Spaces:
Running
Running
File size: 63,708 Bytes
f3429ed 617cddd 0831f73 76e6245 1f489b4 a61d93b 2460c83 0831f73 6c26cfe 4a4fff2 6c26cfe 0831f73 a61d93b 1f489b4 dcc9c31 72d6be4 a8d33ef 1f489b4 dcc9c31 1f489b4 617cddd f3429ed 0831f73 617cddd 38be297 72d6be4 2311ff9 76e6245 2311ff9 38be297 2311ff9 38be297 2311ff9 76e6245 38be297 2311ff9 76e6245 1f489b4 76e6245 1f489b4 76e6245 1f489b4 76e6245 dcc9c31 1f489b4 2311ff9 1f489b4 76e6245 1f489b4 76e6245 1f489b4 76e6245 1f489b4 76e6245 1f489b4 76e6245 2311ff9 4a4fff2 0831f73 4a4fff2 0831f73 617cddd 0831f73 38be297 0831f73 38be297 617cddd 0831f73 617cddd 0831f73 617cddd 0831f73 a61d93b 0831f73 4a4fff2 617cddd 0831f73 4a4fff2 38be297 617cddd 38be297 0831f73 4a4fff2 617cddd 0831f73 4a4fff2 617cddd 38be297 617cddd 38be297 7dc6f47 0831f73 d28ed33 7dc6f47 12e7005 38be297 12e7005 4a4fff2 7dc6f47 4a4fff2 12e7005 d28ed33 12e7005 7dc6f47 12e7005 7dc6f47 12e7005 7dc6f47 12e7005 4a4fff2 7dc6f47 12e7005 7dc6f47 12e7005 7dc6f47 12e7005 0831f73 51096c9 617cddd 12e7005 51096c9 0831f73 617cddd 38be297 0831f73 617cddd 0831f73 617cddd 0831f73 617cddd 0831f73 617cddd 38be297 617cddd 38be297 617cddd 0831f73 38be297 0831f73 617cddd 0831f73 617cddd d28ed33 0831f73 617cddd 0831f73 76e6245 0831f73 617cddd 0831f73 6c26cfe 76e6245 6c26cfe 76e6245 6c26cfe 617cddd 76e6245 617cddd 6c26cfe 617cddd 6c26cfe 617cddd 76e6245 617cddd 76e6245 617cddd 6c26cfe 617cddd 6c26cfe 0831f73 617cddd 38be297 0831f73 38be297 617cddd 0831f73 6ce52dc 617cddd 6ce52dc 617cddd 6ce52dc 617cddd 6ce52dc 6c26cfe 4a4fff2 0831f73 a8d33ef 72d6be4 617cddd 0831f73 f3429ed a61d93b 0831f73 | 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 | # -*- coding: utf-8 -*-
"""Qualitative evidence preparation for SmartPaper's ASQ-oriented track."""
from __future__ import annotations
import hashlib
import json
import logging
import re
from config import APP_CONFIG
from core.study_design_contract import (
STUDY_DESIGN_CONTRACT_VERSION,
build_qualitative_contract,
build_user_research_intent,
)
class QualitativeEngine:
QUALITATIVE_APPENDIX_MANIFEST_VERSION = "qualitative-appendix-manifest-v1"
TARGET_APPENDIX_BINDING_VERSION = "target-qualitative-appendix-binding-v2"
SEMANTIC_CLAIM_FRAME_VERSION = "smartpaper-core-semantic-claim-frame-v1"
STUDY_DESIGN_CONTRACT_VERSION = STUDY_DESIGN_CONTRACT_VERSION
APPENDIX_FIELDS = ("theme", "concept", "stage", "case_label", "evidence_role")
APPENDIX_CANONICAL_LABEL_FIELDS = ("theme", "concept")
APPENDIX_HEADERS = (
"Higher-order Theme", "Empirical Concept", "Process Stage",
"Case/Actor", "Evidence Role", "Representative Quote",
)
_TARGET_APPENDIX_LOCALIZATION = {
"en": {
"heading": "Appendix: Data Structure and Evidence",
"headers": APPENDIX_HEADERS,
},
"zh": {
"heading": "附录：数据结构与证据",
"headers": ("高阶主题", "经验概念", "过程阶段", "案例/参与者", "证据角色", "代表性引文"),
},
"ru": {
"heading": "Приложение: Структура данных и доказательства",
"headers": ("Тема высшего порядка", "Эмпирическое понятие", "Этап процесса", "Случай/участник", "Роль доказательства", "Репрезентативная цитата"),
},
"kk": {
"heading": "Қосымша: Деректер құрылымы және дәлелдер",
"headers": ("Жоғары деңгейлі тақырып", "Эмпирикалық ұғым", "Процесс кезеңі", "Кейс/қатысушы", "Дәлел рөлі", "Репрезентативті дәйексөз"),
},
}
ALLOWED_ANALYTIC_DESIGNS = {
"Process analysis",
"Comparative case analysis",
"Grounded theory",
"Gioia-style inductive coding",
"Longitudinal qualitative analysis",
"Iterative thematic/process analysis",
}
def __init__(self, agent):
self.agent = agent
@staticmethod
def _clean_cell(value, default="Not specified"):
cleaned = str(value or default).replace("\n", " ").strip()
return cleaned or default
@classmethod
def _normalize_analytic_design(cls, value):
cleaned = re.sub(r"\s+", " ", str(value or "")).strip()
if not cleaned:
return "Iterative thematic/process analysis"
lookup = {item.lower(): item for item in cls.ALLOWED_ANALYTIC_DESIGNS}
return lookup.get(cleaned.lower(), "Iterative thematic/process analysis")
@classmethod
def _normalize_string_list(cls, values):
if values is None:
return []
if not isinstance(values, list):
values = [values]
out = []
for value in values:
cleaned = cls._clean_cell(value, "")
if cleaned and cleaned.casefold() != "not specified" and cleaned not in out:
out.append(cleaned)
return out
@classmethod
def _normalize_data_source_roles(cls, values):
rows = []
for item in values or []:
if not isinstance(item, dict):
continue
actor_role = cls._clean_cell(item.get("actor_role"), "")
data_source = cls._clean_cell(item.get("data_source"), "")
count = item.get("count")
if not actor_role or not data_source:
continue
if count is not None and (not isinstance(count, int) or count <= 0):
continue
row = {"actor_role": actor_role, "data_source": data_source, "count": count}
if row not in rows:
rows.append(row)
return rows
@classmethod
def build_qualitative_semantic_claim_frame(cls, metadata):
"""Freeze actor/source, process-evidence, and quote identities in one frame.
The frame is server-side authority. It contains no model-inferred prose
meaning and no verbatim quotation payload, only typed identities/hashes.
