Files
hyungi_document_server/reports/phase2_final_candidate.csv
Hyungi Ahn 120db86d74 docs(search): Phase 2 최종 측정 보고서 (phase2_final.md + csv A/B)
## 결과 요약

Phase 1.3 baseline vs Phase 2 final A/B (평가셋 v0.1, 23 쿼리):
 - Recall@10:  0.730 → 0.737 (+0.007)
 - NDCG@10:    0.663 → 0.668 (+0.005)
 - Top-3 hit:  0.900 → 0.900 (0)
 - p95 latency: 171ms → 256ms (+85)
 - news_crosslingual NDCG: 0.27 → 0.37 (+0.10 ✓)
 - exact_keyword / natural_language_ko: 완전 유지 (회귀 0)

## Phase 2 게이트: 2/6 통과
 ✓ news_crosslingual NDCG ≥ 0.30
 ✓ latency p95 < 400ms
  Recall@10 ≥ 0.78 (0.737)
  Top-3 hit ≥ 0.93 (0.900)
  crosslingual_ko_en NDCG ≥ 0.65 (0.53, bge-m3 한계)
  평가셋 v0.2 작성 (후속)

## 핵심 성과 (게이트 미달이지만 견고한 기반)
 1. QueryAnalyzer async-only 아키텍처 (retrieval 차단 0)
 2. semaphore concurrency=1 (MLX single-inference queue 폭발 방지)
 3. multilingual narrowing (news/global 한정 → 회귀 0 + news 개선)
 4. soft_filter boost 보수적 설정 (0.01, domain only)
 5. prewarm 15개 → cache hit rate 70%+

## infra_inventory.md soft lock 준수
 - config.yaml / Ollama / compose restart 변경 0

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-08 15:52:21 +09:00

4.4 KiB

1labelidcategoryintentdomain_hintqueryrelevant_idsreturned_ids_top10latency_msrecall_at_10mrr_at_10ndcg_at_10top3_hiterror
2singlekw_001exact_keywordfact_lookupdocument산업안전보건법 제6장3856;3868;38793856;3851;3862;3853;3861;3868;3879;3873;3876;387167.91.0001.0000.7931
3singlekw_002exact_keywordfact_lookupdocument중대재해 처벌 등에 관한 법률 제2장 중대산업재해3917;39213921;3917;3919;3923;3916;3874;3918;3854;3922;3920110.01.0001.0001.0001
4singlekw_003exact_keywordfact_lookupdocument화학물질관리법 유해화학물질 영업자39813981;3985;3980;3984;3993;3857;3978;3986;3983;3957119.31.0001.0001.0001
5singlekw_004exact_keywordfact_lookupdocument근로기준법 안전과 보건40414041;3852;3851;3877;3905;3903;3858;3881;3781;3912108.71.0001.0001.0001
6singlekw_005exact_keywordfact_lookupdocument산업안전보건기준에 관한 규칙 보호구38883888;3912;3911;3905;3909;3889;3910;3897;3890;3896125.51.0001.0001.0001
7singlenl_001natural_language_kosemantic_searchdocument기계로 인한 산업재해 관련 법령3856;3868;3879;38543878;3897;3863;3868;3856;3879;3895;3867;3851;385483.91.0000.2500.5710
8singlenl_002natural_language_kosemantic_searchdocument사업주가 도급을 줄 때 산업재해를 예방하기 위해 해야 할 일3855;3867;38783855;3917;3854;3867;3878;3863;3851;3908;3903;3895118.01.0001.0000.8531
9singlenl_003natural_language_kosemantic_searchdocument유해화학물질을 다루는 회사가 지켜야 할 안전 의무3980;3981;39823980;3903;3904;3905;3981;3985;3896;3917;3857;390982.80.6671.0000.6511
10singlenl_004natural_language_kosemantic_searchdocument중대재해가 발생했을 때 경영책임자가 처벌받는 기준3916;3917;3920;39213917;3918;3916;3923;3919;3921;3854;3872;3877;392272.10.7501.0000.7251
11singlenl_005natural_language_kosemantic_searchdocument안전보건교육은 누가 받아야 하고 어떤 내용을 다루는가3853;38653853;4025;3876;3879;3859;3865;3781;3815;3818;378780.21.0001.0000.8321
12singlecl_001crosslingual_ko_ensemantic_searchdocument기계 안전 가드 설계 원리3770;38563770;4540;3817;4541;3774;3816;3787;3758;3793;3773108.30.5001.0000.6131
13singlecl_002crosslingual_ko_ensemantic_searchdocument산업 안전 입문서3755;3775;3776;37773760;3755;3774;3764;3758;3775;3779;3802;3814;381779.50.5000.5000.3851
14singlecl_003crosslingual_ko_ensemantic_searchdocument전기 안전 위험3772;37903897;3772;3771;4018;3773;3790;3819;4020;3807;3755103.71.0000.5000.6051
15singlenews_001news_kosemantic_searchnews이란과 미국의 군사 충돌4303;4304;4307;4316;4322;4323;4327;43354317;4321;4771;4743;4307;4452;4761;4678;4418;43311445.80.1250.2000.0981
16singlenews_002news_kosemantic_searchnews호르무즈 해협 봉쇄4316;4320;4322;43274327;4346;4349;4762;4767;4759;4322;4320;4340;4304185.30.7501.0000.6440
17singlenews_003news_ensemantic_searchnewsTrump Iran ultimatum4258;4260;42624776;4515;4519;4658;4644;4763;4333;4762;4679;432176.30.0000.0000.0001
18singlenews_004news_frsemantic_searchnewsguerre en Iran4199;4202;4210;4361;4363;4507;4519;45214678;4507;4199;4688;4776;4363;4519;4668;4670;4672157.90.5000.5000.4601
19singlenews_005news_crosslingualsemantic_searchnews이란 미국 전쟁 글로벌 반응4202;4258;4262;4536;4303;4304;43164262;4457;4765;4324;4345;4329;4258;4452;4443;4761186.70.2861.0000.3671
20singlemisc_001other_domainfact_lookupdocument강체의 평면 운동학4063;40654063;4065;4064;4067;4071;4068;4069;4062;4060;4066171.61.0001.0001.0001
21singlemisc_002other_domainsemantic_searchdocument질점의 운동역학4060;4061;40624062;4060;4070;4064;4068;4067;4065;4058;4071;4066263.60.6671.0000.7651
22singlefail_001failure_expectedsemantic_searchdocumentRust async runtime tokio scheduler 내부 구조4815;4069;4546;4062;4547;3801;3787;3812;4542;3770121.90.0000.0000.0001
23singlefail_002failure_expectedsemantic_searchdocument양자컴퓨터 큐비트 디코히어런스4058;4057;4067;3800;4065;4068;3817;4063;4064;391575.20.0000.0000.0001
24singlefail_003failure_expectedsemantic_searchnews재즈 보컬리스트 빌리 홀리데이4634;4100;4815;4116;4281;4697;4205;4077;4235;428973.90.0000.0000.0001