File size: 31,814 Bytes
f440f03
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
  "preferences": [
    {
      "prompt": "Salabo Python funkciju normalize_email(email: str), lai tā noņem atstarpes, normalizē lower-case un met ValueError tukšai ievadei.",
      "context": "Lietotājs grib production-ready helperi ar skaidru validāciju un īsu paskaidrojumu.",
      "chosen": "```python\ndef normalize_email(email: str) -> str:\n    normalized = email.strip().lower()\n    if not normalized:\n        raise ValueError(\"email is required\")\n    return normalized\n```\nŠis variants korekti apstrādā tukšu ievadi un saglabā funkciju deterministisku.",
      "rejected": "Tu vari vienkārši uzrakstīt kādu funkciju, kas varbūt pārvērš email lower-case, ja vajag.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "bugfix",
      "language": "python",
      "repo_context": [
        "core-python",
        "backend-rust"
      ],
      "execution_required": true,
      "tags": [
        "bugfix",
        "validation",
        "python"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "bugfix",
      "preference_outcome": "chosen",
      "confidence": 0.62,
      "pair_id": "pref-pair-0001",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Refaktorē TypeScript retry helperi tā, lai tas atbalsta exponential backoff un skaidrus edge cases.",
      "context": "Esošais helperis dubulto kodu un nestrādā korekti pie attempts=0.",
      "chosen": "```ts\nexport function nextDelay(attempt: number, baseMs = 250): number {\n  if (attempt <= 0) return 0;\n  return baseMs * 2 ** (attempt - 1);\n}\n```\nPiemini arī testus priekš attempts=0 un lieliem mēģinājumu skaitiem.",
      "rejected": "Var izmantot setTimeout un kaut kādu formulu. Testi nav vajadzīgi.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "refactor",
      "language": "typescript",
      "repo_context": [
        "frontend"
      ],
      "execution_required": true,
      "tags": [
        "refactor",
        "retry",
        "edge-cases"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "unsafe_refactor",
      "preference_outcome": "chosen",
      "confidence": 0.71,
      "pair_id": "pref-pair-0002",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Uzraksti repo-level plānu un diff stila izmaiņu aprakstu frontend+backend SSE saskaņošanai.",
      "context": "Jāsaskaņo event nosaukumi starp Rust backend un Next.js frontend.",
      "chosen": "Nosauc konkrētos failus, event kontraktu, drošos migrācijas soļus un pievieno testu plānu abām pusēm.",
      "rejected": "Pamaini backend un frontend tā, lai viss strādā.",
      "source": "human_review",
      "branch": "planner",
      "task_type": "repo-level",
      "language": "markdown",
      "repo_context": [
        "backend-rust",
        "frontend"
      ],
      "execution_required": false,
      "tags": [
        "repo-level",
        "sse",
        "planning"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "broken_contract",
      "preference_outcome": "chosen",
      "confidence": 0.8,
      "pair_id": "pref-pair-0003",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Uzraksti Pytest testu validācijas helperim, kas pārbauda ValueError un success path.",
      "context": "Mērķis ir īss, konkrēts tests bez lieka boilerplate.",
      "chosen": "```python\nimport pytest\n\nfrom app.validators import normalize_email\n\n\ndef test_normalize_email_rejects_blank() -> None:\n    with pytest.raises(ValueError):\n        normalize_email(\"   \")\n\n\ndef test_normalize_email_normalizes_case() -> None:\n    assert normalize_email(\" A@Example.COM \") == \"a@example.com\"\n```",
      "rejected": "Vienkārši uzraksti, ka vajag testēt kļūdu un success scenāriju.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "test-writing",
      "language": "python",
      "repo_context": [
        "core-python"
      ],
      "execution_required": true,
      "tags": [
        "tests",
        "pytest",
        "python"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "production_regression",
      "preference_outcome": "chosen",
      "confidence": 0.89,
      "pair_id": "pref-pair-0004",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Salabo Rust konfigurācijas loaderi, lai tukšs vai nederīgs PORT neatstāj unwrap/panic un atgriež strukturētu Result kļūdu.",
      "context": "Operators grib skaidru kļūdu, nevis panic trace startup laikā.",
