正文
获取原文件README.md
v0.2.0 修复发布
新增逐行 structured_row_dispositions,未抽取/未授权记录会阻断 ready;完整保留 raw_dose/raw_route/raw_frequency、就诊键、起止时间。新增可选 project_mimic_tables.py,只做明确字段投影/字典连接,不做脱敏或临床推断;其输出必须重走 D001–D003。
D005 converts D004-approved narrative sections and D003-sanitized structured clinical rows into auditable ClinicalFactDraft objects.
It deliberately stops before context/state resolution, terminology coding, unit normalization, medication action semantics, reconciliation, and timeline construction.
Runtime
Python 3.11+ standard library only. No network. No auto-install.
Typical flow
python scripts/prepare_fact_extraction.py \
--document-manifest /path/to/d004/document_structure_manifest.json \
--deidentified-manifest /path/to/d003/deidentified_manifest.json \
--output-root ./runs/haonan-d005
# Codex may then add explicit-span drafts to agent_fact_drafts.jsonl and must record section completion in agent_extraction_completion.json.
python scripts/finalize_fact_drafts.py \
--run-dir ./runs/haonan-d005/<package-id>
python scripts/validate_fact_drafts.py \
--manifest ./runs/haonan-d005/<package-id>/fact_extraction_manifest.json
当前版本:v0.2.1
prepare及finalize执行包级隐私前置校验,finalize重新检查审批/文件状态,不能因单文件sanitized跳过全包BLOCKED。补充canonical vital名称;保留eICU原始signed minute offset、anchor类型、修订offset和双单位来源。新增显式project_eicu_lab.py投影器,产物必须重新走D001–D003。
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。