project_id":"cp"}"#,
        )
        .unwrap();
        let (tenant, project) = resolve_scope(dir.path(), None, None).unwrap();
        assert_eq!(tenant, "ct");
        assert_eq!(project.as_deref(), Some("cp"));
    }

    #[tokio::test]
    async fn below_threshold_skips_without_force() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        for i in 0..3 {
            store
                .add(
                    MemoryChunk::new(tenant.clone(), format!("chunk {i}"), ChunkType::Summary)
                        .with_project(ProjectId::from("p")),
                )
                .await
                .unwrap();
        }
        let opts = ConsolidateOptions {
            tenant_id: Some("t".to_string()),
            project_id: Some("p".to_string()),
            project_dir: dir.path().to_path_buf(),
            max_region: 50,
            dry_run: false,
            background: false,
            force: false,
            promote: false,
            legacy_immediate: false,
        };
        let consolidator = MockEnvConsolidator;
        let result = consolidate_core(&store, opts, &consolidator).await.unwrap();
        assert_eq!(result["skipped"], "below_threshold");
    }

    #[tokio::test]
    async fn consolidate_writes_and_tombstones() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut ids = Vec::new();
        for i in 0..12 {
            let id = store
                .add(
                    MemoryChunk::new(
                        tenant.clone(),
                        format!("duplicate lesson {i}: cache keys must be tenant-scoped"),
                        ChunkType::Summary,
                    )
                    .with_project(ProjectId::from("p"))
                    .with_tags(vec!["ctx:subsystem:cache".to_string()]),
                )
                .await
                .unwrap();
            ids.push(id.to_string());
        }
        let ids_json = serde_json::to_string(&ids).unwrap();
        let response = format!(
            r#"[{{"text":"Cache keys must be tenant-scoped.","agent_action":"Reuse tenant-scoped keys when repairing this cache failure.","evidence":{ids_json},"confidence":0.9,"supersedes":{ids_json},"priority":8}}]"#,
        );
        let _guard = MOCK_ENV_LOCK.lock().await;
        std::env::set_var(crate::consolidate::MOCK_RESPONSE_ENV, &response);

        let opts = ConsolidateOptions {
            tenant_id: Some("t".to_string()),
            project_id: Some("p".to_string()),
            project_dir: dir.path().to_path_buf(),
            max_region: 50,
            dry_run: false,
            background: false,
            force: false,
            promote: true,
            legacy_immediate: false,
        };
        let consolidator = MockEnvConsolidator;
        let result = consolidate_core(&store, opts, &consolidator).await.unwrap();
        std::env::remove_var(crate::consolidate::MOCK_RESPONSE_ENV);

        assert_eq!(result["consolidated"], 1);
        assert_eq!(result["tombstoned"], 12);
        // Project-scoped runs must not warn about tenant-wide writes.
        assert!(result.get("warning").is_none());

        // Sources are now superseded; the consolidated chunk carries
        // the provenance tag.
        let new_id = result["new_chunk_ids"][0].as_str().unwrap();
        let consolidated = store
            .get(&tenant, &ChunkId::parse(new_id).unwrap())
            .await
            .unwrap()
            .unwrap();
        assert!(consolidated.tags.iter().any(|t| t == "kind:consolidated"));
        assert!(consolidated
            .tags
            .iter()
            .any(|t| t.starts_with("ctx:subsystem:cache")));
    }

    #[tokio::test]
    async fn consolidate_stages_by_default_without_hiding_sources() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut ids = Vec::new();
        for i in 0..12 {
            ids.push(
                store
                    .add(
                        MemoryChunk::new(
                            tenant.clone(),
                            format!("staged lesson {i}: preserve the source until review"),
                            ChunkType::Summary,
                        )
                        .with_project(ProjectId::from("p")),
                    )
                    .await
                    .unwrap()
                    .to_string(),
            );
        }
        let ids_json = serde_json::to_string(&ids).unwrap();
        let response = format!(
            r#"[{{"text":"Sources remain visible until explicit review.","agent_action":"Verify the staged candidate before accepting its replacement.","evidence":{ids_json},"confidence":0.9,"supersedes":{ids_json},"priority":7}}]"#,
        );
        let _guard = MOCK_ENV_LOCK.lock().await;
        std::env::set_var(crate::consolidate::MOCK_RESPONSE_ENV, &response);
        let result = consolidate_core(
            &store,
            ConsolidateOptions {
                tenant_id: Some("t".to_string()),
                project_id: Some("p".to_string()),
                project_dir: dir.path().to_path_buf(),
                max_region: 50,
                dry_run: false,
                background: false,
                force: false,
                promote: false,
                legacy_immediate: false,
            },
            &MockEnvConsolidator,
        )
        .await
        .unwrap();
        std::env::remove_var(crate::consolidate::MOCK_RESPONSE_ENV);

        assert_eq!(result["state"], "validated");
        assert_eq!(result["staged"], 1);
        assert_eq!(result["consolidated"], 0);
        assert_eq!(result["tombstoned"], 0);
        for source_id in ids {
            let source = store
                .metadata()
                .get(&tenant, &ChunkId::parse(&source_id).unwrap())
                .unwrap()
                .unwrap();
            assert_eq!(source.status, ChunkStatus::Final);
        }
        let candidate_id = ChunkId::parse(
            result["candidate_chunk_ids"][0]
                .as_str()
                .expect("candidate id"),
        )
        .unwrap();
        assert_eq!(
            store
                .metadata()
                .get(&tenant, &candidate_id)
                .unwrap()
                .unwrap()
                .status,
            ChunkStatus::Candidate
        );
        let pending = run_consolidate_review(
            &store,
            ConsolidateReviewOptions {
                run_id: None,
                tenant_id: None,
                project_id: None,
                list: true,
                limit: 100,
                accept: false,
                reject: false,
            },
        )
        .await
        .unwrap();
        assert_eq!(pending["count"], 1);
        assert_eq!(pending["staged_runs"][0]["run_id"], result["run_id"]);
    }