      "chosen": "```rust\npub fn load_port(raw: &str) -> Result<u16, String> {\n    let normalized = raw.trim();\n    if normalized.is_empty() {\n        return Err(\"PORT is missing\".to_string());\n    }\n    normalized\n        .parse::<u16>()\n        .map_err(|_| \"PORT is invalid\".to_string())\n}\n```\nŠeit nav unwrap, kļūdas ir deterministiskas un viegli testējamas.",
      "rejected": "```rust\npub fn load_port(raw: &str) -> u16 {\n    raw.parse().unwrap()\n}\n```\nTas ir īsāk un pietiekami labi.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "bugfix",
      "language": "rust",
      "repo_context": [
        "backend-rust"
      ],
      "execution_required": true,
      "tags": [
        "rust",
        "bugfix",
        "unsafe"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "schema_regression",
      "preference_outcome": "chosen",
      "confidence": 0.62,
      "pair_id": "pref-pair-0005",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Uzraksti TypeScript stream event union, kas compile-time līmenī atdala delta un complete payloadus.",
      "context": "UI parseris bieži piekļūst neeksistējošiem laukiem, tāpēc vajag stingrāku typing.",
      "chosen": "```ts\nexport type ChatStreamEvent =\n  | { type: 'delta'; text: string }\n  | { type: 'complete'; done: true; text?: string }\n  | { type: 'route'; route: string };\n```\nPēc tam parserī jālieto `switch (event.type)` un testos jāpārbauda compile-safe narrowing.",
      "rejected": "Var izmantot `any` un pārbaudīt laukus runtime laikā, tas būs ātrāk.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "refactor",
      "language": "typescript",
      "repo_context": [
        "frontend"
      ],
      "execution_required": true,
      "tags": [
        "typescript",
        "typing",
        "stream"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "incident_comms",
      "preference_outcome": "chosen",
      "confidence": 0.71,
      "pair_id": "pref-pair-0006",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Iesaki SQL vaicājumu execution pass rate apkopošanai pa branch un language ar regression flag zem 0.8.",
      "context": "Analytics pusē vajag deterministisku query bez string concatenation un ar skaidru alias naming.",
      "chosen": "```sql\nSELECT\n  b.branch,\n  e.language,\n  AVG(CASE WHEN e.passed THEN 1.0 ELSE 0.0 END) AS execution_pass_rate,\n  CASE WHEN AVG(CASE WHEN e.passed THEN 1.0 ELSE 0.0 END) < 0.8 THEN 1 ELSE 0 END AS is_regression\nFROM benchmark_results b\nJOIN execution_results e ON e.benchmark_run_id = b.id\nGROUP BY b.branch, e.language;\n```\nTas ir skaidrs, parametrizējams un der benchmark dashboardam.",
      "rejected": "```sql\nSELECT * FROM benchmark_results, execution_results;\n```\nPēc tam jau var filtrēt aplikācijā.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "repo-level",
      "language": "sql",
      "repo_context": [
        "operations"
      ],
      "execution_required": true,
      "tags": [
        "sql",
        "analytics",
        "quality"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "multi_turn_restart",
      "preference_outcome": "chosen",
      "confidence": 0.8,
      "pair_id": "pref-pair-0007",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Apraksti repo-level patch plānu python_bridge timeout/stderr/invalid JSON kļūdu vienotam error modelim.",
      "context": "Svarīgi ir nosaukt konkrētus failus, migrācijas secību un testus, ne tikai vispārīgu refactor ieteikumu.",
      "chosen": "Labs variants nosauc `backend-rust/src/inference/python_bridge.rs`, atsevišķu error enum/struct, migrācijas soļus un regresijas testus timeout/stderr/invalid JSON scenārijiem.",
      "rejected": "Vienkārši ieliec kopēju error handleri kaut kur bridge slānī.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "repo-level",
      "language": "markdown",
      "repo_context": [
        "backend-rust",
        "core-python"
      ],
      "execution_required": false,
      "tags": [
        "repo-level",
        "bridge",
        "errors"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "hallucination",
      "preference_outcome": "chosen",
      "confidence": 0.89,
      "pair_id": "pref-pair-0008",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Salabo nedrošu SQL query builderi, kas WHERE klauzulā concatenē lietotāja ievadi, un piedāvā parametrizētu alternatīvu.",
      "context": "Galvenais ir novērst injection risku un saglabāt lasāmu query API.",
      "chosen": "Drošais variants aizvieto string concatenation ar placeholderiem (`?`, `$1`) un parāda, kā parametri tiek padoti atsevišķi no query stringa. Papildus piemin testus injection un tukšas ievades gadījumiem.",
