hyungi/hyungi_document_server
1
1
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity
654 Commits 53 Branches 1 Tag
701113738f92d034c749f3dea3781a2ebed26aa8
Commit Graph

654 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
hyungi 701113738f merge: 편집형 /digest(57de6a1) + UI 세이지 셸 통합 2026-06-04 05:02:11 +00:00
hyungi cc8bdee6c1 feat(ui): 셸 재구성 — nav 4그룹·데스크탑 상시 사이드바·모바일 하단탭바 (F2) 
+layout.svelte: 상단 nav 11개 flat → 4그룹(홈·문서▾·뉴스▾·질문, 드롭다운) +
브랜드(DS)·받은편지함·⋮(설정/로그아웃). 데스크탑(lg+)=상시 좌측 사이드바,
모바일(<lg)=하단 탭바(문서·뉴스·질문·메모·더보기) + 사이드바 드로어.
세이지 토큰 Tailwind. /news=풀스크린(상시 사이드바 없음). frontend docker build PASS.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-04 05:02:11 +00:00
hyungi e968236796 feat(ui): app.css 테마 다크블루 → 세이지 그린 라이트 (F1) 
UI 전면 개선 파운데이션. @theme + :root 토큰 값을 세이지 라이트로 교체
(bg #e7ebe4·surface #f4f7f1·text #23291f·accent #4f8a6b·도메인색 세이지 조화).
토큰 규율(lint:tokens) 덕에 값 교체만으로 전 페이지 전환. markdown zebra
rgba(255,255,255,.02)→rgba(35,41,31,.03) 1곳 라이트 보정. frontend docker build PASS.
검토 대상 = text-white 14 + bg-white 2 (대부분 강조색 버튼 위, 시각확인 시 점검).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-04 04:53:39 +00:00
hyungi 57de6a1072 feat(digest): 편집형 1면 레이아웃 (안1 채택) 
/digest 단순 카드 → 신문 1면형 편집 뷰. 웜톤(크림+clay) self-contained — 앱 다크토큰 충돌 방지 위해 .digest-page 래퍼에 웜 팔레트 로컬 재정의.
- 슬롯 매핑: ALL=전국가 imp 내림차순 / country=rank 오름차순 → lead·featured 2·sidebar 3·심층 grid, graceful 생략
- 국가 nav(ALL+국가별 주제수)·edition line·중요도 막대. date picker URL sync·기사 /documents/{id} 라우팅·국가사전 재사용
- 검정·이모지·외부폰트 0. 구현+적대적 리뷰 2(ok). docker build PASS

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-04 02:55:19 +00:00
hyungi 696d8b71b0 Merge pull request 'Feat/digest ui followup' (#26) from feat/digest-ui-followup into main 
Reviewed-on: #26
 2026-06-04 08:44:16 +09:00
hyungi f269e0df27 ops(news): chunk_worker news_source 매핑 실패 가시성 가드 
_lookup_news_source prefix 미일치 시 silent (None) 반환 → warn 로그 추가.
loader 의 drop 로그와 대칭, 신규 source / RSS category 오염 재발 즉시 가시. 동작 변경 0.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 23:39:14 +00:00
hyungi aa2d7814e3 feat(digest): date picker URL sync + article→문서 라우팅 + country 국기·한국어 
- GET /api/digest/dates 신설 (브리핑 /briefing/dates 패턴 미러, read-only)
- topic article 제목 enrich (documents 배치 1쿼리 + dedupe(set) + map-miss=null → 프론트 '(제목 없음)')
- /digest 재작성: ?date=&country= URL sync(공유·뒤로가기), 국가 탭=인라인 SVG 국기+한국어, 기사=/documents/{id} 링크(상위5+펼치기)
- Phase 4.5(PR #22) 후속. 검증: py_compile·dates/enrich 쿼리(275 resolve·miss 0)·frontend docker build PASS. 시각 렌더 검증=preview 게이트 대기

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 23:39:07 +00:00
hyungi cd33ded7a8 docs(search): passage-RAG go/no-go = NO-GO (hier evidence 동등, diagnose c4+c5) 
PR-DocSrv-Hier-PassageRAG-Diagnose-1 c4+c5. 조건부 N=12(retrieval 통제) blind pairwise
(hypothesis-blind subagent, 익명 3-file split). 결과 4-way 수렴 = 동등:
pairwise prehier4/hier3/tie5(no edge) + axis ±0.08 + objective 동일(halluc36/36) +
variance~0(byte-identical 재생성). verbosity artifact 없음(prehier 더 길었으나 승+1).
=> NO-GO: hier-leaf evidence 무이득. hier leaf = section-outline UI 전용 완전 확정
(UI yes / doc-search NO-GO / passage-RAG NO-GO 3영역 종결). 2026-06-21 freeze input only.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 07:02:46 +00:00
hyungi 9c039139ef feat(search): passage-RAG capture runner + raw JSONL (diagnose c3) 
PR-DocSrv-Hier-PassageRAG-Diagnose-1 c3. 22Q x {prehier,hier_sim_clean} /ask?debug=true
exact_knn capture (44 rec). ai_answer/evidence/target_doc_present/target_span_used/
objective signals(hallucination/grounding/completeness/refused) 박제.
관찰: hier 일부 타깃 retrieval 실패(exam_005/006,cl_007=doc-search NO-GO 일관) + 일부 gain
(cl_001/002). empty-answer 케이스(cl_005/cl_007 prehier, cl_006/exam_004 skipped) 존재.
JWT 15min 만료로 1차 부분실패 → cache-warm 재실행 완주.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 06:53:11 +00:00
hyungi 698510bc0e feat(search): passage-RAG answer-seeking question subset (diagnose c2) 
PR-DocSrv-Hier-PassageRAG-Diagnose-1 c2. queries.yaml v0.2 의 answer-seeking 22문항
(exam 7 + korean_only 7 + mixed 8, decomposed-target 필터). targets_g2/g3 = 조건부 subset
산출용. broad seed (조건부 ~65-70% → N≥12 확보). 신규 authoring 0 (기존 graded 재사용).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 06:20:20 +00:00
hyungi 2f152911f7 feat(search): /ask corpus_variant + exact_knn (EVAL-ONLY) for passage-RAG diagnose 
PR-DocSrv-Hier-PassageRAG-Diagnose-1 c1. /ask evidence retrieval 의 chunk leg 를
측정 뷰(prehier/hier_sim_*)로 교체 + exact_knn — passage evidence 단위(hier 절 vs
legacy 윈도우) 비교용. /search 와 동일 패턴, run_search 전달. EVAL-ONLY 박제,
default(미지정) 시 기존 /ask byte/behavior 동일(회귀 0). pattern 검증 → 잘못된 값 422.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 06:14:59 +00:00
hyungi 6e9d73278f docs(search): pin hier measurement views as EVAL-ONLY (replace-diagnose) 
COMMENT ON VIEW + header — corpus_chunks_{prehier,hier_sim_raw,hier_sim_clean} 은
?corpus_variant= eval dispatch 전용. production retrieval default-path 는 corpus_chunks
(partial ivfflat) 만. 재측정/passage-RAG 재평가 자산으로 보존, 오용 방지 박제.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 05:53:04 +00:00
hyungi 6a9142a2e5 docs(search): hier vs legacy go/no-go = NO-GO (replace-diagnose c6) 
PR-DocSrv-Hier-Replace-Diagnose-1 c6 측정+결정. prehier exact vs hier_sim exact, dedup 0/51.
결정權(분해-subset n=41): prehier 0.748 -> hier_sim_clean 0.675 (-0.074 회귀). raw 0.673 (robust).
카테고리: standards(법령, hier 최적가설) flat -0.002 / exam -0.183 / korean -0.109 / english -0.088.
법령 제N조조차 개선 없음 + 대체로 회귀 → 짧은 절 leaf 가 맥락 손실. dedup clean = 실제값.
=> NO-GO: 검색 코퍼스 hier 교체 안 함. Apply PR 미진입. hier leaf 는 in_corpus=false 잔존
(section-outline UI 재료, doc-level 검색 무관). 측정은 doc-level NDCG 한정.