"""
meta = dict(metadata or {})
facts = dict(meta.get("study_facts") or {})
relations = []
for row in cls._normalize_data_source_roles(facts.get("data_source_roles") or []):
actor = cls._clean_cell(row.get("actor_role"), "")
source = cls._clean_cell(row.get("data_source"), "")
folded = source.casefold()
if "interview" in folded:
family = "interview"
elif "focus" in folded and "group" in folded:
family = "focus_group"
elif "observ" in folded:
family = "observation"
elif "log" in folded:
family = "log"
elif "document" in folded or "archiv" in folded or "record" in folded:
family = "document"
else:
family = "other"
relation = {
"actor_role": actor,
"data_source": source,
"source_family": family,
"count": row.get("count"),
"relation": "EVIDENCED_BY",
}
relation_id_payload = json.dumps(relation, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
relation["relation_id"] = hashlib.sha256(relation_id_payload.encode("utf-8")).hexdigest()[:16]
relations.append(relation)
quotes = []
for item in meta.get("evidence") or []:
if not isinstance(item, dict):
continue
quote_id = str(item.get("tag") or "").strip()
if not re.fullmatch(r"QUOTE-\d+", quote_id):
continue
quote_hash = str(item.get("quote_hash") or "").strip()
if not quote_hash and str(item.get("quote") or "").strip():
normalized = re.sub(r"\s+", " ", str(item.get("quote") or "")).strip()
quote_hash = hashlib.sha256(normalized.encode("utf-8")).hexdigest()
quotes.append({
"quote_id": quote_id,
"quote_hash": quote_hash,
"theme_id": str(item.get("theme_id") or "").strip(),
"stage": cls._clean_cell(item.get("stage")),
"case_label": cls._clean_cell(item.get("case_label")),
"evidence_role": str(item.get("evidence_role") or "typical").strip().lower(),
})
process_evidence = []
for theme in meta.get("themes") or []:
if not isinstance(theme, dict):
continue
theme_id = str(theme.get("theme_id") or "").strip()
if not theme_id:
continue
process_evidence.append({
"theme_id": theme_id,
"stages": [cls._clean_cell(x, "") for x in (theme.get("stages") or []) if cls._clean_cell(x, "")],
"has_deviant_evidence": bool(theme.get("has_deviant_evidence")),
"quote_ids": [
str(row.get("tag") or "").strip()
for row in (theme.get("evidence") or []) if isinstance(row, dict) and str(row.get("tag") or "").strip()
],
})
frame = {
"version": cls.SEMANTIC_CLAIM_FRAME_VERSION,
"mode": "qualitative",
"actor_data_source_relations": sorted(relations, key=lambda row: (row["actor_role"].casefold(), row["data_source"].casefold())),
"process_evidence": sorted(process_evidence, key=lambda row: row["theme_id"]),
"quote_identities": sorted(quotes, key=lambda row: row["quote_id"]),
}
canonical = json.dumps(frame, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
frame["frame_hash"] = hashlib.sha256(canonical.encode("utf-8")).hexdigest()
return frame
@classmethod
def build_scoped_evidence_md(
cls,
metadata,
*,
include_themes=True,
theme_titles=None,
assigned_evidence=None,
include_quotes=False,
include_quote_ids=True,
):
"""Render the minimum qualitative evidence needed by one drafting unit.
Study facts/theme summaries are an internal empirical channel. Verbatim
quotations are a separate channel and are exposed only to their owning
Findings unit (or the final appendix when explicitly requested by the
caller). This prevents a quotation visible in one unit from being
opportunistically reused elsewhere and keeps external literature from
becoming evidence for this study's own facts.
"""
metadata = metadata or {}
facts = metadata.get("study_facts") or {}
sources = facts.get("data_sources") or ["Not specified"]
if not isinstance(sources, list):
sources = [sources]
clean_sources = [cls._clean_cell(item) for item in sources]
participants = facts.get("participant_count")
interviews = facts.get("interview_count")
analytic_design = cls._normalize_analytic_design(
facts.get("analytic_design") or facts.get("analysis_method")
)
analytic_logic = cls._clean_cell(
facts.get("analytic_logic"),
"Iteratively compare empirical observations to refine a process/mechanism explanation.",
)
temporal_structure = cls._clean_cell(facts.get("temporal_structure"))
case_structure = cls._clean_cell(facts.get("case_structure"))
setting = cls._clean_cell(facts.get("research_setting"))
data_source_roles = cls._normalize_data_source_roles(facts.get("data_source_roles") or [])
trustworthiness = cls._normalize_string_list(facts.get("trustworthiness_procedures"))
ethics = cls._normalize_string_list(facts.get("ethics_procedures"))
reflexivity = cls._normalize_string_list(facts.get("reflexivity_procedures"))
lines = [
"### Frozen Qualitative Study Facts",
"| Field | Frozen value |",
"| :--- | :--- |",
f"| Research setting | {setting} |",
f"| Data sources | {', '.join(clean_sources)} |",
f"| Participants | {participants if participants is not None else 'Not specified'} |",
f"| Interviews | {interviews if interviews is not None else 'Not specified'} |",
f"| Analytic design | {analytic_design} |",
f"| Analytic logic | {analytic_logic} |",
f"| Temporal structure | {temporal_structure} |",
f"| Case structure | {case_structure} |",
f"| Trustworthiness procedures | {'; '.join(trustworthiness) if trustworthiness else 'Not specified'} |",
f"| Ethics procedures | {'; '.join(ethics) if ethics else 'Not specified'} |",
f"| Reflexivity procedures | {'; '.join(reflexivity) if reflexivity else 'Not specified'} |",
]
if data_source_roles:
lines.extend([
"",
"### Frozen Actor / Data-Source Relations",
"| Actor role | Data source | Count |",
"| :--- | :--- | :--- |",
])
lines.extend(
f"| {row['actor_role']} | {row['data_source']} | {row['count'] if row['count'] is not None else 'Not specified'} |"
for row in data_source_roles
)
lines.extend([
"",
"[QUALITATIVE EVIDENCE CHANNELS]",
"1. STUDY FACTS and THEME SUMMARIES below are this study's internal empirical evidence. Never attach an external [CITE-XX] tag to a sentence merely because it reports or interprets these frozen facts.",
"2. External literature may support only prior-theory, prior-evidence, or methodological-rationale claims. Keep those claims in distinct sentences from this study's setting, sample, data, events, quotations, and findings.",
"3. ASSIGNED QUOTATIONS, when present, are analysis-only verbatim evidence owned by the current stable theme. Do not reproduce the verbatim wording or any internal quotation identifier in generated prose; the Executor renders frozen quotations deterministically after narrative QA.",
"4. Study-owned trustworthiness, ethics, reflexivity, recruitment, consent, observation, or audit procedures may be stated as performed only when explicitly frozen above. Never convert general methodological best practice into a fact about this study.",
])
allowed_theme_titles = None
if theme_titles:
allowed_theme_titles = {
re.sub(r"\s+", " ", str(item or "")).strip().casefold()
for item in theme_titles
if str(item or "").strip()
}
if include_themes:
theme_rows = []
for theme in metadata.get("themes") or []:
if not isinstance(theme, dict):
continue