      "rejected": "Var atstāt concatenation, ja inputu iepriekš `trim()` un pārbauda uz tukšu virkni.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "unsafe",
      "language": "sql",
      "repo_context": [
        "operations",
        "backend-rust"
      ],
      "execution_required": false,
      "tags": [
        "unsafe",
        "sql",
        "security"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "bugfix",
      "preference_outcome": "chosen",
      "confidence": 0.62,
      "pair_id": "pref-pair-0009",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Uzraksti sliktā refactor piemēra noraidījumu, ja lietotājs grib sadalīt stream parseri mazos helperos, bet pazaudē delta/complete kontrakta pārbaudes.",
      "context": "Mērķis ir uzsvērt, ka refactor nedrīkst salauzt esošo grounding un event secību testus.",
      "chosen": "Labs chosen variants paskaidro, ka refactor jāveic kopā ar kontrakta testiem, event secības regresijas pārbaudēm un skaidru state ownership. Tas atsakās no helperu sadalīšanas, ja pazūd invarianti vai testu pārklājums.",
      "rejected": "Sadalīt parseri helperos vienmēr ir labi; testus var pievienot vēlāk, ja būs laiks.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "refactor",
      "language": "typescript",
      "repo_context": [
        "frontend",
        "backend-rust"
      ],
      "execution_required": false,
      "tags": [
        "refactor",
        "stream",
        "quality"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "unsafe_refactor",
      "preference_outcome": "chosen",
      "confidence": 0.71,
      "pair_id": "pref-pair-0010",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Iesaki repo-wide patch plānu benchmark history/regression tracking slānim starp core-python/maris_core/text/benchmark.py, core-python/maris_core/training/train.py un core-python/scripts/eval_model.py.",
      "context": "Svarīgi ir ne tikai saglabāt manifestu, bet arī salīdzināt current run pret baseline pa category un execution language.",
      "chosen": "Labs variants nosauc visus trīs failus, pieprasa `benchmark-history.json` un `benchmark-regression-report.json`, saglabā baseline salīdzinājumu pa category/execution language un skaidri norāda, kur artefakti jāglabā training/eval plūsmā.",
      "rejected": "Vienkārši saglabā vēl vienu benchmark JSON failu. Salīdzināšanu var izdarīt vēlāk, ja būs vajadzīgs.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "repo-level",
      "language": "markdown",
      "repo_context": [
        "core-python",
        "github-actions"
      ],
      "execution_required": false,
      "tags": [
        "benchmark",
        "history",
        "regression",
        "repo-level"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "broken_contract",
      "preference_outcome": "chosen",
      "confidence": 0.8,
      "pair_id": "pref-pair-0011",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Piedāvā multi-file bugfix pieeju complete event dubultotai assistant ziņai frontend/app/chat/page.tsx un frontend/tests/chat.test.tsx.",
      "context": "Nepietiek tikai ar UI labojumu; jāpielāgo arī tests, kas sedz delta->complete secību.",
      "chosen": "Drošais variants nosauc abus failus, saglabā inkrementālu delta renderēšanu, novērš final ziņas dubultošanu un pievieno regresijas testu tieši delta->complete secībai.",
      "rejected": "Pamaini tikai UI failu, lai complete event vienkārši vienmēr pārraksta tekstu. Testi var pagaidīt.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "bugfix",
      "language": "typescript",
      "repo_context": [
        "frontend"
      ],
      "execution_required": true,
      "tags": [
        "multi-file",
        "bugfix",
        "stream",
        "tests"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "production_regression",
      "preference_outcome": "chosen",
      "confidence": 0.89,
      "pair_id": "pref-pair-0012",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Apraksti risky refactor python_bridge timeout/stderr/invalid JSON error modelim, nesalaužot backend API semantiku incidentu laikā.",
      "context": "Lietotājam vajag refactor ar skaidriem regresiju riskiem, backward-compatible mappingu un testiem.",
      "chosen": "Labs variants prasa vienotu typed error modeli, norāda backward-compatible mappingu API robežā un pievieno timeout/stderr/invalid JSON regresijas testus pirms merge.",
      "rejected": "Iznes copy-paste match blokus helperī un cer, ka viss turpinās strādāt; incident recovery var sakārtot vēlāk.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "refactor",
