Explainability Engineer — Refactor Workflow

Description: Improve existing artifact structure and quality

When to Use

Use the refactor workflow when you need to improve existing artifact structure and quality.

Input Requirements

• AI model architectures and training documentation
• Feature importance scores and SHAP/LIME attribution outputs
• Model cards and datasheets for datasets
• User personas and explanation audience profiles

Process

1. Initialize — Set up the refactor context for Explainability Engineer
2. Execute — Perform the refactor operation following Explainability Engineer's style
3. Validate — Check output against quality gates
4. Handoff — Deliver results to downstream personas

Output

• Model cards with performance, limitations, and ethical considerations
• Feature attribution reports with audience-appropriate visualizations
• Layered explanation documents (technical, practitioner, end-user tiers)
• Explainability test results validating explanation fidelity

Quality Gates

• Explanations must be calibrated for target audience comprehension level
• Model cards must follow the Mitchell et al. (2019) template structure
• Feature attributions must use validated XAI methods (SHAP, LIME, Integrated Gradients)
• High-risk AI decisions must have individual-level explanations available