산출물: decision md + 4 eval csv(sanity/prehier/clean/raw exact) + subset analysis script.
in_corpus 634 전 구간 불변. default 검색 path 회귀 0.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 05:46:14 +00:00
hyungi 100aaa3b0c feat(search): corpus_variant + exact_knn measurement dispatch (replace-diagnose c4+c5) 
PR-DocSrv-Hier-Replace-Diagnose-1 c4+c5. hier vs prehier(legacy) go/no-go 비파괴 측정 hook.
- 측정 뷰 3종 (hier_measure_views.sql, additive/droppable): corpus_chunks_prehier
  (legacy+null-source 375 포함) / hier_sim_raw / hier_sim_clean (childless-tiny<30 제외,
  all-tiny doc 은 legacy fallback 정합).
- retrieval_service: _resolve_corpus_variant + CORPUS_VARIANT_MAP + _VALID_CHUNKS_TABLE
  3 뷰 추가 + exact_knn(SET LOCAL enable_indexscan/bitmapscan=off, eval 전용).
  chunk leg 만 영향 (doc-level + fts/trgm = documents 무관). baseline/None path 회귀 0.
- search_pipeline.run_search + search.py: corpus_variant/exact_knn 전달, unknown→400,
  embedding_backend cand 와 동시 사용 금지(400).
- run_eval: --corpus-variant + --exact-knn flag.
- tests/test_corpus_variant.py 22 PASS (resolver/map/allowlist + SQL injection 거부).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 05:37:15 +00:00
hyungi e860baa179 ops(hier): Phase A law/library decompose + snapshot freeze (replace-diagnose c3) 
47 eval-target undecomposed non-news docs (law21+library24+document2) 분해+임베딩
(--skip-analysis, additive). 1005 leaf 생성 fail0, in_corpus 634 무손상 검증.
snapshot doc_id_max=25912 chunk_id_max=71164 docs_decomposed 301->348. 측정 drift 0.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 05:23:38 +00:00
hyungi fc9e0f1d8f feat(search): hier backfill --skip-analysis + --doc gate-bypass flags 
PR-DocSrv-Hier-Replace-Diagnose-1 c2. 구조화 소형 문서(법령 등) eval coverage
보정용 — --doc 명시 리스트로 DOC_MIN_CHARS=4000 게이트 우회, --skip-analysis 로
절분석(Mac mini) 생략하고 분해+임베딩만. retrieval go/no-go 측정 준비. additive,
in_corpus 무영향. NOT EXISTS hier 멱등 가드 유지.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 05:21:00 +00:00
hyungi f7198d9d68 feat(search): expose hier section outline & summaries in document detail 
PR-DocSrv-Hier-Section-UI-1 Phase 1 (코드+커밋만, 배포는 Phase 2 backfill 완주 후).

- backend: GET /documents/{id}/sections — hier leaf 목차 + chunk_section_analysis
  요약. document_chunks 직접 조회(retrieval 아닌 목차 표시라 corpus_chunks 뷰
  의도적 우회 — docstring 명시). DISTINCT ON 으로 최신 분석 1행.
- frontend: SectionOutline.svelte(좌측 목차, per-doc 동적 그룹/flat, window
  dedupe, 클릭 시 요약/breadcrumb 인라인), headingPath.ts 순수 유틸(+node:test
  단위테스트 8케이스). [id]/+page.svelte 3-zone 레이아웃 + 우측 메타 Tabs
  [정보|AI|관리] 로 카드 스프롤 해소.
- 절 없는 문서/404 는 목차 숨김(graceful). 본문 점프는 follow-up.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-25 00:22:34 +00:00
hyungi ec174fc1e7 ops(hier): default backfill scope to all-except-news 
기본 범위 = 뉴스 도메인만 제외, 나머지 전부(>4000자 미분해). --domains 로 allowlist override.
신규 후보 50건(general 29 + programming 13 + engineering 8). additive(in_corpus=false) 유지.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 22:51:13 +00:00
hyungi c2f9dca62d ops(hier): add section analysis backfill runner 
hier 분해(additive, in_corpus=false) + 절 분석(Mac mini gemma-26B BACKGROUND gate)
오버나이트 backfill 러너. time-box deadline + per-doc commit + 멱등 선별(NOT EXISTS).
section_summary_pilot 상수 재사용(PROMPT_VERSION 단일화). no silent fallback.
검증: Engineering+Industrial_Safety 245 doc / 6066 절 요약 / fail 0 (2026-05-24~25).
컨테이너 TZ=UTC → deadline KST 환산 주의. 종료는 컨테이너 내부 PID kill 필수.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 22:47:06 +00:00
hyungi cfadaaffd9 feat(search): hier section per-leaf analysis scaffold (Section-Summary-1 c1) 
chunk_section_analysis 테이블(migration 286) + ORM model + pilot script.
document_chunks(retrieval-hot)와 분리된 절-레벨 분석 축. domain 상속,
section_type 절-전용 역할 enum, status로 skip 박제, source_content_hash로 stale 탐지.
script-only(scripts mount, rebuild 불필요). LLM 0 dry-run 검증 = 5225 147 analyze + 17 skip.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 13:45:30 +00:00
hyungi a7b16b63db feat(search): doc-level atomic corpus replace + isolation test (Hier-Decomp-1 c5) 
replace_doc_corpus(dry_run): G5 precond(doc-local embed 100% + parent 무결성 + leaf>0) 검증 후
단일 트랜잭션 atomic 교체(legacy in_corpus=false / hier leaf in_corpus=true,
predicate=is_leaf AND embedding NOT NULL, node_type 미사용). 물리삭제 없음. rollback_doc_corpus 역토글.
precond 미충족 시 변경 0(legacy 유지).