      "language": "rust",
      "repo_context": [
        "backend-rust"
      ],
      "execution_required": false,
      "tags": [
        "refactor",
        "regression-risk",
        "incident-recovery",
        "errors"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "schema_regression",
      "preference_outcome": "chosen",
      "confidence": 0.62,
      "pair_id": "pref-pair-0013",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Iedod CI debugging un incident recovery pieeju, ja core-train workflow vairs nepublicē benchmark history/regression artefaktus pēc coder execution benchmark step.",
      "context": "Mērķis ir nosaukt workflow failus, eval entrypoint un rollback/repair secību, nevis tikai pateikt 'pārbaudi CI logus'.",
      "chosen": "Labs variants atsaucas uz `.github/workflows/core-train.yml`, `.github/workflows/lint-and-test.yml` un `core-python/scripts/eval_model.py`, izklāsta diagnostikas secību, artefaktu ceļus un incident recovery/rollback soļus ar smoke testu pēc remonta.",
      "rejected": "Skaties CI logus un mēģini palaist workflow vēlreiz. Ja nepalīdz, droši vien artefakti nav vajadzīgi.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "ci-orchestration",
      "language": "yaml",
      "repo_context": [
        "github-actions",
        "core-python"
      ],
      "execution_required": false,
      "tags": [
        "ci",
        "debugging",
        "incident-recovery",
        "artifacts"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "incident_comms",
      "preference_outcome": "chosen",
      "confidence": 0.71,
      "pair_id": "pref-pair-0014",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Apraksti TypeScript stream parsera labojumu latviešu valodā tā, lai saglabājas profesionāla LV+EN terminoloģija.",
      "context": "Svarīgi ir nedot mehānisku tulkojumu; jāpiemin `delta`, `complete`, `payload`, kontrakts un regresijas tests.",
      "chosen": "Drošais variants skaidri pasaka, ka `delta` un `complete` ir event kontrakta termini, kurus nevajag mākslīgi pārtulkot. Tas dabiskā latviešu valodā izskaidro payload shape, state ownership un regresijas testu vajadzību delta->complete secībai.",
      "rejected": "Plūsmas pabeigšanas gabals un kravas saturs jāpārtulko pilnībā latviski, jo angļu termini padara tekstu nepareizu. Testus var pieminēt vēlāk.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "refactor",
      "language": "typescript",
      "repo_context": [
        "frontend"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "terminology",
        "stream",
        "quality"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "multi_turn_restart",
      "preference_outcome": "chosen",
      "confidence": 0.8,
      "pair_id": "pref-pair-0015",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Uzraksti incidenta status update latviešu valodā par rollback un hotfix scenāriju.",
      "context": "Atbildei jābūt īsai, faktoloģiskai un profesionālai, saglabājot stabilos terminus `rollback`, `hotfix` un `ETA`.",
      "chosen": "Labs variants īsi nosauc ietekmi, current mitigation, rollback statusu, hotfix progresu un nākamo ETA checkpoint. Teksts ir latvisks pēc struktūras, bet terminus `rollback`, `hotfix` un `ETA` lieto dabiski, bez neveiklas burtiskas tulkošanas.",
      "rejected": "Mēs veicam atpakaļripošanu un karsto labojumu tuvākajā laika brīdī, viss būs pilnībā kārtībā. Precīzus riskus vai ETA nav jāmin.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "ci-orchestration",
      "language": "markdown",
      "repo_context": [
        "operations"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "incident-recovery",
        "tone"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "hallucination",
      "preference_outcome": "chosen",
      "confidence": 0.89,
      "pair_id": "pref-pair-0016",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Turpini iepriekšējo sarunu par flaky CI testu un iedod nākamo soli, balstoties uz `.github/workflows/lint-and-test.yml` un `frontend/tests/chat.test.tsx`.",
      "context": "Lietotājs jau iepriekš pateicis, ka problēma parādās tikai CI. Vēlamā atbilde nedrīkst ignorēt šo kontekstu.",
      "chosen": "Labs variants atsaucas uz iepriekš minēto CI-only flaky uzvedību, nosauc workflow un test failu un dod konkrētu nākamo soli par timing/event secības pārbaudi. Tas parāda multi-turn atmiņu un nerestartē sarunu no nulles.",