title = cls._clean_cell(theme.get("title"), "Emergent pattern")
if allowed_theme_titles is not None and title.casefold() not in allowed_theme_titles:
continue
concepts = ", ".join(
cls._clean_cell(item, "")
for item in (theme.get("concepts") or [])
if cls._clean_cell(item, "")
) or "Not specified"
stages = ", ".join(
cls._clean_cell(item, "")
for item in (theme.get("stages") or [])
if cls._clean_cell(item, "") and cls._clean_cell(item, "") != "Not specified"
) or "Not specified"
deviant = "yes" if theme.get("has_deviant_evidence") else "no"
theme_rows.append((title, concepts, stages, deviant))
if theme_rows:
lines.extend([
"",
"### Frozen Qualitative Theme Summary",
"| Higher-order Theme | Empirical Concepts | Process Stages | Deviant/Boundary Evidence Present |",
"| :--- | :--- | :--- | :--- |",
])
lines.extend(
f"| {title} | {concepts} | {stages} | {deviant} |"
for title, concepts, stages, deviant in theme_rows
)
if include_quotes:
evidence_items = assigned_evidence
if evidence_items is None:
evidence_items = metadata.get("evidence") or []
quote_rows = []
for item in evidence_items or []:
if not isinstance(item, dict):
continue
tag = cls._clean_cell(item.get("tag"), "")
quote = cls._clean_cell(item.get("quote"), "").replace('"', "'")
if not tag or not quote:
continue
stage = cls._clean_cell(item.get("stage"))
case_label = cls._clean_cell(item.get("case_label"))
evidence_role = str(item.get("evidence_role") or "typical").strip().lower()
if evidence_role not in {"typical", "deviant"}:
evidence_role = "typical"
quote_rows.append((tag, stage, case_label, evidence_role, quote))
if quote_rows:
lines.extend(["", "### Assigned Qualitative Quotations"])
if include_quote_ids:
lines.extend([
"| Tag | Process Stage | Case/Actor | Evidence Role | Verbatim Quote |",
"| :--- | :--- | :--- | :--- | :--- |",
])
lines.extend(
f'| {tag} | {stage} | {case_label} | {role} | "{quote}" |'
for tag, stage, case_label, role, quote in quote_rows
)
else:
lines.extend([
"| Process Stage | Case/Actor | Evidence Role | Verbatim Quote |",
"| :--- | :--- | :--- | :--- |",
])
lines.extend(
f'| {stage} | {case_label} | {role} | "{quote}" |'
for _tag, stage, case_label, role, quote in quote_rows
)
return "\n".join(lines).strip()
@staticmethod
def _appendix_target_language_code(target_lang):
value = str(target_lang or "").strip().casefold().replace("_", "-")
if value.startswith("zh") or "chinese" in value or "中文" in value or "简体" in value:
return "zh"
if value.startswith("ru") or "russian" in value or "рус" in value:
return "ru"
if value.startswith("kk") or value.startswith("kz") or "kazakh" in value or "қазақ" in value:
return "kk"
return "en"
@classmethod
def build_appendix_manifest(cls, metadata):
"""Freeze Appendix row/cell identity before any target-language operation."""
rows = []
seen = set()
for index, item in enumerate((metadata or {}).get("evidence") or [], start=1):
if not isinstance(item, dict):
continue
quote = cls._clean_cell(item.get("quote"), "")
if not quote:
continue
quote_id = str(item.get("tag") or "").strip().upper()
if not re.fullmatch(r"QUOTE-\d+", quote_id) or quote_id in seen:
return {}
seen.add(quote_id)
row = {
"row_id": quote_id,
"quote_id": quote_id,
"quote_hash": hashlib.sha256(re.sub(r"\s+", " ", quote).strip().encode("utf-8")).hexdigest(),
"theme": cls._clean_cell(item.get("theme"), "Emergent pattern"),
"concept": cls._clean_cell(item.get("concept"), "Empirical observation"),
"stage": cls._clean_cell(item.get("stage")),
"case_label": cls._clean_cell(item.get("case_label")),
"evidence_role": cls._clean_cell(
str(item.get("evidence_role") or "typical").strip().lower(), "typical"
),
"quote": quote,
}
rows.append(row)
if not rows:
return {}
manifest = {
"version": cls.QUALITATIVE_APPENDIX_MANIFEST_VERSION,
"columns": list(cls.APPENDIX_HEADERS),
"rows": rows,
}
manifest["manifest_hash"] = cls.appendix_manifest_hash(manifest)
return manifest
@classmethod
def appendix_manifest_hash(cls, manifest):
record = dict(manifest or {})
record.pop("manifest_hash", None)
payload = json.dumps(record, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
return hashlib.sha256(payload.encode("utf-8")).hexdigest()
@classmethod
def appendix_translation_cells(cls, manifest):
"""Return typed non-quote Appendix cells; IDs stay server-side."""
cells = {}
for row in (manifest or {}).get("rows") or []:
if not isinstance(row, dict):
continue
row_id = str(row.get("row_id") or "").strip()
if not row_id:
continue
for field in cls.APPENDIX_FIELDS:
value = cls._clean_cell(row.get(field), "Not specified")
cells[f"{row_id}:{field}"] = value
return cells
@classmethod
def appendix_translation_plan(cls, manifest, glossary=None):
"""Split Appendix cells into server-bound labels and model-owned surfaces.
Theme/concept cells are frozen research labels. When the task-level
canonical glossary already owns their target identity, the translation
model must not rewrite those labels or their internal anchors. Remaining
ordinary cells (stage/case/evidence-role) continue through the typed
translation compiler and are rebound by server-side cell ID.
"""
expected = cls.appendix_translation_cells(manifest)
glossary_value = dict(glossary or {})
folded_glossary = {
str(key or '').strip().casefold(): str(value or '').strip()
for key, value in glossary_value.items() if str(key or '').strip() and str(value or '').strip()
}
prebound = {}
model_cells = {}
for cell_id, source in expected.items():
field = str(cell_id or '').rsplit(':', 1)[-1]
canonical_target = str(glossary_value.get(source) or '').strip()
if not canonical_target:
canonical_target = folded_glossary.get(str(source or '').strip().casefold(), '')
if field in cls.APPENDIX_CANONICAL_LABEL_FIELDS and canonical_target:
prebound[cell_id] = canonical_target
else:
model_cells[cell_id] = source
if set(prebound) & set(model_cells) or set(prebound) | set(model_cells) != set(expected):
raise RuntimeError("qualitative_appendix_translation_plan_identity_mismatch")
return prebound, model_cells
@staticmethod
def _escape_appendix_markdown_cell(value):
return re.sub(r"\s+", " ", str(value or "").strip()).replace("|", r"\|")
@classmethod
def _validate_appendix_manifest(cls, manifest):
record = dict(manifest or {})
if record.get("version") != cls.QUALITATIVE_APPENDIX_MANIFEST_VERSION:
return False
if list(record.get("columns") or []) != list(cls.APPENDIX_HEADERS):
return False
rows = [row for row in (record.get("rows") or []) if isinstance(row, dict)]
ids = [str(row.get("row_id") or "") for row in rows]
quote_ids = [str(row.get("quote_id") or "") for row in rows]
if not rows or len(ids) != len(set(ids)) or ids != quote_ids:
return False
if any(not re.fullmatch(r"QUOTE-\d+", item) for item in ids):
return False
expected_hash = str(record.get("manifest_hash") or "")
return bool(expected_hash) and expected_hash == cls.appendix_manifest_hash(record)
@classmethod
def bind_target_language_appendix(cls, manifest, target_lang, cell_translations, quote_translations):
"""Render one six-column target Appendix from frozen semantic objects.