tests/hier_decomp/test_corpus_isolation.py: in_corpus=false leaf 가 corpus_chunks 누출 0 단언
(부분 ivfflat + 뷰 이중 choke point 회귀 가드).

c5: dry-run 3 pilot precond_ok(5140 158L→271leaf / 5186 381→199 / 5225 18→164), 격리 테스트 PASS.
실제 replace 는 c6(1-doc-first).

plan: hierarchical-decomposition-tiered-nesting-marmot.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 13:14:36 +00:00
hyungi fa82bd495b feat(search): hier persist + partial ivfflat index on in_corpus (Hier-Decomp-1 c4) 
persist_hier_tree(): build_hier_tree → document_chunks insert. source_type=hier_section,
in_corpus=false, is_leaf 노드만 bge-m3 embedding. idempotent(기존 hier 행 삭제 후 재삽입).
chunk_index = doc 별 (max+1) offset → 기존 (doc_id,chunk_index) unique 충돌 회피.
embedding NULL 파라미터 asyncpg 타입추론 → cast(cast(:emb AS text) AS vector) 이중캐스트.

migration 284/285: ivfflat 오염 fix. full 인덱스는 in_corpus=false hier 벡터까지 색인 →
근사 검색이 비활성 벡터에 오염(corpus_chunks 필터해도 근사 이웃 셋 흔들림). partial index
(WHERE in_corpus=true)로 교체 → in_corpus=false 는 검색 인덱스에 부재 = 무영향 인덱스 레벨 보장.

c4 pilot(5140/5186/5225) G3: 트리 insert, embed_coverage 1.0(doc-local 100%), in_corpus_true=0,
dangling_parent=0, dup 0. **부분인덱스 후 검색 baseline IDENTICAL to 원래(pre-hier)** = 691 hier
행 영향 0 검증(오염 fix 효과). replace 는 c5/c6.

plan: hierarchical-decomposition-tiered-nesting-marmot.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 13:12:42 +00:00
hyungi d982dce7d1 feat(search): rule hierarchy builder (Hier-Decomp-1 c3) 
순수 함수 build_hier_tree(text) → heading 경계 segment 트리 (DB 미접근, c4 에서 insert).
- 경계 규칙: ATX 마크다운(#{1,6}) > 한국 제N장/절/조 > 영문 Chapter/Section/Article.
- segment = heading + 다음 heading 전까지 본문 (disjoint, 100% 커버). parent/level = heading 깊이 정규화 트리.
- 과대 own-text(>HARD_MAX 5000) = 무overlap window 분해(자식 유무 무관), 부모 is_leaf=false(heading 마커, 코퍼스 제외).
- 구조 전용 heading(자식 보유 + own body<30자) = is_leaf=false. is_leaf = replace 코퍼스 편입 대상.

dry-run G2 (insert 없음, 5 pilot + headingless):
- 5140/5186/5225/5151/5124 md_content: coverage 0.9993~1.0, dup_hash 0, empty 0, dangling 0, bad_level 0, leaf_max<=4973(<5000).
- 5152 headingless extracted_text(238k): window 89 leaf, coverage 1.0, dup 0, leaf_max 3000.
관찰: tiny heading-only leaf(7~19자) 잔존(무해, tuning 후보).

plan: hierarchical-decomposition-tiered-nesting-marmot.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 13:05:06 +00:00
hyungi f940f50c60 feat(search): route retrieval through corpus_chunks view (Hier-Decomp-1 c2) 
baseline chunk 벡터검색을 document_chunks → corpus_chunks 뷰(in_corpus=true)로 rewire.
in_corpus=false(비활성 hier leaf 등) 자동 제외 = 검색 오염 구조적 차단(B choke point).

- retrieval_service: baseline chunks_table=corpus_chunks, _VALID_CHUNKS_TABLE 에 corpus_chunks 허용,
  snapshot_clause 조건 corpus_chunks 포함(eval snapshot 보존). candidate(cand_*) 경로 불변.
  documents 측(FTS+doc embedding) 무변경 — doc row 는 교체 무관.
- models/chunk: 5 신규 컬럼 매핑(parent_id/level/node_type/is_leaf/in_corpus). server_default 로
  기존 chunk_worker INSERT 무영향(legacy=in_corpus true/is_leaf false).
- subject_note_rag/explanation_rag: RAG chunk 로드에 in_corpus=true 필터(교체 doc legacy 중복 방지).

게이트: G4b(rewire 불변) before/after IDENTICAL(현재 view==table no-op) / G4a(누출) synthetic
in_corpus=false leaf 가 corpus_chunks 0건·document_chunks raw top(dist 0.0) 양방향 증명. /health 200.

plan: hierarchical-decomposition-tiered-nesting-marmot.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 12:58:28 +00:00
hyungi 7971e69e3e feat(search): hier decomposition schema + corpus_chunks view (Hier-Decomp-1 c1) 
PR-DocSrv-Hierarchical-Decomposition-1 c1 (G1).
- migration 282: document_chunks ADD parent_id/level/node_type/is_leaf/in_corpus
  (단일 statement ALTER, additive, IF NOT EXISTS). legacy 행 = in_corpus=true/is_leaf=false 기본값.
- migration 283: corpus_chunks 뷰 (WHERE in_corpus=true) = 검색 코퍼스 단일 choke point.
  c2 에서 retrieval 을 이 뷰로 rewire. node_type 은 hint, replace 는 is_leaf 사용.

검증: schema_migrations 282/283, 30952 행 in_corpus=true 보존, corpus_chunks 30952,
/health 200, restarts=0. dry-run(BEGIN/ROLLBACK) 선검증 후 적용.

plan: hierarchical-decomposition-tiered-nesting-marmot.md

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 12:47:41 +00:00
hyungi 0854c72c70 fix(search): sync doc md_status to failed on permanent markdown queue failure 
marker_worker 는 변환 시작 시 doc.md_status=processing 으로 표시하는데, 변환이
_fail()/_set_skipped() 를 거치지 않고 예외(예: 대형 batch ReadTimeout)로 죽으면
queue_consumer 가 큐 행만 failed 처리하고 doc.md_status 는 processing 에 영구 고착
= orphan (큐 failed, 문서 processing). markdown consumer 분리 후 이 orphan 이
tail 재처리에서 재발(5149/5201)하여 근본 원인 차단.

_process_stage except 블록에서 큐 항목이 영구 실패(attempts>=max)할 때 stage가
markdown 이고 doc.md_status=processing 이면 failed 로 동기화. 재시도 중
(attempts<max)엔 pending 큐 행이 남아 orphan 아니므로 미터치.