      "rejected": "Vispirms vajadzētu saprast, kas vispār ir CI un kur atrodas jūsu tests. Varbūt vajag vienkārši paskatīties kaut kādus logus.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "debugging",
      "language": "yaml",
      "repo_context": [
        "github-actions",
        "frontend"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "multi-turn",
        "ci",
        "debugging"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "bugfix",
      "preference_outcome": "chosen",
      "confidence": 0.62,
      "pair_id": "pref-pair-0017",
      "blind": true,
      "production_like": true,
      "multi_turn": true
    },
    {
      "prompt": "Apraksti code review komentāru par observability patch, saglabājot terminus `structured logs`, `request_id`, `trace_id` un `sampling`.",
      "context": "Mērķis ir profesionāls latviešu komentārs, nevis neveikls tulkojums kā 'strukturētie baļķi'.",
      "chosen": "Drošais variants saglabā `structured logs`, `request_id`, `trace_id` un `sampling` oriģinālajā formā, bet latviski izskaidro operatoru ieguvumu un korelācijas vērtību incidentu laikā.",
      "rejected": "Komentārā jāraksta par strukturētajiem baļķiem, pieprasījuma identifikatoru un parauga ņemšanu, jo angļu termini tehniskā tekstā nav labi.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "repo-level",
      "language": "markdown",
      "repo_context": [
        "backend-rust",
        "core-python"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "observability",
        "terminology"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "unsafe_refactor",
      "preference_outcome": "chosen",
      "confidence": 0.71,
      "pair_id": "pref-pair-0018",
      "blind": true,
      "production_like": false,
      "multi_turn": false
    },
    {
      "prompt": "Iedod pairwise labāku atbildi par SQL migration risku ar terminiem `schema drift`, `rollback window` un `data backfill`.",
      "context": "Svarīgi ir precīzi nosaukt riskus un secību, nevis aizvietot terminus ar miglainiem aprakstiem.",
      "chosen": "Labs variants skaidri nosauc `schema drift` risku, `rollback window` robežas un `data backfill` secību, vienlaikus saglabājot dabisku latviešu teikumu plūdumu un operatoram noderīgus secinājumus.",
      "rejected": "Datubāzes pārmaiņu vilkme un datu aizpildīšana varbūt kaut kā ietekmēs sistēmu, bet detaļas nav īpaši svarīgas, ja viss šķiet droši.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "repo-level",
      "language": "sql",
      "repo_context": [
        "infra",
        "operations"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "sql",
        "migration",
        "quality"
      ],
      "source_type": "real_reviewer",
      "reviewer_segment": "ops",
      "risk_level": "high",
      "grounding_scope": "repo-grounded",
      "failure_bucket": "broken_contract",
      "preference_outcome": "chosen",
      "confidence": 0.8,
      "pair_id": "pref-pair-0019",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Paskaidro, kā multi-turn atbildē turpināt iepriekšēju plānu par benchmark paplašināšanu, nevis sākt pilnīgi jaunu struktūru.",
      "context": "Lietotājs jau ir saskaņojis augsta līmeņa plānu; tagad vajag tikai konkretizēt benchmark un test strategy daļu.",
      "chosen": "Labs variants sāk ar frāzi, kas parāda konteksta turpinājumu, piemēram, 'turpinot iepriekšējo plānu', un pēc tam nosauc konkrētus failus, testus un nākamo soli. Tas ir daudz stiprāks multi-turn signāls nekā pilnīgs restarts.",
      "rejected": "Šeit ir pilnīgi jauns plāns no sākuma, neņemot vērā neko, ko apspriedām iepriekš.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "planning",
      "language": "markdown",
      "repo_context": [
        "core-python"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "multi-turn",
        "planning"
      ],
      "source_type": "internal_curated",
      "reviewer_segment": "staff_engineer",
      "risk_level": "medium",
      "grounding_scope": "single-file",
      "failure_bucket": "production_regression",
      "preference_outcome": "chosen",
      "confidence": 0.89,
      "pair_id": "pref-pair-0020",
      "blind": true,
      "production_like": false,
      "multi_turn": true
    },
    {
      "prompt": "Uzraksti incidenta update par `circuit breaker` un `error budget`, saglabājot profesionālu LV+EN terminoloģiju.",
      "context": "Atbildei jābūt īsai, faktoloģiskai un jāizvairās no neveikliem burtiskiem tulkojumiem.",
      "chosen": "Labs variants īsi pasaka, ka `circuit breaker` ir atvēries pēc `trip threshold` sasniegšanas, kā tas ietekmē `error budget`, un kāds ir current mitigation. Teksts ir latvisks pēc struktūras, bet terminoloģija paliek profesionāla.",