Markdown structure, row identity, column identity, and quote identity are
program-owned. The translation model may supply only ordinary cell
surfaces through ``cell_translations``.
"""
if not cls._validate_appendix_manifest(manifest):
raise RuntimeError("target_qualitative_appendix_manifest_invalid")
expected_cells = cls.appendix_translation_cells(manifest)
translations = dict(cell_translations or {})
if set(translations) != set(expected_cells):
missing = sorted(set(expected_cells) - set(translations))
extra = sorted(set(translations) - set(expected_cells))
raise RuntimeError(
f"target_qualitative_appendix_cell_identity_mismatch:missing={missing[:4]} extra={extra[:4]}"
)
quote_map = dict(quote_translations or {})
required_quotes = {str(row.get("quote_id") or "") for row in manifest.get("rows") or []}
missing_quotes = sorted(tag for tag in required_quotes if not str(quote_map.get(tag) or "").strip())
if missing_quotes:
raise RuntimeError(f"target_qualitative_appendix_quote_identity_missing:{missing_quotes[:6]}")
code = cls._appendix_target_language_code(target_lang)
profile = dict(cls._TARGET_APPENDIX_LOCALIZATION.get(code) or cls._TARGET_APPENDIX_LOCALIZATION["en"])
headers = list(profile.get("headers") or cls.APPENDIX_HEADERS)
heading = str(profile.get("heading") or cls._TARGET_APPENDIX_LOCALIZATION["en"]["heading"])
lines = [
"| " + " | ".join(cls._escape_appendix_markdown_cell(item) for item in headers) + " |",
"| :--- | :--- | :--- | :--- | :--- | :--- |",
]
for row in manifest.get("rows") or []:
row_id = str(row.get("row_id") or "")
values = [translations[f"{row_id}:{field}"] for field in cls.APPENDIX_FIELDS]
values.append(quote_map[str(row.get("quote_id") or "")])
lines.append(
"| " + " | ".join(cls._escape_appendix_markdown_cell(value) for value in values) + " |"
)
rendered = f"# {heading}\n\n" + "\n".join(lines)
cls.validate_bound_target_appendix(rendered, manifest, target_lang)
return rendered.strip()
@classmethod
def validate_bound_target_appendix(cls, text, manifest, target_lang):
"""Fail before DOCX packaging if semantic Appendix identity drifted."""
if not cls._validate_appendix_manifest(manifest):
raise RuntimeError("target_qualitative_appendix_manifest_invalid")
code = cls._appendix_target_language_code(target_lang)
profile = dict(cls._TARGET_APPENDIX_LOCALIZATION.get(code) or cls._TARGET_APPENDIX_LOCALIZATION["en"])
heading = str(profile.get("heading") or "")
headers = list(profile.get("headers") or cls.APPENDIX_HEADERS)
value = str(text or "")
heading_pattern = re.compile(rf"(?m)^#\s+{re.escape(heading)}\s*$")
matches = list(heading_pattern.finditer(value))
if len(matches) != 1:
raise RuntimeError("target_qualitative_appendix_heading_identity_missing_or_ambiguous")
tail = value[matches[0].end():]
table_blocks = re.findall(r"(?m)(?:^[ \t]*\|[^\n]+\|[ \t]*$\n?)+", tail)
valid_blocks = []
for block in table_blocks:
lines = [line.strip() for line in block.splitlines() if line.strip().startswith("|")]
if not lines:
continue
if all(len(re.findall(r"(?<!\\)\|", line)) == 7 for line in lines):
valid_blocks.append(lines)
if len(valid_blocks) != 1:
raise RuntimeError("target_qualitative_appendix_semantic_table_identity_missing_or_ambiguous")
lines = valid_blocks[0]
expected_rows = len(manifest.get("rows") or []) + 2
if len(lines) != expected_rows:
raise RuntimeError(
f"target_qualitative_appendix_row_identity_mismatch:expected={expected_rows} got={len(lines)}"
)
expected_header = "| " + " | ".join(cls._escape_appendix_markdown_cell(item) for item in headers) + " |"
if lines[0] != expected_header or not re.fullmatch(
r"\|\s*:---\s*\|\s*:---\s*\|\s*:---\s*\|\s*:---\s*\|\s*:---\s*\|\s*:---\s*\|",
lines[1],
):
raise RuntimeError("target_qualitative_appendix_column_identity_mismatch")
if re.search(r"\bQUOTE-\d+\b", "\n".join(lines), flags=re.I):
raise RuntimeError("target_qualitative_appendix_internal_quote_id_leak")
return True
@classmethod
def build_appendix_evidence_md(cls, metadata):
"""Render the reader-facing English Appendix from the same frozen manifest."""
manifest = cls.build_appendix_manifest(metadata)
if not cls._validate_appendix_manifest(manifest):
raise RuntimeError("qualitative_appendix_manifest_invalid")
lines = [
"| " + " | ".join(cls.APPENDIX_HEADERS) + " |",
"| :--- | :--- | :--- | :--- | :--- | :--- |",
]
for row in manifest.get("rows") or []:
values = [row.get(field) for field in cls.APPENDIX_FIELDS] + [f'"{row.get("quote")}"']
lines.append(
"| " + " | ".join(cls._escape_appendix_markdown_cell(value) for value in values) + " |"
)
return "\n".join(lines).strip()
@staticmethod
def _intent_data_source_family(value):
folded = re.sub(r"\s+", " ", str(value or "")).strip().casefold()
if re.search(r"interview|访谈|интервью|сұхбат", folded):
return "interview"
if re.search(r"focus.{0,4}(?:group|групп|топ)|焦点小组", folded):
return "focus_group"
if re.search(r"observ|观察|наблюден|бақылау", folded):
return "observation"
if re.search(r"archiv|document|档案|文档|文件|архив|документ|мұрағат|құжат", folded):
return "document"
if re.search(r"field.{0,4}note|田野笔记|полев.{0,6}замет|далалық.{0,6}жазб", folded):
return "field_note"
if re.search(r"digital.{0,4}trace|数字痕迹|цифров.{0,6}след|цифрлық.{0,6}із", folded):
return "digital_trace"
if re.search(r"\blog(?:s)?\b|日志|\bлог\b", folded):
return "log"
return "other"
def _build_request(self, topic, keywords, raw_interviews):
user_intent = build_user_research_intent(topic, keywords, mode="qualitative")
intent_block = json.dumps(user_intent, ensure_ascii=False)
schema = """Return ONLY one valid JSON object with this schema:
{
"study_facts": {
"research_setting": "string or Not specified",
"data_sources": ["string"],
"data_source_roles": [{"actor_role":"string", "data_source":"string", "count":integer or null}],
"participant_count": integer or null,
"interview_count": integer or null,
"trustworthiness_procedures": ["explicitly supported/frozen procedure"],
"ethics_procedures": ["explicitly supported/frozen procedure"],
"reflexivity_procedures": ["explicitly supported/frozen procedure"],
"analytic_design": "one allowed design",
"analytic_logic": "one concise sentence",
"temporal_structure": "string or Not specified",
"case_structure": "string or Not specified"
},
"evidence": [
{
"theme": "higher-order pattern/theme",
"concept": "empirical concept",
"quote": "verbatim quotation",
"stage": "temporal/process stage or Not specified",
"case_label": "generic case/actor label or Not specified",
"evidence_role": "typical or deviant"
}
]
}
Allowed analytic designs: Process analysis; Comparative case analysis; Grounded theory; Gioia-style inductive coding; Longitudinal qualitative analysis; Iterative thematic/process analysis.
