Metrics Board: Building an Agent-ready Metrics Layer

This post details the architecture and design choices behind Pinterest's Metrics Board platform. Key contributions include: a YAML-based inventory system for metric definitions treated as code, automated Airflow DAG generation for metric computation using SparkSQL/PrestoSQL and sqlglot for dependency analysis, a dual-engine quality assurance system (rule-based and ML-based anomaly detection via Warden), and a multi-phase publishing mechanism to integrate with PinCat, Helium, Superset, and StarRocks. The platform aims to streamline metric creation from days/weeks to hours and provide a trusted, discoverable source of metrics for human and agentic analytics.