Files
C.E.L_Slide_test2/tests/matching/INTENT_DIAGNOSTIC_REPORT.md
T
KyeongminandClaude Opus 4.8 b836e79ee1 wip: phase_z2 evidence 파이프라인 + matching 실험(phase2~26) + 프론트 trace 패널 진행분 스냅샷
- src: phase_z2 composition/mapper/pipeline/placement_planner/retry, ai_fallback(prompts/schema/validate), mdx_text_atoms 신규
- Front: PipelineTracePanel 신규, FramePanel/SlideCanvas/Home/designAgentApi 등 갱신 + 테스트 4종 추가
- templates/phase_z2: catalog(component_expansion_registry, node_slot_mapping 신규), frames, families, slide_base 갱신
- tests/matching: phase2~26 매칭 실험 스크립트·리포트·온톨로지 전체 (미커밋 진행분)
- tests: b4_v4 evidence, task5~28.5 시리즈, regression(imp95 baseline) 등 신규 테스트 대량 추가
- docs/reference: MDX 구조 인벤토리, MDX→Frame 구조 계약 문서
- scripts: mdx 계약/parity/coverage/viewport 체크, gitea comment, run sync 유틸
- .gitignore: tmp*.json, chromedriver, .orchestrator, *.pkl, Front_test* 등 임시/스냅샷 제외

미완성 작업의 보존용 스냅샷 커밋 (2026-07-02)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-02 17:03:42 +09:00

8.9 KiB
Raw Blame History

Structure Intent Diagnostic — 점수 통합 전 검수

목적: structure_intent 추론이 사람 직관과 맞는지 확인. 맞으면 Phase 24/25 점수에 통합 진입.

범위: 11 문제 frame 에만 intent 태깅. 나머지 21 frames 은 neutral 0.5 처리.

대상 11 frames: 13, 14, 17, 18, 20, 23, 24, 28, 29, 30, 31

정의된 intent 종류 (10): category_comparison, concept_comparison, industry_comparison, multi_attribute_comparison, persona_benefit, problem_diagnosis, process_product_split, requirement_list, requirement_or_pillar, transformation_story

MDX01-2 — BIM vs DX 다면 비교

  • 정답 Frame: 18
  • MDX detected_intents: ['concept_comparison', 'multi_attribute_comparison']
  • MDX item_count: 2
  • MDX relation_type: compare
순위 sid template_id frame_intents compat 비고
1 18 bim_dx_comparison_table concept_comparison, multi_attribute_comparison 1.00 🎯 정답
2 23 app_sw_package_vs_solution concept_comparison, multi_attribute_comparison 1.00
3 24 engn_sw_three_types category_comparison, multi_attribute_comparison 1.00
4 30 industry_current_status_three_ industry_comparison, multi_attribute_comparison 1.00
5 31 industry_characteristics_three industry_comparison, multi_attribute_comparison 1.00
6 21 solution_engn_split_diagram (미태깅, neutral) 0.50
7 25 commercial_sw_four_categories (미태깅, neutral) 0.50
8 26 sw_dependency_four_problems (미태깅, neutral) 0.50
9 19 design_method_distortion_three (미태깅, neutral) 0.50
10 27 bim_adoption_central_split (미태깅, neutral) 0.50
11 12 construction_goals_three_circl (미태깅, neutral) 0.50
12 15 policy_goals_plus_execution_re (미태깅, neutral) 0.50

하위 5개 (intent 불일치 확인용):

순위 sid template_id frame_intents compat
28 13 three_parallel_requirements requirement_list 0.30
29 20 dx_sw_necessity_three_perspect requirement_or_pillar 0.30
30 28 sw_reality_three_emphasis problem_diagnosis 0.25
31 17 bim_current_problems_paired problem_diagnosis 0.25
32 29 process_product_two_way process_product_split, transformation_story 0.20

🔎 문제 후보 진단:

  • Frame 29 process_product_two_way: intent=[process_product_split, transformation_story], compat=0.20 ✓ 낮음 (정상)

MDX02 — 주체별 기대효과

  • 정답 Frame: 14
  • MDX detected_intents: ['persona_benefit']
  • MDX item_count: 3
  • MDX relation_type: parallel
순위 sid template_id frame_intents compat 비고
1 14 three_persona_benefits persona_benefit 1.00 🎯 정답
2 21 solution_engn_split_diagram (미태깅, neutral) 0.50
3 25 commercial_sw_four_categories (미태깅, neutral) 0.50
4 26 sw_dependency_four_problems (미태깅, neutral) 0.50
5 19 design_method_distortion_three (미태깅, neutral) 0.50
6 27 bim_adoption_central_split (미태깅, neutral) 0.50
7 12 construction_goals_three_circl (미태깅, neutral) 0.50
8 15 policy_goals_plus_execution_re (미태깅, neutral) 0.50
9 32 policy_achievement_five_goals (미태깅, neutral) 0.50
10 01 sw_development_cycle_six_nodes (미태깅, neutral) 0.50
11 04 domestic_bim_actor_relations (미태깅, neutral) 0.50
12 06 compensation_complaint_map (미태깅, neutral) 0.50