"""
if raw_interviews and len(raw_interviews) > 100:
return f"""Topic: {topic}
Keywords: {keywords}
<USER_RESEARCH_INTENT>{intent_block}</USER_RESEARCH_INTENT>
Analyze the supplied interview or field material for an ASQ-oriented qualitative manuscript.
Rules:
0. USER RESEARCH INTENT AUTHORITY: Preserve the user's research phenomenon, explicitly requested qualitative design, data-source family, temporal scope, actor/population, and context when the supplied material supports them. Never silently substitute a different method or phenomenon merely because literature or a generic template is easier. If the supplied material genuinely conflicts with the requested design, preserve evidence truth rather than inventing support; the deterministic validator will surface the conflict.
1. Treat the supplied material as the only source of study facts. Use null, empty arrays, or "Not specified" for any setting, sample, count, date, case, actor/data-source relation, trustworthiness procedure, ethics procedure, reflexivity procedure, or data source not explicitly supported by the material.
2. Extract 8-15 representative quotations VERBATIM. Never repair, paraphrase, translate, combine, or invent a quotation.
3. Choose the analytic design that best fits the actual material. Do NOT default to Gioia. Use Gioia-style coding only if the evidence structure genuinely warrants it.
4. Organize evidence around a central organizational puzzle and, when the material supports it, preserve temporal stages, actor/case contrasts, process transitions, and at least one deviant or boundary observation.
5. Do not manufacture a longitudinal sequence when the material is cross-sectional, and do not manufacture multiple cases when only one is present.
6. Freeze actor-to-data-source relations explicitly in data_source_roles whenever the material supports them (for example extension officers -> interviews; farmers -> focus groups). Never infer that an actor participated in a data source not supported by the material. The study_facts object and quotations will become the frozen single source of truth for the entire manuscript.
{schema}
<QUALITATIVE_MATERIAL>
{raw_interviews[:30000]}
</QUALITATIVE_MATERIAL>
"""
return f"""Topic: {topic}
Keywords: {keywords}
<USER_RESEARCH_INTENT>{intent_block}</USER_RESEARCH_INTENT>
Construct one internally coherent qualitative evidence packet for a complete ASQ-oriented manuscript because no interview material was supplied.
Rules:
0. USER RESEARCH INTENT AUTHORITY: The topic/keywords define what the study is asked to investigate. If they explicitly request an analytic design (for example grounded theory, thematic analysis, comparative case, process, longitudinal, or Gioia) or a data-source family (for example interviews, focus groups, observation, documents), preserve that request in the generated evidence packet. Sparse literature is not a reason to substitute another method.
1. Choose one generic organizational research setting directly relevant to the topic and keep it fixed. Do not name a real organization, institution, or identifiable person.
2. Freeze one participant count, interview count, data-source set, actor-to-data-source relation map, analytic design, temporal/case structure, and analytic logic. Keep them internally consistent. Every actor/data-source relationship later stated in the paper must be represented in data_source_roles.
3. Produce exactly 12 natural representative quotations distributed across at least three higher-order themes. Quotations must sound like distinct participants rather than repeated template variants.
4. Choose the analytic design that best fits the constructed evidence; do NOT default to Gioia. Use process, comparative-case, grounded-theory, longitudinal, or Gioia-style logic only when the packet supports it.
5. Include at least one clearly labelled deviant/boundary observation so the eventual theory is not built only from confirming examples.
6. The evidence should support a process/mechanism explanation, not merely descriptive topic categories.
7. Freeze concise, internally coherent trustworthiness, ethics, and reflexivity procedures in their typed arrays so later Methods prose has an explicit study-owned source. Do not name a real ethics board, institution, vendor, approval number, or identifiable organization.
8. The study_facts object and evidence packet will become the frozen single source of truth for the entire manuscript.
9. Avoid decorative language, generic AI phrasing, and repetitive quotation patterns.
10. Keep the JSON compact: each representative quotation should be one concise complete sentence; keep labels short, analytic_logic to one sentence, and each procedure array to at most three concise items. Do not add explanatory prose outside the required fields.
{schema}
"""
@staticmethod
def _quote_has_substantive_content(value):
"""Language-aware minimum-content gate for participant quotations.
The previous whitespace-token rule rejected ordinary Chinese sentences
because an entire CJK sentence can contain zero spaces. Article Agent
supports Chinese/English/Russian/Kazakh, so accept either four or more
whitespace-delimited tokens or a substantive CJK character sequence.
"""
quote = re.sub(r"\s+", " ", str(value or "")).strip()
if not quote:
return False
if len(quote.split()) >= 4:
return True
cjk_chars = re.findall(r"[\u3400-\u4dbf\u4e00-\u9fff]", quote)
return len(cjk_chars) >= 8
def _study_facts_validation_error(self, facts, *, has_user_data, user_intent=None):
"""Validate frozen qualitative study facts without repairing model output."""