검증: synthetic 영구 실패 경로 → md_status processing→failed 동기화 PASS.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 12:06:32 +00:00
hyungi 2edc80d4bb fix(search): split markdown into dedicated queue consumer to prevent pipeline stall 
대형 PDF split 변환(5210 ≈ 40분 실측)이 단일 consume_queue 코루틴을 점유해
extract/classify/embed/chunk 등 전 파이프라인을 stall 시키던 문제 제거.

- consume_markdown_queue 신규 — markdown 전용 scheduler job (id=markdown_consumer)
- consume_queue 는 MAIN_QUEUE_STAGES (markdown 제외) 만 처리
- _process_stage / _load_workers 헬퍼로 per-stage 로직 공유
- reset_stale_items(stages, threshold_minutes) 파라미터화: main=10min(markdown 제외),
  markdown=MARKDOWN_STALE_MINUTES(기본 120). marker_worker 는 heartbeat 미기록이라
  40분 변환을 10분 stale 로 오인하던 함정 차단
- enqueue flow (classify -> embed,chunk,markdown) 불변

STT/deep_summary 분리 + GPU 동시성 튜닝은 out of scope (follow-up).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 10:33:45 +00:00
hyungi 826f66f8f5 fix(search): correct large-doc manifest wording after commit 4 drop 
PR-DocSrv-LargeDoc-Split-Markdown-1 follow-up (plan brisk-paging-quokka.md).

commit 4(marker_section→document_chunks) 드롭으로, split md_content/manifest 의
「권위 검색본 = document_chunks (source_type=marker_section)」 문구가 실제와 불일치.
실제 = 검색 인덱스는 기존 document_chunks(extracted_text long_pdf window chunks),
marker_section chunk 부재, md_content 는 Markdown 렌더링 preview.

- _build_large_md_content 헤더: 「검색 인덱스 = 기존 document_chunks long_pdf/
  extracted_text window chunks. 아래는 Markdown 렌더링 preview.」
- _split_manifest: canonical_storage(marker_section) → search_index(legacy/extracted_text)
- 상수 주석 + _process_split docstring: commit 4 드롭/이중적재 회피 반영

뷰어에 없는 source_type 으로 디버깅 오도 방지. 이미 처리된 5 docs 의 md_content 는
즉시 재처리 X — 자연 reprocess 시 갱신(사용자 결정).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 09:48:03 +00:00
hyungi cf0d75fe84 fix(search): handle markdown/fileless docs without marker conversion 
PR-DocSrv-LargeDoc-Split-Markdown-1 commit 5 (plan brisk-paging-quokka.md).

이미 마크다운인 문서는 marker 변환 불필요 → _process_markdown_passthrough 로
파일 내용(없으면 extracted_text)을 md_content 에 직접 적재(success), 비면 skipped.
- _is_markdown_doc: file_format=md/markdown 또는 .md/.markdown 확장자
- 분기 위치 = file_path validation 이전 (fileless md = file_path NULL 처리 위함)
- engine=passthrough 로 marker 변환본과 구분

기존 버그 해소: fileless md 43건=「no file_path」 fail / .md 파일=unsupported extension
skip → 둘 다 md_content 미생성이었음.

검증(docker cp 격리): 13948(.md+file_path)→success md_len=1805(파일) /
23409(fileless 931자)→success(extracted_text) / 20237(fileless 6자)→success.
PDF 경로 무영향(_is_markdown_doc=False).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 08:02:30 +00:00
hyungi 7aaabe2c75 feat(search): split markdown processing for large PDFs (>threshold) 
PR-DocSrv-LargeDoc-Split-Markdown-1 commit 3 (plan brisk-paging-quokka.md).

- page_count gauge 분기: 소형(<=120p)=_process_single 통째 1-shot / 대형(>120p)=_process_split
- MAX_PAGES=200 hard skip 제거 → 대형은 BATCH_PAGES=40 page-range 윈도우 순차 변환
- 각 batch /convert start_page/end_page(1-based) 호출 + slug 충돌 회피 batch별 ref rewrite + stitch
- _persist_images_to_nas seq_offset → batch 간 image_key(img_NNN) 연속
- md_status success/partial/failed (전부/일부/전무) + failed batch manifest JSON
- 대형 md_content = head+manifest (LARGE_DOC_MD_CONTENT_HEAD_CHARS=50000), canonical=document_chunks(commit 4)
- MARKER_MAX_SPLIT_PAGES=5000 초과 = skipped_too_large 안전상태

검증: G1 소형회귀 doc6675 동일(success,6292,14)/single경로 / G2 doc5180 453p→12batch success
manifest+207img(img_001~207 연속) / G4 stuck0 restart0 각batch<300s. 섹션 chunk적재(G3)=commit 4.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 07:39:49 +00:00
hyungi 2528996dee feat(marker): support page-range conversion in /convert 
ConvertRequest.start_page/end_page (1-based inclusive); per-request PdfConverter with config page_range, reuses loaded models. 1-based->0-based contained in marker adapter. PR-DocSrv-LargeDoc-Split-Markdown-1 commit 2.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 07:01:34 +00:00
hyungi 72190cf90a feat(search): add document_chunks page/source columns + unique idx 
migrations 279-281: page_start/end + source_type/chunker_version/source_hash/chunk_content_hash, legacy backfill (30,952 rows), unique (doc_id,source_type,chunker_version,chunk_index). PR-DocSrv-LargeDoc-Split-Markdown-1 commit 1.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 07:01:34 +00:00
hyungi 329c9eac76 feat(documents): PR-Chore-OCR-Column-1 add ocr_derived column 
RAG-independent data hygiene. ocr_derived 식별 컬럼 부재 = PR-Eval-V0_2
TBD-O FAILED 원인. 향후 OCR/Marker Diagnose, markdown 품질 분류,
ingest 품질 통계 어디에서나 재사용 가능.

Schema: documents.ocr_derived BOOLEAN NOT NULL DEFAULT false.
Backfill rule R1 단독 (실측 audit 후): extract_meta ? ocr_attempted
AND ocr_attempted = true. 8 rows true / 21727 false.

R2 (file_format IN png/jpg) 폐기 — 1건 R1 흡수 + 1건 marker 미처리.
R3 (marker PDF extract_meta 부재 283 rows) 폐기 — born-digital
false positive 위험. UPDATE 전 candidate preview + source rule별
count + 표본 audit gate 통과 후 적용.

asyncpg single-statement 제약으로 ALTER (277) + UPDATE (278) 분리.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 06:11:29 +00:00
hyungi c4a40ab18a docs(search): Phase 2Q closed as evaluated experiment (deprecated, not recommended for production) 
사용자 결정 (2026-05-24, measurement chain 4-layer 정정 완료 후):

> Phase 2Q Query Rewrite is closed as an evaluated experiment.
> After result-level dedup correction, true net gain was marginal
> (NDCG +0.019, Recall t≥2 +0.030) while latency cost was high
> (cold +876%, warm +320%). Therefore, multi-query rewrite is not
> recommended for default production rollout. Keep opt-in path as
> experimental/deprecated reference only; do not proceed to
> Cache-Prewarm unless future real-query evidence shows a stronger gain.

