Analyze Data

Use this skill for decision science and analytical work: querying canonical metrics, profiling data in-process, evaluating statistical drift, conducting causal inference, and producing verifiable stakeholder reports.

When to Use

• answering business, product, and KPI questions using conformed lakehouse tables
• querying certified metrics via Semantic Metric Layers (dbt MetricFlow, Cube.js, Apache Ossie)
• executing fast in-process exploratory analytics on Parquet/Iceberg extracts via DuckDB and Polars
• auditing statistical feature and cohort distribution drift (PSI, Two-Sample KS-test, Tukey's IQR)
• evaluating A/B tests with automated Sample Ratio Mismatch (SRM) checks and causal DAG inference
• producing decision-ready quantitative reports with explicit fact vs interpretation separation
• tracking agent query costs as a FinOps line item for AI agent data access
• Apache Ossie (Incubating) universal semantic layer for vendor-neutral metric portability
• DuckDB v2.0 readiness (breaking storage format, VARIANT type, Quack server mode)
• FastMCP sqlglot AST validation for anti-hallucination gateway

Core Rules