if not isinstance(facts, dict):
return "invalid_study_facts_schema"
for key in ("participant_count", "interview_count"):
value = facts.get(key)
if value is not None and (not isinstance(value, int) or value <= 0):
return f"invalid_{key}"
raw_relations = facts.get("data_source_roles") or []
if raw_relations and not isinstance(raw_relations, list):
return "invalid_data_source_roles"
declared_sources = [
re.sub(r"\s+", " ", str(item or "")).strip().casefold()
for item in (facts.get("data_sources") or [])
if str(item or "").strip()
]
for item in raw_relations:
if not isinstance(item, dict) or not str(item.get("actor_role") or "").strip() or not str(item.get("data_source") or "").strip():
return "invalid_data_source_role_item"
count = item.get("count")
if count is not None and (not isinstance(count, int) or count <= 0):
return "invalid_data_source_role_count"
relation_source = re.sub(r"\s+", " ", str(item.get("data_source") or "")).strip().casefold()
if declared_sources and not any(
relation_source == source or relation_source in source or source in relation_source
for source in declared_sources
):
return "data_source_role_not_in_declared_sources"
for key in ("trustworthiness_procedures", "ethics_procedures", "reflexivity_procedures"):
value = facts.get(key) or []
if not isinstance(value, list):
return f"invalid_{key}"
if any(not str(item or "").strip() for item in value):
return f"invalid_{key}"
if not has_user_data:
if not raw_relations:
return "generated_packet_missing_data_source_roles"
for key in ("trustworthiness_procedures", "ethics_procedures", "reflexivity_procedures"):
if not (facts.get(key) or []):
return f"generated_packet_missing_{key}"
intent = dict(user_intent or {})
requested_design = str(intent.get("qualitative_design") or "").strip()
if requested_design:
observed_design = self._normalize_analytic_design(facts.get("analytic_design") or facts.get("analysis_method"))
if observed_design.casefold() != requested_design.casefold():
return f"user_intent_analytic_design_mismatch:{observed_design}->{requested_design}"
requested_sources = set(intent.get("qualitative_data_source_families") or [])
if requested_sources:
observed_sources = {self._intent_data_source_family(item) for item in (facts.get("data_sources") or [])}
missing_sources = sorted(requested_sources - observed_sources)
if missing_sources:
return "user_intent_data_source_mismatch:" + ",".join(missing_sources)
return ""
def _qualitative_packet_validation_error(self, data, *, has_user_data, user_intent=None):
"""Return a bounded local validation reason, or an empty string when safe.
The validator never repairs JSON, invents quotations, or mutates model
output. A completed but unsafe response is discarded and, when
appropriate, regenerated through smaller structured stages.
"""
if not isinstance(data, dict):
return "packet_not_json_object"
facts = data.get("study_facts") or {}
evidence = data.get("evidence") or data.get("quotes") or []
if not isinstance(facts, dict) or not isinstance(evidence, list):
return "invalid_packet_schema"
facts_reason = self._study_facts_validation_error(
facts,
has_user_data=has_user_data,
user_intent=user_intent,
)
if facts_reason:
return facts_reason
seen_quotes = set()
theme_keys = set()
retained = 0
has_deviant = False
for item in evidence[:15]:
if not isinstance(item, dict):
continue
quote = self._clean_cell(item.get("quote"), "").replace('"', "'")
normalized_quote = re.sub(r"\s+", " ", quote).strip().casefold()
if not self._quote_has_substantive_content(quote) or normalized_quote in seen_quotes:
continue
seen_quotes.add(normalized_quote)
retained += 1
theme = self._clean_cell(item.get("theme"), "Emergent pattern")
theme_keys.add(re.sub(r"\s+", " ", theme).strip().casefold())
if str(item.get("evidence_role") or "typical").strip().lower() == "deviant":
has_deviant = True
validation = APP_CONFIG.get("manuscript_validation", {})
minimum_themes = int(validation.get("minimum_qualitative_themes", 3))
minimum_quotes = int(validation.get("minimum_qualitative_quotes", 8))
if len(theme_keys) < minimum_themes or retained < minimum_quotes:
return f"incomplete_packet_themes_{len(theme_keys)}_quotes_{retained}"
if not has_user_data and not has_deviant:
return "generated_packet_missing_deviant_evidence"
return ""
@staticmethod
def _theme_plan_validation_error(value):
if not isinstance(value, list):
return "invalid_theme_plan"
cleaned = [re.sub(r"\s+", " ", str(item or "")).strip() for item in value]
cleaned = [item for item in cleaned if item]
unique = {item.casefold() for item in cleaned}
if len(unique) < 3:
return "incomplete_theme_plan"
return ""
def _call_qualitative_json(self, system_prompt, prompt, *, max_tokens):
"""Make one bounded structured call and parse it strictly; never repair JSON."""
response = self.agent.call_deepseek(
system_prompt,
prompt,
max_tokens=max_tokens,
temp=0.0,
model_role="reasoning",
thinking=False,
allow_fallback=False,
response_format={"type": "json_object"},
logical_node="structured_json",
)
response = re.sub(
r"<think>.*?</think>",
"",
str(response or ""),
flags=re.DOTALL,
).strip()
return self.agent._extract_json_from_text(response)
def _build_study_facts_stage_request(self, topic, keywords, raw_interviews, user_intent):
intent_block = json.dumps(user_intent, ensure_ascii=False)
facts_schema = """Return ONLY one valid JSON object with this schema:
{
"study_facts": {
"research_setting": "string or Not specified",
"data_sources": ["string"],
"data_source_roles": [{"actor_role":"string", "data_source":"string", "count":integer or null}],
"participant_count": integer or null,
"interview_count": integer or null,
"trustworthiness_procedures": ["concise procedure"],
"ethics_procedures": ["concise procedure"],
"reflexivity_procedures": ["concise procedure"],
"analytic_design": "one allowed design",
"analytic_logic": "one concise sentence",
"temporal_structure": "string or Not specified",
"case_structure": "string or Not specified"
},
"theme_plan": ["concise theme 1", "concise theme 2", "concise theme 3"]
}
Allowed analytic designs: Process analysis; Comparative case analysis; Grounded theory; Gioia-style inductive coding; Longitudinal qualitative analysis; Iterative thematic/process analysis.
"""
if raw_interviews and len(raw_interviews) > 100:
material_block = f"""Use ONLY the supplied qualitative material as the source of study facts. Use null, empty arrays, or \"Not specified\" when the material does not support a fact. Preserve the user's requested design/data-source family only when evidence truth permits it. Do not invent actors, counts, procedures, cases, dates, or settings. The theme_plan must contain at least three concise evidence-grounded higher-order themes and must not contain quotations.
<QUALITATIVE_MATERIAL>
{raw_interviews[:30000]}
</QUALITATIVE_MATERIAL>"""
else:
material_block = """No interview material was supplied. Construct one generic, internally coherent organizational qualitative study directly relevant to the topic. Do not name a real organization or identifiable person. Freeze one coherent setting, sample/data-source structure, analytic design, procedures, temporal/case structure, and at least three concise higher-order themes. Keep every field compact."""
return f"""Topic: {topic}
Keywords: {keywords}
<USER_RESEARCH_INTENT>{intent_block}</USER_RESEARCH_INTENT>
This is FALLBACK STAGE A. Produce STUDY FACTS and a compact THEME PLAN only. Do not produce quotations or an evidence array.
{material_block}
{facts_schema}
"""
def _build_evidence_stage_request(
self,
topic,
keywords,
raw_interviews,
user_intent,
*,
facts,
theme_plan,
prior_reason="",
):
intent_block = json.dumps(user_intent, ensure_ascii=False)
facts_block = json.dumps(facts, ensure_ascii=False, separators=(",", ":"))
themes_block = json.dumps(theme_plan, ensure_ascii=False)
retry_note = (
f"\nThe previous completed Stage B response failed deterministic validation with reason={prior_reason}. "
"Generate a fresh evidence object from the same frozen facts and original inputs; do not repair or continue the previous response."
if prior_reason
else ""
)
if raw_interviews and len(raw_interviews) > 100:
evidence_rules = f"""Use ONLY the supplied qualitative material. Return 8-12 representative quotations VERBATIM; never repair, paraphrase, translate, combine, or invent a quotation. Use at least three themes from the frozen theme plan when supported. Preserve stage/case labels only when supported by the material.