변경:
- docs/phase_2q_apply_opt_in.md: 🛑 DEPRECATED / EXPERIMENTAL status 박제. measurement chain
  정정 history (4-layer) + 진짜 효과 + Phase 2Q 성과 보존.
- app/api/search.py: rewrite_backend query param description 갱신 (⚠️ EXPERIMENTAL/DEPRECATED,
  production 추천 문구 제거, opt-in 실험 reference 만 유지 명시).

5 액션 박제 (사용자 결정):
  1. opt-in 코드 유지 (recommended=false / experimental)
  2. docs/ deprecated 박제
  3. search.py description production 추천 제거
  4. PR-2Q-Cache-Prewarm + PR-2Q-Apply-Default-ON-1 폐기
  5. Extended 4건 중 SynonymDict (deterministic, LLM 우회) 만 별도 후보 보존

신규 feedback memory: [[feedback_measurement_chain_audit]] — Diagnose 측정이 Apply/rollout
결정 기준일 때 retrieval/fusion/rerank/eval 모든 layer audit 필수. Phase 2Q 4-iteration
정정 chain (0.927→0.876→0.641→0.663) origin.

Phase 2Q 성과 (실패가 아닌 좋은 실험):
- chunk_id/doc_id 중복 inflation 발견 + measurement chain audit pattern 확립
- LLM rewrite 는 현재 DS 검색 기본값으로는 ROI 낮음 결론 확보
- search_pipeline 의 multi-query 합성 + 3-layer dedup 인프라 보존 (Extended SynonymDict
  또는 미래 cloud LLM scaffold 재사용 가능)
- 신규 feedback memory 4건: fixture-first-call-shape / apply-prereq-structural-fix /
  graded-ndcg-dedup-invariant / measurement-chain-audit

main 위 직접 commit (read-only docs / API description, retrieval path 영향 0).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 04:57:11 +00:00
hyungi 5e480d6d6e merge(search): PR-2Q-Search-Result-Dedup closed — 진짜 multi-query 효과 측정 (NDCG +0.019 / latency +876% cold) 2026-05-24 04:48:50 +00:00
hyungi 3b753f18d6 fix(search): Phase 2Q result dedup — apply_diversity unlimited path doc_id inflation 차단 
PR-2Q-Search-Result-Dedup. measurement chain 의 마지막 cleanup. plan inline.

root cause: apply_diversity 의 top_score ≥ 0.90 → unlimited path (diversity 제약 해제)
→ 같은 doc 의 N chunks 가 results 에 박제 → returned_ids 에 doc.id 중복 → 모든 graded
metric inflation. multi-query 의 reranker score 가 자주 0.90+ → 다수 case 영향.

변경 (baseline path 영향 0, multi-query 전용 invariant):
- app/services/search/search_pipeline.py:
  · _dedup_results_by_doc_id() helper 신규 (doc.id first-only, top score 보존)
  · search_with_rewrite() 의 rerank path 에 apply_diversity(top_score_threshold=2.0)
    강제 + 후속 _dedup_results_by_doc_id 적용
  · rerank=False path 도 _dedup_results_by_doc_id(unified_docs) 적용
- tests/test_query_rewriter.py — 신규 4 test (55/55 PASS)

🎯 진짜 측정값 (모든 dedup layer 적용, 51 case gemma):
  cold: NDCG 0.663 / Recall t≥2 0.729 / Recall t≥3 0.761 / p50 3692ms / p95 9992ms
  warm: NDCG 0.659 / Recall t≥2 0.721 / Recall t≥3 0.739 / p50 1588ms / p95 3514ms
  baseline (rewrite_backend=null): NDCG 0.644 / Recall t≥2 0.699 / Recall t≥3 0.761 / p50 378ms
  Dedup audit: gemma 0/51 ✓ 정상 (fix 작동, eval-dedup 42/51 → 0/51 회복)

Δ vs baseline (진짜 multi-query 효과):
  NDCG +0.019 (cold) / +0.015 (warm) — sub-noise level
  Recall t≥2 +0.030 (cold) / +0.022 (warm) — 소량 개선
  Recall t≥3 0.000 / -0.022 — 동등~약간 회귀
  latency p50 +876% (cold) / +320% (warm) — major cost
  category: english/standards/mixed 약간 우세 / exam/korean 약간 회귀

measurement chain 정정 history:
  Phase 3 (a41adb6) 0.927 — chunk_id 중복 inflation
  Rerank-Fix (b734fc5) 0.876 — doc_id 중복 잔재
  Eval-Dedup (3553573) 0.641 — eval layer 만 dedup
  Result-Dedup (본 PR) 0.663 — production + eval 둘 다 dedup ← 정확값

사용자 결정 필요 (3 path, json 박제):
  (a) rollback — marginal 개선이 latency cost 정당화 X
  (b) opt-in 유지 + PR-2Q-Cache-Prewarm 진입 (warm path 만 노출)
  (c) 1주 관찰 종료 후 (2026-05-31) 재결정 (현 상태 유지)

산출물:
  reports/v0_2_phase2q_result_dedup_gemma_{cold,warm}_2026-05-24.csv
  tests/search_eval/baselines/v0_2_phase2q_result_dedup_2026-05-24.json (요약 + 사용자 결정 옵션)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 04:48:50 +00:00
hyungi 3553573595 merge(eval): PR-Eval-GradedNDCG-Dedup closed — Phase 2Q 측정 inflation 정정 (NDCG 0.876 → 0.641, multi-query 실제 효과 ≈ 0) 2026-05-24 04:35:33 +00:00
hyungi 9dad5e6289 chore(eval): graded NDCG dedup + warning + audit stats (Phase 2Q inflation 정정) 
PR-Eval-GradedNDCG-Dedup. [[feedback_graded_ndcg_dedup_invariant]] cleanup.
plan pr-eval-graded-ndcg-dedup-stormy-tide.md.

변경:
- tests/search_eval/run_eval.py:
  · _dedup_returned_ids() helper — returned[:k] 첫 등장 순서 보존 dedup + count 반환
  · count_dedup() wrapper (audit 용)
  · ndcg_at_k + graded_ndcg_at_k 진입 시 dedup (NDCG > 1.0 invariant 강제)
  · QueryResult.dedup_count 필드 + csv schema 신규 column
  · evaluate() 에서 dedup_count > 0 시 stderr WARNING
  · print_summary 에 dedup audit stats (cases/total chunks + 정상/⚠️ flag)
- tests/search_eval/test_eval_graded_ndcg_dedup.py 신규 — 13 test:
  · _dedup_returned_ids 6 (empty / no-dup / dup-first / k-limit / count helper / Phase 2Q kw_001)
  · graded_ndcg invariant 5 (baseline 회귀 0 / dup 차단 / all-dup / exam_001 regression / empty grades)
  · ndcg_at_k binary dedup 1 + graded_recall set 변환 1

51/51 test PASS (13 신규 + 38 기존 회귀 0).

