正文
获取原文件README.md
v0.2.0 修复发布
本阶段核心结构化逻辑保持原样；仅纳入统一回归。真实数据集无病历全文，因此本轮没有验证中文全文章节准确率。
D004 is the document-structure layer of the Haonan D-series medical-data pipeline.
It reads only D003-sanitized surfaces and converts narrative clinical documents into auditable StructuredDocument JSON objects with:
- conservative document type
- canonical Chinese/English clinical sections
- exact character offsets
- source provenance
- privacy/model-visibility state
- structural review flags
It intentionally does not extract clinical facts. Diagnosis/symptom/medication/lab/procedure extraction begins in D005.
Quick start
python scripts/structure_documents.py \
--deidentified-manifest /path/to/d003/deidentified_manifest.json \
--phi-report /path/to/d003/phi_gate_report.json \
--output-root ./runs/haonan-d004
If D003 marked free text sanitized_review_required, supply a local privacy review artifact before allowing Codex/D005 to read that text:
python scripts/structure_documents.py \
--deidentified-manifest /path/to/d003/deidentified_manifest.json \
--phi-report /path/to/d003/phi_gate_report.json \
--privacy-review-decisions /path/to/privacy_review_decisions.json \
--output-root ./runs/haonan-d004
Validate:
python scripts/validate_structure.py \
./runs/haonan-d004/<package-id>/document_structure_manifest.json
Self-test:
cd scripts
python self_test.py
Safety posture
D004 uses only the Python standard library, never accesses the network, never installs packages, never executes input code, and never opens the raw pre-D003 source package.
Instruction-like text inside a clinical document is treated as untrusted data and may only generate a security flag.
当前版本：v0.2.1
读取D003数据前校验包级downstream/model gate、清单与报告package一致性、声明的产物摘要和路径边界。blocked/报告缺失/文件篡改立即拒绝。model review_required不等同模型可见许可，仍执行逐文件审核。
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。