<QUALITATIVE_MATERIAL>
{raw_interviews[:30000]}
</QUALITATIVE_MATERIAL>"""
else:
evidence_rules = """Generate exactly 12 concise natural representative quotations distributed across at least three themes from the frozen theme plan. Quotations must sound like distinct participants, each should be one concise complete sentence, and at least one evidence item must be clearly labelled deviant. Do not change or contradict any frozen study fact."""
return f"""Topic: {topic}
Keywords: {keywords}
<USER_RESEARCH_INTENT>{intent_block}</USER_RESEARCH_INTENT>
This is FALLBACK STAGE B. The study facts below are frozen authority and MUST NOT be changed.
<FROZEN_STUDY_FACTS>{facts_block}</FROZEN_STUDY_FACTS>
<FROZEN_THEME_PLAN>{themes_block}</FROZEN_THEME_PLAN>
{evidence_rules}
{retry_note}
Return ONLY one valid JSON object with this schema:
{{
"evidence": [
{{
"theme": "one frozen theme",
"concept": "concise empirical concept",
"quote": "representative quotation",
"stage": "temporal/process stage or Not specified",
"case_label": "generic case/actor label or Not specified",
"evidence_role": "typical or deviant"
}}
]
}}
No Markdown and no explanatory prose.
"""
def _generate_staged_qualitative_packet(
self,
topic,
keywords,
raw_interviews,
*,
has_user_data,
user_intent,
trigger_reason,
):
"""Bounded small-JSON fallback after a completed unsafe full packet."""
system_prompt = (
"You are a rigorous qualitative research analyst. "
"Evidence fidelity has priority over narrative completeness."
)
facts_data = None
facts = None
theme_plan = None
facts_reason = "packet_not_json_object"
facts_prompt = self._build_study_facts_stage_request(
topic, keywords, raw_interviews, user_intent
)
for attempt in range(2):
prompt = facts_prompt
if attempt:
prompt += (
"\n\nLOCAL STAGE-A RETRY: The previous completed response failed deterministic "
f"validation with reason={facts_reason}. Generate a fresh complete Stage A JSON object "
"from the ORIGINAL inputs. Do not repair or continue the previous response."
)
facts_data = self._call_qualitative_json(
system_prompt,
prompt,
max_tokens=2600,
)
if not isinstance(facts_data, dict):
facts_reason = "packet_not_json_object"
continue
facts = facts_data.get("study_facts")
theme_plan = facts_data.get("theme_plan")
facts_reason = self._study_facts_validation_error(
facts,
has_user_data=has_user_data,
user_intent=user_intent,
) or self._theme_plan_validation_error(theme_plan)
if not facts_reason:
break
logging.warning(
"Qualitative staged fallback facts retry | attempt=%s/2 | trigger=%s | reason=%s",
attempt + 1,
trigger_reason,
facts_reason,
)
if facts_reason:
raise RuntimeError(
"Qualitative evidence facts fallback failed deterministic validation: "
f"{facts_reason}."
)
evidence_reason = "packet_not_json_object"
for attempt in range(2):
evidence_prompt = self._build_evidence_stage_request(
topic,
keywords,
raw_interviews,
user_intent,
facts=facts,
theme_plan=theme_plan,
prior_reason=evidence_reason if attempt else "",
)
evidence_data = self._call_qualitative_json(
system_prompt,
evidence_prompt,
max_tokens=4600,
)
evidence = evidence_data.get("evidence") if isinstance(evidence_data, dict) else None
merged = {"study_facts": facts, "evidence": evidence}
evidence_reason = self._qualitative_packet_validation_error(
merged,
has_user_data=has_user_data,
user_intent=user_intent,
)
if not evidence_reason:
return merged
logging.warning(
"Qualitative staged fallback evidence retry | attempt=%s/2 | trigger=%s | reason=%s",
attempt + 1,
trigger_reason,
evidence_reason,
)
raise RuntimeError(
"Qualitative evidence fallback failed deterministic validation after bounded regeneration: "
f"{evidence_reason}."
)
def _generate_qualitative_packet(self, topic, keywords, raw_interviews, *, has_user_data):
"""Generate one compact packet, then fall back to bounded smaller JSON stages."""
system_prompt = (
"You are a rigorous qualitative research analyst. "
"Evidence fidelity has priority over narrative completeness."
)
user_intent = build_user_research_intent(topic, keywords, mode="qualitative")
base_prompt = self._build_request(topic, keywords, raw_interviews)
data = self._call_qualitative_json(
system_prompt + "\n\nDIRECT OUTPUT MODE: Produce the complete compact evidence packet directly without extended hidden reasoning.",
base_prompt,
max_tokens=8000,
)
reason = self._qualitative_packet_validation_error(
data,
has_user_data=has_user_data,
user_intent=user_intent,
)
if not reason:
return data
logging.warning(
"Qualitative evidence full packet unsafe; entering staged fallback | reason=%s",
reason,
)
return self._generate_staged_qualitative_packet(
topic,
keywords,
raw_interviews,
has_user_data=has_user_data,
user_intent=user_intent,
trigger_reason=reason,
)
def resolve_qualitative_data(self, topic, keywords, raw_interviews=""):
has_user_data = bool(raw_interviews and len(raw_interviews) > 100)
# R3-Free engine code is headless. Streamlit rendering belongs to app.py;
# this method only produces frozen evidence + metadata.
data = self._generate_qualitative_packet(
topic, keywords, raw_interviews, has_user_data=has_user_data
)
facts = data.get("study_facts") or {}
evidence = data.get("evidence") or data.get("quotes") or []
data_sources = facts.get("data_sources") or ["Not specified"]
if not isinstance(data_sources, list):
data_sources = [data_sources]
setting = self._clean_cell(facts.get("research_setting"))
clean_sources = [self._clean_cell(item) for item in data_sources]
sources = ", ".join(clean_sources)
data_source_roles = self._normalize_data_source_roles(facts.get("data_source_roles") or [])
trustworthiness = self._normalize_string_list(facts.get("trustworthiness_procedures"))
ethics = self._normalize_string_list(facts.get("ethics_procedures"))
reflexivity = self._normalize_string_list(facts.get("reflexivity_procedures"))
participants = facts.get("participant_count")
interviews = facts.get("interview_count")
analytic_design = self._normalize_analytic_design(
facts.get("analytic_design") or facts.get("analysis_method")
)
analytic_logic = self._clean_cell(
facts.get("analytic_logic"),
"Iteratively compare empirical observations to refine a process/mechanism explanation.",
)
temporal_structure = self._clean_cell(facts.get("temporal_structure"))
case_structure = self._clean_cell(facts.get("case_structure"))
for value, label in ((participants, "participant_count"), (interviews, "interview_count")):
if value is not None and (not isinstance(value, int) or value <= 0):
raise RuntimeError(f"Qualitative evidence packet has invalid {label}.")