🚨 CRITICAL 측정 발견:
  dedup audit baseline = 0/51 정상 (single-query path 의 retrieval 가 doc unique 박제)
  dedup audit gemma = 42/51 (totaling 81 chunks dedup) ⚠️
  → _rrf_fuse_variants 의 representative 보존 logic 이 같은 doc_id 의 여러 SearchResult
    를 unique 가정. chunk_id dedup (Rerank-Fix) 이후에도 doc_id 중복 잔재.

정정값 (이번이 가장 정확):
  baseline NDCG 0.644 (이전 0.659 와 noise level diff)
  gemma NDCG 0.641 → Δ vs baseline = -0.003 (사실상 동일, multi-query 실제 net 효과 ≈ 0)
  latency p50 +1005ms (+266%) — 회귀
  Recall t≥3 -0.033 (회귀)

이전 박제값 (모두 inflation):
  Phase 3 (a41adb6) NDCG 0.927 — chunk_id 중복
  Rerank-Fix (b734fc5) NDCG 0.876 — doc_id 중복 잔재
  Category-Analysis (b00d9f5) NDCG 0.876 정정 박제 — 위와 동일

산출물:
  reports/v0_2_phase2q_eval_dedup_baseline_2026-05-24.csv (baseline 회귀 verify)
  reports/v0_2_phase2q_eval_dedup_gemma_2026-05-24.csv (실제 효과 측정)
  tests/search_eval/baselines/v0_2_phase2q_eval_dedup_2026-05-24.json (요약 + critical 권고)

권고 (사용자 결정 필요):
  1. Apply rollback 검토 — multi-query 의 실제 net 효과 ≈ 0 + latency 4x 회귀
  2. 또는 PR-2Q-Search-Result-Dedup 진입 (real fix _rrf_fuse_variants representative)
     후 재측정 → 실제 multi-query 효과 측정 후 Apply 결정

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 04:35:33 +00:00
hyungi b00d9f5e15 docs(eval): Phase 2Q Category-Analysis — standards/exam 회귀 진단 (inflation 정정) 
Apply rollout 후속 read-only 진단. Phase 3 측정 (commit a41adb6) 의 NDCG 0.927 + standards 1.441 + exam 1.109 = **측정 artifact (top-N doc 중복 박제 → graded NDCG inflation)**.

진단 path:
- script category_analysis_phase2q.py (csv parse + queries.yaml graded lookup + standards/exam 18 case 3-way top-5 박제)
- 회귀 큰 case top: kw_004/kw_009/kw_010 = Phase 3 inflation 1.631 → Rerank-Fix 정상 1.000 (baseline 동일, 회귀 0)
- kw_001/exam_004 = Rerank-Fix 가 baseline 대비도 회귀 (reranker chunk-level relevance 우선 → doc grade 3 가 rank 5 밀림)

정정값 박제:
- Phase 3 NDCG 0.927 → **Rerank-Fix 0.876 (정확값)**
- Δ vs baseline: +0.268 (inflated) → **+0.217 (실제 multi-query 효과)**
- standards 1.441 → 1.157 (vs baseline 0.873, +0.284)
- exam 1.109 → 0.918 (vs baseline 0.738, +0.180)

결론:
- **Apply rollout 결정 = 정정값 기준 invariant 유지** — +0.217 vs baseline = 유의미 net 개선
- standards -0.28 / exam -0.19 회귀 = false alarm (inflation 정정)
- 실제 회귀 case (kw_001/exam_004) = Apply 후 telemetry 박제 항목

산출물:
- tests/search_eval/baselines/v0_2_phase2q_category_analysis_2026-05-24.md (180+ lines, §1~8)
- tests/search_eval/scripts/category_analysis_phase2q.py (read-only csv parse script, reproducibility)

신규 feedback memory: graded-ndcg-dedup-invariant (NDCG > 1.0 = inflation 의심 invariant + dedup audit 필수)

후속 별 chore 후보:
- PR-Eval-GradedNDCG-Dedup — run_eval.py 의 graded NDCG 계산 dedup + NDCG > 1.0 warning
- PR-2Q-Search-Result-Dedup — _rrf_fuse_variants 의 representative doc_id 중복 audit

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 04:23:58 +00:00
hyungi fef5ddc5c8 merge(search): PR-2Q-Apply-Query-Rewrite-1 closed — opt-in rollout 시작, 1주 관찰 (~2026-05-31) 2026-05-24 04:01:49 +00:00
hyungi 59bde9a399 feat(search): phase-2q apply opt-in — production rollout 시작, 1주 관찰 (gemma-4) 
plan pr-2q-apply-query-rewrite-1-bright-meadow.md. Phase 2Q Diagnose closure +
Rerank-Payload-Fix (main 0257a5d) 완료 후 Apply rollout. opt-in path 가 Phase 1B/2
부터 이미 production 가동 중 → 본 PR 의 production 영향 0 (marker PR).

rollout 정책:
  · default = rewrite_backend null (single-query path, baseline 회귀 0 invariant)
  · 명시 opt-in = ?rewrite_backend=cand_multi_query_macmini (추천 gemma-4)
  · 대안 = cand_multi_query_macbook (qwen3.6, mixed/english 강점, MacBook 가동 시)
  · 1주 관찰 (2026-05-24 ~ 2026-05-31) → metric 정상 시 default ON 별 PR

변경 (production 영향 0):
- docs/phase_2q_apply_opt_in.md 신규 — 사용자 가시화:
  · 사용 방법 (query param + SvelteKit fetch 예시)
  · 1주 관찰 metric 목표 (cache hit ≥ 50% / LLM warm p50 ≤ 1500 / 503 ≤ 5/day / Recall t≥3 ≥ 0.74)
  · 추천 LLM 사유 (decision md §4 4-factor) + 대안 명시
  · Phase 2 QueryAnalyzer sequencing 박제 (영향 0, ask_events 0건 운영 관찰 후 확정)
  · Follow-up PR 5건 명시 (Telemetry / Alert / Default-ON / Cache-Prewarm / Category-Analysis)
- app/api/search.py — rewrite_backend query param description 갱신.
  Apply 진입 박제 + 추천 LLM 표시 + docs 링크. 동작 변경 0.
- tests/search_eval/baselines/v0_2_phase2q_apply_smoke_2026-05-24.json — production smoke:
  · opt-in path HTTP 200 + total_ms 957 (cache hit) + rerank_ms 109 (정상 호출) + fallback 0
  · baseline path HTTP 200 + total_ms 207 + rerank_ms 19 + fallback 0 (회귀 0 확정)

38/38 unit test PASS (회귀 0). main HEAD 0257a5d 위 branch.

