D005 临床事实提取

从已去标识化的临床章节和结构化记录中提取明确出现的症状、诊断、检验、用药与操作事实，保留原文和字段依据。

个人 Skill知识加工22 个文件浏览次数 0

获取完整包获取 SKILL.md

开始使用

下载前阅读 SKILL.md 中的使用条件，并确认所用工具支持对应能力。

输入资料: Required: D004 document_structure_manifest.json D003 deidentified_manifest.json Optional but useful: D002 schema_linkage_profile.json for patient/encounter key hints D005 may consume two approved source classes: narrativ… 阅读本节完整说明
交付内容: clinical_fact_drafts.jsonl — finalized local fact drafts fact_extraction_manifest.json — text-free aggregate manifest fact_extraction_worklist.json — text-free extraction plan agent_extraction_completion.json — text-free… 阅读本节完整说明

获取完整包，保留配套文件的目录结构。
按照原文准备所需工具、依赖与输入资料。
阅读对应使用步骤与输入输出说明，在自己的环境中验证结果。

使用提示摘自原文，配套文件保留来源版本；实际运行条件以各包说明为准。

正文

获取原文件

references/clinical-fact-draft-contract.md

D005 emits mention-level, pre-context facts. A draft is not yet a final ClinicalFact.

Hard boundaries:

concept.surface_text must be explicitly present in the cited source.
concept.normalized_name is always null in D005.
Polarity, certainty, temporality, experiencer, active/resolved state, medication action resolution, terminology codes, and OMOP concept IDs are not decided here.
Narrative provenance uses exact zero-based offsets into the D004 full_text.
Structured provenance uses a zero-based data-row index plus field name.
Every draft carries context_resolution = pending_d006 and terminology_resolution = pending_d012.

A D005 fact can still carry raw attributes that are explicitly encoded in the same source record/span, such as:

raw lab value/unit/reference range
raw medication dose/route/frequency/action text
source diagnosis code
specimen/accession identifiers
structured event timestamp

These are copied, not normalized or interpreted.