#&idZddlmZddlZddlmZmZmZddlm Z eje Z dZ dZee fzZdZ d dd Z d dd Z d dd Zdd Zy)uG3.1 evidence/inference 4-field helpers — single source. LLM strict schema 가 강제 못하는 contract consistency 와 옛 checkpoint backfill 을 처리. 두 step (scene_consistency, scene_detail) 이 공유. ) annotationsN)AnyDictUnion)AppErrorlegacy)highmediumlow) source_factsvisual_inferencescreative_decisionsct|ts|Sd}tD]!}||vst||trg||<d}#d|vs |dtvr t |d<d}n |r t |d<|r|rt jd||S)uG3.1 lazy backfill — 4 evidence 필드 누락 시 default 주입. **옛 checkpoint 전용** (CRITICAL #2). 새 LLM 응답에는 호출 금지 — strict schema 가 4 필드 강제. 이 함수가 빈자리 채우면 strict 우회 silent backfill 회귀 발생. Mutation in-place. file 안 건드림. Type guard (Cl-M1 fix): item 이 dict 아니면 그대로 반환 — caller 가 list-of-X iter 중 비-dict element 를 만나도 AttributeError 안 남. FT confidencez(evidence backfill marker=legacy where=%s) isinstancedictEVIDENCE_FIELDSlistVALID_STORED_CONFIDENCELEGACY_CONFIDENCEloggerdebug)itemwhere backfilledfields V/Users/manta/Documents/Projects/TheRoad-I1/backend/app/core/steps/_evidence_helpers.py_normalize_evidence_fieldsrs dD ! J   JtE{D$ADKJ! 44 #5=T#T.\ .\e ?G Kct|ts|S|jd}t|tsg|d<|S|D]}t |||S)u fixed_elements 각 element normalize. PROBLEM #3: malformed container type guard. Codex BLOCKING 2 fix: helper 가 'list 아니면 silent return' 으로 끝내면 downstream consumer (shot_dependency_t2i / t2i_review / scene_checkpoint_loaders) 가 list 가정하고 ``v.get()`` 호출 → AttributeError. fixed_elements/t2i_variations 가 malformed 일 때 helper 가 in-place 로 빈 list 강제 → consumer 정상 (빈 iter, noop). 옛 데이터 손실은 어차피 malformed 라 의미 없음. fixed_elementsrrrgetrr) scene_resultrfixedfes r#_normalize_scene_consistency_resultr(7sZ lD )   - .E eT ")+ %&"2U3 rct|ts|S|jd}t|tsg|d<|S|D]}t |||S)ut2i_variations 각 variation normalize. type guard. Codex BLOCKING 2 fix: malformed t2i_variations 가 list 아니면 in-place 빈 list 강제 (consumer AttributeError 차단). t2i_variationsr"r#)r%rvars_vars r_normalize_scene_detail_resultr-NsZ lD )   - .E eT ")+ %&"3e4 rcP|jd}|tk(rtd|dd|jdxsg}|jdxsg}|jdxsg}|s|d k7rtd|d |d d|s|s|s|d k7rtd|d |dy y y y )upost-parse contract validator (IMPROVEMENT #2). LLM strict schema = 4 필드 존재만 강제. contract consistency 추가 검증: - confidence == "legacy" → reject (LLM 출력 금지, adapter-only) - source_facts == [] AND confidence != "low" → violation - 4 lists 모두 [] AND confidence != "low" → violation AppError(code="step.contract_violation") raise → step_runner retry path. **새 LLM 응답에만 호출**. 옛 cp 는 normalize 가 처리. rzstep.contract_violationu9: LLM 출력에 confidence='legacy' 금지 (adapter-only)i)codemessage status_coder r rr u$: source_facts=[] 인데 confidence=u (low 여야 함)u-: 4 lists 모두 비어있음 — confidence=N)r$rr)r step_namersfvicds rassert_fresh_llm_evidencer6as,'J&&* k!Z[  . ! 'RB % & ,"B & ' -2B *%* k!Ej\Qbc  " e 3* k!Nzl[  !4"Br))rUnion[Dict[str, Any], Any]rstrreturnr8)r%Dict[str, Any]rr9r:r;)rr;r2r9r:None)__doc__ __future__rloggingtypingrrrapp.core.errorsr getLogger__name__rrVALID_LLM_CONFIDENCErrrr(r-r6rrrFs ###$   8 $0.2C1EEM46 $-0F02 ),002 ),&  r