Closure gate PASS:
  · docs 가시화 / search.py description / smoke json 박제
  · production smoke 양쪽 path 정상 + 회귀 0 verify
  · 메모리 갱신 + 1주 관찰 종료일 2026-05-31 박제

Follow-up: 1주 후 PR-2Q-Apply-Default-ON-1 (metric 정상 시) 또는 fix PR.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 04:01:49 +00:00
hyungi 0257a5d49e merge(search): PR-2Q-Rerank-Payload-Fix closed — Apply prereq invariant, 413 98%↓ + latency -48% 2026-05-24 03:55:15 +00:00
hyungi b734fc54af fix(search): Phase 2Q rerank payload — chunk_id dedup + cap 60 + TEI batch 64 (Apply prereq) 
plan pr-2q-rerank-payload-fix-resolute-haven.md. Phase 2Q multi-query path 의 reranker
413 Payload Too Large root cause = TEI 의 MAX_CLIENT_BATCH_SIZE=32 default (batch entries
한도) + multi-query 의 chunks 누적이 32 초과. MAX_BATCH_TOKENS 와 별개 (token sum 한도).

4 iteration 진단 history (json 박제):
  1) cap 60 + dedup = 413 다수 (batch 54 > 32)
  2) cap 30 + chunks_per_doc=1 = 413 0건 + NDCG 0.666 catastrophic (-0.261)
  3) cap 60 + dedup + TEI 16384 only = 413 46건 (batch size 한도 별개)
  4) cap 60 + dedup + TEI 16384/64 = 413 1건 + NDCG 0.876 (FINAL)

변경:
- app/services/search/search_pipeline.py:
  · _dedup_chunks_by_id() 신규 helper — chunk_id (None 시 doc.id) 기준 first-only.
    variant 별 same chunk 중복 누적 회피, 첫 등장 variant 보존.
  · PHASE2Q_RERANK_INPUT_CAP=60 + PHASE2Q_CHUNKS_PER_DOC=2 신규 상수 (baseline
    MAX_RERANK_INPUT=200 / MAX_CHUNKS_PER_DOC=2 와 별도).
  · search_with_rewrite() merge 후 dedup wire-up + rerank input cap swap.
- docker-compose.yml reranker env (사용자 결정, plan out-of-scope 정정):
  · MAX_BATCH_TOKENS 8192 → 16384 (token sum 한도)
  · MAX_CLIENT_BATCH_SIZE 32 → 64 신규 추가 (batch entries 한도 — root cause)
  · GPU VRAM free 6199MiB 충분 사전 verify.
- tests/test_query_rewriter.py: _dedup_chunks_by_id 5 test + PHASE2Q_* constants test.
  38/38 PASS (기존 32 + 신규 6).

측정 결과 (51 case, gemma backend, snapshot 25180/56526):
  vs Phase 3 (commit a41adb6 NDCG 0.927, 413 다수):
  · NDCG 0.876 (-0.051 acceptable, plan 변수 격리 invariant 충족)
  · Recall t≥2 0.721 (+0.034 회복)
  · Recall t≥3 0.739 (+0.011)
  · latency p50 1421ms (-1336ms, -48%) / p95 3392ms (-6292ms, -65%) major win
  · 413 fallback 1/51 (98%↓ from 다수) + reranker batch error 0
  · 카테고리 english_only +0.34 / standards -0.28 / exam -0.19 (Apply 후 분석 항목)

closure gate PASS:
  · unit test 38/38, production smoke 413 0
  · 51 case 413 < 5/51 (1건만)
  · latency 대폭 개선
  · NDCG threshold 0.92 미달 단 plan invariant (production 평가 단일 변수) 충족
  · Apply PR-2Q-Apply-Query-Rewrite-1 진입 ready

산출물:
  · reports/v0_2_phase2q_rerank_fix_2026-05-24.csv (raw)
  · tests/search_eval/baselines/v0_2_phase2q_rerank_fix_2026-05-24.json (4 iter 진단 박제)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 03:54:59 +00:00
hyungi 1ae7802485 Merge pull request 'Feat/ds ai routing policy' (#23) from feat/ds-ai-routing-policy into main 
Reviewed-on: #23
 2026-05-24 12:20:49 +09:00
hyungi 711d4952a2 merge(search): Phase 2Q Query Rewrite Diagnose closed — H1 multi-query gemma-4 추천 2026-05-24 02:57:59 +00:00
hyungi c57e4c52dc docs(eval): Phase 2Q Diagnose Phase 4 — decision tree md + Apply PR 백로그 
phase-2q-query-rewrite-diagnose.md v6 plan §7 Phase 4 closure.
Phase 3 commit a41adb6 의 3 측정 결과 + 4 factor weighted decision.

decision = H1 (both backends NDCG net 개선 ≥ +0.26):
- 추천 Apply LLM = cand_multi_query_macmini (gemma-4)
- 사유: F3 ⭐ 24/7 가동 + F1 NDCG 0.927 dominant + F4 cold latency 우세
- 대안: qwen (mixed/english 강점 + MacBook always-on 의향 시)

산출물:
- tests/search_eval/baselines/v0_2_phase2q_decision_2026-05-24.md (180 lines)
  · §1 결정 요약 / §2 측정 표 / §3 카테고리 회복 / §4 4-factor weighted
  · §5 분석 노트 5건 (multi-query 효과 / variants 구성 / cache hit / Recall 회귀 / Phase 3 incident)
  · §6 closure gate (branch close 사용자 결정 보류)
  · §7 follow-up PR 백로그: Apply 1 + 별 chore 2 + Extended 4 + Cloud 1 + Cleanup 1
  · §9 사용자 검토 항목 5건

Phase 2Q Diagnose closure 완료. Apply PR 진입 = 사용자 LLM 선택 + sequencing 결정 후.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 00:57:48 +00:00
hyungi a41adb63a0 fix(search): Phase 2Q variants bug fix + Phase 3 3 measurement 박제 
Phase 3 cold 측정 1차에서 NDCG 0.033 catastrophic 발견 — 모든 query 에 동일 variants
반환. root cause = _call_llm 이 user 메시지 1개에 prompt template 전체 박음. LLM 이
actual query 인식 못 함. fixture request_body 형식 (system=prompt / user=query) 과
mismatch. fixture-first invariant 위반.

fix:
- app/services/search/query_rewriter.py _call_llm — system/user 메시지 분리.
  fixture request_body 와 단일 source-of-truth. _render_prompt 는 [deprecated] 유지.
- tests/test_query_rewriter.py — Phase 3 regression test 2:
  · _call_llm 가 system + user 분리 호출 verify (httpx.AsyncClient monkeypatch)
  · qwen backend = response_format 미사용 verify
- 32/32 unit test PASS.

