正文
获取原文件README.md
v0.2.0 修复发布
补齐本次来源数据中明确的项目名称和 meq/L 单位别名;不推导新的临床换算公式。按项目/单位兼容性保留已知原单位。重复候选必须限定同患者、同就诊等身份;不同就诊重用 specimen_id 不自动合并。
Codex Skill for conservative, auditable normalization of de-identified laboratory and vital-sign facts.
Position in the D-series
D001 -> D002 -> D003 -> D004 -> D005 -> D006 -> D007
D007 is one branch of domain normalization after D006. D008 handles medications and D009 handles diagnoses/problems.
Main outputs
- normalized measurement facts with parsed numeric/comparator values;
- conservative unit normalization using a small reviewed allowlist;
- parsed reference ranges and derived low/normal/high flags;
- specimen/accession identity preservation;
- duplicate/repeat candidates without deleting source facts;
- text-free review queue and aggregate manifest.
Runtime
Python 3.11+ standard library only. No network, no auto-install, no external medical framework.
Important boundary
D007 does not assign LOINC or OMOP codes and never guesses missing units.
当前版本:v0.2.1
用已识别目录kind纠正lab/vital类型,保存classification_lineage,不把生命体征再按lab传下去。保留unit_system_raw与unit_interface_raw,冲突来源单位仍review,不臆造换算。
以SKILL.md顶部发布合同和references/repair-v0.2.1.md为准,历史自测描述不代表全面临床验收。