하위 5개 (intent 불일치 확인용):

순위 sid template_id frame_intents compat
28 31 industry_characteristics_three industry_comparison, multi_attribute_comparison 0.40
29 13 three_parallel_requirements requirement_list 0.30
30 29 process_product_two_way process_product_split, transformation_story 0.30
31 28 sw_reality_three_emphasis problem_diagnosis 0.20
32 17 bim_current_problems_paired problem_diagnosis 0.20

🔎 문제 후보 진단:

  • Frame 21 solution_engn_split_diagram: intent=[(미태깅)], compat=0.50 ⚠️ 높음 (문제)

MDX03-1 — DX 필수요건 3요소

  • 정답 Frame: 13
  • MDX detected_intents: ['requirement_list']
  • MDX item_count: 3
  • MDX relation_type: parallel
순위 sid template_id frame_intents compat 비고
1 13 three_parallel_requirements requirement_list 1.00 🎯 정답
2 20 dx_sw_necessity_three_perspect requirement_or_pillar 0.70
3 21 solution_engn_split_diagram (미태깅, neutral) 0.50
4 25 commercial_sw_four_categories (미태깅, neutral) 0.50
5 26 sw_dependency_four_problems (미태깅, neutral) 0.50
6 19 design_method_distortion_three (미태깅, neutral) 0.50
7 27 bim_adoption_central_split (미태깅, neutral) 0.50
8 12 construction_goals_three_circl (미태깅, neutral) 0.50
9 15 policy_goals_plus_execution_re (미태깅, neutral) 0.50
10 32 policy_achievement_five_goals (미태깅, neutral) 0.50
11 01 sw_development_cycle_six_nodes (미태깅, neutral) 0.50
12 04 domestic_bim_actor_relations (미태깅, neutral) 0.50

하위 5개 (intent 불일치 확인용):

순위 sid template_id frame_intents compat
28 30 industry_current_status_three_ industry_comparison, multi_attribute_comparison 0.30
29 31 industry_characteristics_three industry_comparison, multi_attribute_comparison 0.30
30 29 process_product_two_way process_product_split, transformation_story 0.20
31 28 sw_reality_three_emphasis problem_diagnosis 0.15
32 17 bim_current_problems_paired problem_diagnosis 0.15

🔎 문제 후보 진단:

  • Frame 28 sw_reality_three_emphasis: intent=[problem_diagnosis], compat=0.15 ✓ 낮음 (정상)

MDX03-2 — 과정/결과 혁신

  • 정답 Frame: 29
  • MDX detected_intents: ['process_product_split', 'transformation_story']
  • MDX item_count: 2
  • MDX relation_type: compare
순위 sid template_id frame_intents compat 비고
1 29 process_product_two_way process_product_split, transformation_story 1.00 🎯 정답
2 21 solution_engn_split_diagram (미태깅, neutral) 0.50
3 25 commercial_sw_four_categories (미태깅, neutral) 0.50
4 26 sw_dependency_four_problems (미태깅, neutral) 0.50
5 19 design_method_distortion_three (미태깅, neutral) 0.50
6 27 bim_adoption_central_split (미태깅, neutral) 0.50
7 12 construction_goals_three_circl (미태깅, neutral) 0.50
8 15 policy_goals_plus_execution_re (미태깅, neutral) 0.50
9 32 policy_achievement_five_goals (미태깅, neutral) 0.50
10 01 sw_development_cycle_six_nodes (미태깅, neutral) 0.50
11 04 domestic_bim_actor_relations (미태깅, neutral) 0.50
12 06 compensation_complaint_map (미태깅, neutral) 0.50

하위 5개 (intent 불일치 확인용):

순위 sid template_id frame_intents compat
28 13 three_parallel_requirements requirement_list 0.20
29 18 bim_dx_comparison_table concept_comparison, multi_attribute_comparison 0.20
30 23 app_sw_package_vs_solution concept_comparison, multi_attribute_comparison 0.20
31 28 sw_reality_three_emphasis problem_diagnosis 0.15
32 17 bim_current_problems_paired problem_diagnosis 0.15

🔎 문제 후보 진단:

  • Frame 18 bim_dx_comparison_table: intent=[concept_comparison, multi_attribute_comparison], compat=0.20 ✓ 낮음 (정상)

검수 가이드

  1. 정답 Frame 이 상위에 있는가 — compat ≥ 0.8 예상
  2. 문제 후보 (MDX01-2×29, MDX03-1×28, MDX03-2×18) 가 0.4 이하인가 — 낮으면 의도대로 intent 가 걸러냄
  3. MDX detected_intents 가 자연스러운가 — AI 추론 결과 검토
  4. 미태깅 frame (neutral 0.5) 이 정답보다 위에 오는가 — 있으면 부분 적용 한계 = 추가 태깅 필요

모두 통과하면 → Phase 24/25 점수 통합 진입.