Phase 3 측정 (fix 후 재측정, 51 case × 3 candidate × cold/warm = 5 run):
- baseline_rebaseline (rewrite_backend=null): NDCG 0.659 = Phase 2A 0.659, diff 0.000 PASS
- cand_multi_query_macmini cold: NDCG 0.927 (Δ +0.268), p50 2757ms / p95 9684ms
- cand_multi_query_macmini warm: NDCG 0.927 동일, p50 998ms (cache hit -64%)
- cand_multi_query_macbook cold: NDCG 0.919 (Δ +0.260), p50 3647ms / p95 5202ms
- cand_multi_query_macbook warm: NDCG 0.919 동일, p50 873ms (cache hit -76%)

핵심 약점 회복 (gemma / qwen):
- mixed 0.39 → 0.57 / 0.65
- korean_only 0.51 → 0.71 / 0.67
- standards 0.87 → 1.44 / 1.31
- exam 0.74 → 1.11 / 1.04

decision = H1 (both backends 유의미 net 개선). LLM 선택 = Phase 4 decision md 별 step.

산출물:
- reports/v0_2_phase2q_*.csv (5 raw run_eval output)
- tests/search_eval/baselines/v0_2_phase2q_results_2026-05-24.json (요약 + incident 박제)

follow-up:
- rerank 413 Payload Too Large 다수 관찰 (RRF fallback 작동, NDCG 영향 없음). Apply PR 전 별 chore — chunk dedup 또는 reranker batch cap 검토.
- p95 cold 9684ms 매우 큼. production rollout 시 cache prewarm 정책 필수.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-24 00:51:56 +00:00
hyungi ecd2350c15 feat(search): Phase 2Q Diagnose Phase 2 — multi-query retrieval fusion 
phase-2q-query-rewrite-diagnose.md v6 plan §5.5 + §7 Phase 2.
Phase 1B 3e6866b (scaffold + dispatcher) 위 retrieval 합성 wire-up.

신규:
- search_pipeline._rrf_fuse_variants() — N variant ranked list RRF 합성.
  fusion_service.RRFOnly 알고리즘 동일 (k=60), 첫 등장 variant representative 보존.
- search_pipeline.search_with_rewrite() — variant N 별 retrieval+fusion 후
  unified RRF (cap 60) → reranker 1회 (query=원본 q) → diversity+freshness+display.
  · per-variant K = 50//3 = 16 (PHASE2Q_PRODUCTION_TOPK//N, A1 채택)
  · variant 별 retrieval asyncio.gather 병렬
  · chunks_by_doc merge (variant 무관 unified reranker input)
  · production fusion_service.get_strategy() + rerank_chunks() 재사용
- 상수: PHASE2Q_PRODUCTION_TOPK=50, PHASE2Q_UNIFIED_CAP=60, PHASE2Q_RRF_K=60.

수정:
- search_pipeline.run_search() — rewrite_backend param 추가. hybrid + cand_<slug> 시
  search_with_rewrite() 위임. baseline/None 시 기존 single-query path 그대로 (invariant).
- app/api/search.py — Phase 1B scaffold discard call 제거. run_search 에 rewrite_backend
  전달. ValueError → 400 (unknown_rewrite_backend 우선 분기) / RuntimeError → 503
  (rewrite_llm_unavailable).
- tests/test_query_rewriter.py — Phase 2 test 9개 추가:
  · _rrf_fuse_variants 6 (single / overlap accumulation / representative / cap limit /
    empty / rank position)
  · search_pipeline import + run_search rewrite_backend default=None signature 1
  · PHASE2Q_* constants 1
  · DATABASE_URL dummy 주입 (api.search import → SQLAlchemy engine init 회피)

30/30 unit test PASS (Phase 1B 21 + Phase 2 9).

baseline 회귀 0 invariant:
- run_search(rewrite_backend=None) → 기존 path 100% 그대로 (분기 first line guard)
- run_search(rewrite_backend=baseline) → 동일
- mode != hybrid → multi-query path 비활성 (text-only/vector-only/trgm 영향 0)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-23 22:41:50 +00:00
hyungi 3e6866b4ae feat(search): Phase 2Q Diagnose Phase 1B — scaffold + dispatcher 
phase-2q-query-rewrite-diagnose.md v6 plan Phase 1 의 fixture 외 잔여.
Phase 1A 446ba82 위 dispatcher + cache + LLM call + API param + eval flag + 21 unit test.
retrieval 합성 (search_with_rewrite) 은 Phase 2 별 commit.

신규:
- app/services/search/query_rewriter.py — LLM_BACKEND_MAP + _resolve + cache + rewrite()
  · slug-based allowlist (no silent fallback), httpx 직접, Priority.FOREGROUND semaphore
  · sampling 박제 (gemma response_format json_object / qwen prompt rule only — Phase 0 inspect 9)
  · manual TTL cache (query_analyzer 패턴 1:1, sha256[:32] NFKC key, LLM_REWRITE_TIMEOUT_MS=15000)
- tests/test_query_rewriter.py — 21 test PASS (resolve / cache key / parser / cache TTL / constants)

수정:
- app/api/search.py — ?rewrite_backend= query param + 400 unknown / 503 unavailable.
  scaffold = call but discard variants (retrieval path 영향 0). Phase 2 에서 합성.
- tests/search_eval/run_eval.py — --rewrite-backend flag + 4 hot spot wire-up.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-23 22:25:03 +00:00
hyungi 446ba82c91 feat(eval): Phase 2Q Diagnose Phase 1A — fixture (4 카테고리 × 2 LLM) + prompt v1 
phase-2q-query-rewrite-diagnose.md v6 plan 의 Phase 1 fixture 박제 (G0-1 + G0-2).

산출물:
- app/prompts/query_rewrite.txt — multi-query rewrite prompt v1 (3 variants: 원본 + 한국어 rephrase + 영어 번역)
- tests/fixtures/macmini_gemma4_query_rewrite_response.json — 4 카테고리 (korean_only/mixed/english_only/exam)
- tests/fixtures/macbook_qwen_query_rewrite_response.json — 4 카테고리 동일

inspect 9 결과 (2026-05-24):
- Mac mini gemma-4-26B-A4B :8801 = response_format json_object 지원
- MacBook qwen3.6-27B-8bit :8810 = response_format json_object 미지원 (120s hang) — prompt rule only
- prompt rule \"no markdown, no code fence\" 강제 시 둘 다 strict JSON (gemma 도 fence wrap 없음)
- parser fallback (markdown fence regex) 유지 — 첫 호출 prompt 없을 때 wrap 관찰 사례

8 호출 측정:
- gemma 1.16~1.36s / qwen 1.93~2.24s (warm)
- variants 의미 일관 + 도메인 용어 (ASME/Section VIII/압력용기/가스기사) verbatim preserve
- 한국어→영어 cross-lingual translation 자연

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-23 22:09:29 +00:00