    #[tokio::test]
    async fn dry_run_emits_prompt_without_calling_llm() {
        let dir = tempdir().unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        for i in 0..12 {
            store
                .add(
                    MemoryChunk::new(tenant.clone(), format!("chunk {i}"), ChunkType::Summary)
                        .with_project(ProjectId::from("p")),
                )
                .await
                .unwrap();
        }
        let opts = ConsolidateOptions {
            tenant_id: Some("t".to_string()),
            project_id: Some("p".to_string()),
            project_dir: dir.path().to_path_buf(),
            max_region: 50,
            dry_run: true,
            background: false,
            force: false,
            promote: false,
            legacy_immediate: false,
        };
        // A mock with no response env set must NOT be invoked.
        let consolidator = MockEnvConsolidator;
        let result = consolidate_core(&store, opts, &consolidator).await.unwrap();
        assert_eq!(result["dry_run"], true);
        assert!(result["prompt"].as_str().unwrap().contains("CHUNKS:"));
    }

    #[tokio::test]
    async fn explicit_tenant_ignores_scope_file_project() {
        let dir = tempdir().unwrap();
        // cwd scope file names another tenant/project; an explicit
        // --tenant-id must not inherit its project_id, so the region
        // stays tenant-wide (the old code narrowed it to 0 chunks).
        std::fs::create_dir_all(dir.path().join(".memd")).unwrap();
        std::fs::write(
            dir.path().join(".memd/project_scope.json"),
            r#"{"tenant_id":"other_tenant","project_id":"other_project"}"#,
        )
        .unwrap();
        let store = persistent_store(&dir.path().join("store"));
        let tenant = TenantId::new("t").unwrap();
        let mut ids = Vec::new();
        for i in 0..12 {
            let id = store
                .add(
                    MemoryChunk::new(tenant.clone(), format!("lesson {i}"), ChunkType::Summary)
                        .with_project(ProjectId::from("p")),
                )
                .await
                .unwrap();
            ids.push(id.to_string());
        }
        let ids_json = serde_json::to_string(&ids).unwrap();
        let response = format!(
            r#"[{{"text":"One tenant-wide lesson.","agent_action":"Reuse this tenant-wide lesson only after checking its sources.","evidence":{ids_json},"confidence":0.9,"supersedes":{ids_json},"priority":8}}]"#,
        );
        let _guard = MOCK_ENV_LOCK.lock().await;
        std::env::set_var(crate::consolidate::MOCK_RESPONSE_ENV, &response);
        let opts = ConsolidateOptions {
            tenant_id: Some("t".to_string()),
            project_id: None,
            project_dir: dir.path().to_path_buf(),
            max_region: 50,
            dry_run: false,
            background: false,
            force: false,
            promote: true,
            legacy_immediate: false,
        };
        let result = consolidate_core(&store, opts.clone(), &MockEnvConsolidator)
            .await
            .unwrap();
        std::env::remove_var(crate::consolidate::MOCK_RESPONSE_ENV);

        // Old behavior borrowed "other_project" from the scope file:
        // region_size 0 -> skipped:below_threshold.
        assert_eq!(result["region_size"], 12);
        assert_eq!(result["consolidated"], 1);
        assert!(result["project_id"].is_null());
        assert!(result["warning"]
            .as_str()
            .unwrap()
            .contains("project-scoped searches"));
        assert_eq!(result["tombstoned"], 0);
        let new_id = ChunkId::parse(result["new_chunk_ids"][0].as_str().unwrap()).unwrap();
        let tenant_wide = store.get(&tenant, &new_id).await.unwrap().unwrap();
        assert!(tenant_wide
            .tags
            .iter()
            .any(|tag| tag.starts_with("derives_from:")));
        assert!(!tenant_wide
            .tags
            .iter()
            .any(|tag| tag.starts_with("supersedes:")));

        // A tenant-wide synthesis is additional machine-wide knowledge, not
        // a project-visible replacement. Its project-scoped sources must stay
        // active until a replacement exists in the same project scope.
        for id in &ids {
            let resolved = store
                .get_with_lifecycle(&tenant, &ChunkId::parse(id).unwrap())
                .await
                .unwrap()
                .expect("tenant-wide source remains stored");
            assert_eq!(
                resolved.status,
                ChunkStatus::Final,
                "tenant-wide consolidation must retain project source {id}"
            );
            assert!(resolved.lifecycle.superseded_by.is_none());
        }

        let response = handle_memory_search(
            &store,
            SearchParams {
                tenant_id: "t".to_string(),
                project_id: Some("p".to_string()),
                query: "lesson".to_string(),
                k: 50,
                ..Default::default()
            },
        )
        .await
        .unwrap();
        let text = response["content"][0]["text"].as_str().unwrap();
        let payload: Value = serde_json::from_str(text).unwrap();
        let hits = payload["results"]
            .as_array()
            .unwrap()
            .iter()
            .filter_map(|row| row["chunk_id"].as_str())
            .collect::<std::collections::HashSet<_>>();
        for id in &ids {
            assert!(
                hits.contains(id.as_str()),
                "project search must retain tenant-wide source {id}"
            );
        }

        let repeated = consolidate_core(&store, opts, &MockEnvConsolidator)
            .await
            .unwrap();
        assert_eq!(repeated["skipped"], "below_threshold");
        assert_eq!(repeated["region_size"], 0);
    }
}