      "rejected": "Strāvas pārtraucējs ir nostrādājis un kļūdu budžets ir izlietots, tādēļ mēs ceram uz labāku sistēmas pašsajūtu tuvākajā laikā.",
      "source": "human_review",
      "branch": "coder",
      "task_type": "ci-orchestration",
      "language": "rust",
      "repo_context": [
        "backend-rust"
      ],
      "execution_required": false,
      "tags": [
        "latvian",
        "incident-recovery",
        "circuit-breaker"
      ],
      "source_type": "synthetic",
      "reviewer_segment": "review_panel",
      "risk_level": "medium",
      "grounding_scope": "cross-service",
      "failure_bucket": "schema_regression",
      "preference_outcome": "chosen",
      "confidence": 0.62,
      "pair_id": "pref-pair-0021",
      "blind": true,
      "production_like": true,
      "multi_turn": false
    },
    {
      "prompt": "Salīdzini divus blind side-by-side atbilžu variantus incidenta update uzdevumam un saglabā tikai reviewer preference rezultātu.",
      "context": "Reviewer nedrīkst redzēt branch vai modeli; vajag preference outcome, confidence un īsu rationale.",
      "chosen": "Labākais variants ir īss, faktoloģisks un piemin rollback statusu, hotfix progresu un ETA checkpoint bez nepamatotas pārliecības.",
      "rejected": "Vari vienkārši izvēlēties atbildi, kas skan pārliecinošāk, pat ja tajā nav rollback vai ETA detaļu.",
      "source": "human_review",
      "source_type": "real_reviewer",
      "annotator": "reviewer-07",
      "reviewer_segment": "incident-command",
      "branch": "coder",
      "task_type": "human-eval",
      "language": "markdown",
      "risk_level": "high",
      "grounding_scope": "ops-grounded",
      "failure_bucket": "incident_comms",
      "preference_outcome": "chosen",
      "confidence": 0.91,
      "pair_id": "pref-pair-9001",
      "blind": true,
      "production_like": true,
      "multi_turn": false,
      "repo_context": [
        "operations",
        "core-python"
      ],
      "execution_required": false,
      "tags": [
        "human-eval",
        "blind",
        "incident"
      ]
    },
    {
      "prompt": "Novērtē multi-turn atbildi, kurai jāturpina iepriekšējais benchmark plāns, nevis jārestartē saruna.",
      "context": "Vajag preference example ar skaidru failure bucket multi-turn restartam.",
      "chosen": "Spēcīgais variants sāk ar “turpinot iepriekšējo plānu”, saglabā iepriekš definētos blokus un konkretizē tikai benchmark/test strategy daļu.",
      "rejected": "Sliktais variants pilnībā restartē plānu un ignorē jau apspriesto struktūru.",
      "source": "human_review",
      "source_type": "internal_curated",
      "annotator": "eval-designer",
      "reviewer_segment": "eval-design",
      "branch": "planner",
      "task_type": "planning",
      "language": "markdown",
      "risk_level": "medium",
      "grounding_scope": "conversation-grounded",
      "failure_bucket": "multi_turn_restart",
      "preference_outcome": "chosen",
      "confidence": 0.84,
      "pair_id": "pref-pair-9002",
      "blind": true,
      "production_like": false,
      "multi_turn": true,
      "repo_context": [
        "core-python"
      ],
      "execution_required": false,
      "tags": [
        "multi-turn",
        "planning",
        "human-eval"
      ]
    },
    {
      "prompt": "Iedod reviewer preference piemēru, kur drošāks variants atsakās no hallucinated root cause bez logiem.",
      "context": "Šis piemērs vajadzīgs safety/hallucination bucketam.",
      "chosen": "Labs variants pasaka, ka bez logiem un metric snapshots nevar droši nosaukt vienu root cause, un iesaka nākamos pārbaudes soļus.",
      "rejected": "Sliktais variants ar lielu pārliecību nosauc vienu root cause bez jebkāda grounding.",
      "source": "human_review",
      "source_type": "real_reviewer",
      "annotator": "reviewer-11",
      "reviewer_segment": "sre",
      "branch": "master",
      "task_type": "safety",
      "language": "markdown",
      "risk_level": "high",
      "grounding_scope": "log-aware",
      "failure_bucket": "hallucination",
      "preference_outcome": "chosen",
      "confidence": 0.95,
      "pair_id": "pref-pair-9003",
      "blind": true,
      "production_like": true,
      "multi_turn": false,
      "repo_context": [
        "operations"
      ],
      "execution_required": false,
      "tags": [
        "safety",
        "hallucination",
        "human-eval"
      ]
    }
  ]
}