md = (
"### Frozen Qualitative Study Design\n"
"| Field | Frozen value |\n"
"| :--- | :--- |\n"
f"| Research setting | {setting} |\n"
f"| Data sources | {sources} |\n"
f"| Participants | {participants if participants is not None else 'Not specified'} |\n"
f"| Interviews | {interviews if interviews is not None else 'Not specified'} |\n"
f"| Analytic design | {analytic_design} |\n"
f"| Analytic logic | {analytic_logic} |\n"
f"| Temporal structure | {temporal_structure} |\n"
f"| Case structure | {case_structure} |\n"
f"| Trustworthiness procedures | {'; '.join(trustworthiness) if trustworthiness else 'Not specified'} |\n"
f"| Ethics procedures | {'; '.join(ethics) if ethics else 'Not specified'} |\n"
f"| Reflexivity procedures | {'; '.join(reflexivity) if reflexivity else 'Not specified'} |\n\n"
"[QUALITATIVE EVIDENCE CONTRACT]\n"
"Use the frozen design summary and tagged evidence below as the only source of this study's empirical facts. "
"Do not change the setting, sample, interview count, data sources, actor/data-source relations, analytic design, temporal structure, cases, procedures, or quotations. "
"Do not infer an event sequence, actor comparison, trustworthiness/ethics/reflexivity procedure, negative case, or quotation that is absent from this packet.\n\n"
"### Table: Qualitative Evidence Structure\n"
"| Higher-order Theme | Empirical Concept | Process Stage | Case/Actor | Evidence Role | Representative Quote |\n"
"| :--- | :--- | :--- | :--- | :--- | :--- |\n"
)
if data_source_roles:
md += (
"### Frozen Actor / Data-Source Relations\n"
"| Actor role | Data source | Count |\n"
"| :--- | :--- | :--- |\n"
)
for row in data_source_roles:
md += f"| {row['actor_role']} | {row['data_source']} | {row['count'] if row['count'] is not None else 'Not specified'} |\n"
md += "\n"
clean_evidence = []
seen_quotes = set()
for item in evidence[:15]:
if not isinstance(item, dict):
continue
quote = self._clean_cell(item.get("quote"), "").replace('"', "'")
concept = self._clean_cell(item.get("concept"), "Empirical observation")
theme = self._clean_cell(item.get("theme"), "Emergent pattern")
stage = self._clean_cell(item.get("stage"))
case_label = self._clean_cell(item.get("case_label"))
evidence_role = str(item.get("evidence_role") or "typical").strip().lower()
if evidence_role not in {"typical", "deviant"}:
evidence_role = "typical"
normalized_quote = re.sub(r"\s+", " ", quote).strip().lower()
if not self._quote_has_substantive_content(quote) or normalized_quote in seen_quotes:
continue
seen_quotes.add(normalized_quote)
tag = f"QUOTE-{len(clean_evidence) + 1:02d}"
md += (
f'| {theme} | {concept} | {stage} | {case_label} | {evidence_role} | '
f'"{quote}" [{tag}] |\n'
)
clean_evidence.append(
{
"theme": theme,
"concept": concept,
"quote": quote,
"quote_hash": hashlib.sha256(re.sub(r"\s+", " ", quote).strip().encode("utf-8")).hexdigest(),
"tag": tag,
"stage": stage,
"case_label": case_label,
"evidence_role": evidence_role,
}
)
theme_map = {}
theme_ids = {}
for item in clean_evidence:
theme_key = re.sub(r"\s+", " ", item["theme"]).strip().casefold()
theme_id = theme_ids.setdefault(theme_key, f"THEME-{len(theme_ids) + 1:02d}")
item["theme_id"] = theme_id
theme_entry = theme_map.setdefault(
item["theme"],
{
"theme_id": theme_id,
"title": item["theme"],
"concepts": [],
"evidence": [],
"stages": [],
"has_deviant_evidence": False,
},
)
if item["concept"] not in theme_entry["concepts"]:
theme_entry["concepts"].append(item["concept"])
if item["stage"] != "Not specified" and item["stage"] not in theme_entry["stages"]:
theme_entry["stages"].append(item["stage"])
if item["evidence_role"] == "deviant":
theme_entry["has_deviant_evidence"] = True
theme_entry["evidence"].append(
{
"theme_id": item["theme_id"],
"tag": item["tag"],
"quote": item["quote"],
"quote_hash": item["quote_hash"],
"stage": item["stage"],
"case_label": item["case_label"],
"evidence_role": item["evidence_role"],
}
)
themes = list(theme_map.values())
validation = APP_CONFIG.get("manuscript_validation", {})
minimum_themes = int(validation.get("minimum_qualitative_themes", 3))
minimum_quotes = int(validation.get("minimum_qualitative_quotes", 8))
if len(themes) < minimum_themes or len(clean_evidence) < minimum_quotes:
raise RuntimeError(
"Qualitative evidence packet is incomplete: "
f"themes={len(themes)}, quotes={len(clean_evidence)}; "
f"required themes>={minimum_themes}, quotes>={minimum_quotes}."
)
if not has_user_data and not any(item["evidence_role"] == "deviant" for item in clean_evidence):
raise RuntimeError("Generated qualitative evidence packet lacks a deviant/boundary observation.")
metadata = {
"mode": "qualitative",
"source_type": (
"user_provided" if has_user_data else "system_generated"
),
"disclosure_required": not has_user_data,
"study_facts": {
"research_setting": setting,
"data_sources": clean_sources,
"data_source_roles": data_source_roles,
"participant_count": participants,
"interview_count": interviews,
"trustworthiness_procedures": trustworthiness,
"ethics_procedures": ethics,
"reflexivity_procedures": reflexivity,
"analytic_design": analytic_design,
"analysis_method": analytic_design,
"analytic_logic": analytic_logic,
"temporal_structure": temporal_structure,
"case_structure": case_structure,
},
"evidence_count": len(clean_evidence),
"generation_parameters": {
"analytic_design": analytic_design,
"requested_quote_count_when_generated": 12,
"maximum_retained_quotes": 15,
"minimum_themes": minimum_themes,
"minimum_quotes": minimum_quotes,
"requires_generated_deviant_evidence": not has_user_data,
"consistency_constraints": [
"one frozen research setting",
"one frozen participant count",
"one frozen interview count",
"one frozen data-source set",
"one frozen analytic design",
"no unsupported temporal or case structure",
"quotation tags must remain unique",
],
},
"evidence": clean_evidence,
"themes": themes,
"user_research_intent": build_user_research_intent(topic, keywords, mode="qualitative"),
}
metadata["study_design_contract"] = build_qualitative_contract(topic, keywords, metadata)
metadata["semantic_claim_frame"] = self.build_qualitative_semantic_claim_frame(metadata)
status_message = "✅ 定性研究事实、分析设计与证据结构已冻结。"
return md, status_message, metadata
def get_qualitative_engine(agent):
return QualitativeEngine(agent)
|