extractor-quality-check

Persona-driven datalake quality assessment for the extractor pipeline. Two named assessors co-preside over extraction quality using dynamic annealing thresholds.

Prompt Iteration Rule (NON-NEGOTIABLE)

Margaret/Jennifer persona assessment prompts MUST be validated through /prompt-lab before deployment. NEVER hand-craft assessment system prompts in Python strings.

• Before deploying: /prompt-lab eval the assessment prompt against ground truth
• Comparing variants: /prompt-lab compare across models

Assessors

Architecture