building-gold-corpus

Builds synthetic OpenMed annotation corpora for clinical NER and de-identification evaluation.

5.0k|615|Updated Oct 4, 2025

Search Tags:#synthetic data #de-identification #label studio #clinical annotation #gold corpus #brat #evaluation fixtures

Ready to run

npx skills add https://github.com/maziyarpanahi/openmed --skill building-gold-corpus

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: building-gold-corpus
Source: https://github.com/maziyarpanahi/openmed/tree/main/skills/building-gold-corpus
Command: npx skills add https://github.com/maziyarpanahi/openmed --skill building-gold-corpus

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps teams create trustworthy evaluation data for OpenMed clinical NER and de-identification models without committing real patient information or licensed clinical corpora.

Core Features & Use Cases

Label Schema Design: Define canonical entity labels, boundary rules, examples, and edge-case decisions.
Annotation Project Setup: Create annotation guidelines and configure BRAT or Label Studio workflows.
Synthetic Gold Fixtures: Generate validated, character-offset annotations with disciplined document-level train, development, and test splits.
Use Case: Build a synthetic held-out corpus for measuring model quality, checking subgroup performance, and enforcing CI release gates while keeping sensitive datasets outside the repository.

Quick Start

Use the building-gold-corpus skill to create a synthetic OpenMed annotation project with a label schema, guidelines, tool configuration, and disjoint train, development, and test fixtures.

Frequently Asked Questions about building-gold-corpus

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I create synthetic clinical evaluation data for de-identification models?▼

Synthetic gold-standard corpora provide fabricated clinical text with character-offset validated annotations, enabling trustworthy evaluation of named-entity recognition and de-identification models without risking patient privacy or committing licensed clinical corpora to your repository.

How do I set up an annotation project in Label Studio for clinical NER?▼

Configure Label Studio by defining canonical OpenMed entity labels, establishing boundary rules and edge-case decisions, and applying strict annotation guidelines to ensure consistent clinical named-entity recognition tagging across annotators.

What is a gold corpus in clinical named-entity recognition evaluation?▼

A gold corpus is a validated set of annotated clinical text with canonical OpenMed labels and character-offset validation, used to measure named-entity recognition model quality, check subgroup performance, and enforce CI release gates without using licensed external corpora.

Can I use BRAT configuration for clinical de-identification annotation workflows?▼

Yes, BRAT configuration supports clinical de-identification annotation workflows by applying canonical OpenMed label schemas, character-offset validation, and annotation guidelines to generate synthetic gold-standard evaluation fixtures for de-identification models.

How do I split synthetic clinical data into train development and test fixtures safely?▼

Split synthetic clinical data safely by enforcing strict document-level partitioning to create disjoint train, development, and test fixtures. This leakage-safe separation keeps committed synthetic data strictly distinct from licensed external corpora during evaluation.

What are the limitations of using synthetic data for clinical NER evaluation fixtures?▼

Synthetic clinical NER evaluation fixtures require canonical OpenMed labels and character-offset validation, and cannot substitute for licensed external corpora in training. They serve strictly as committed synthetic data for measuring model quality and enforcing CI release gates.