What problem does it solve? Tabular analysis often produces wrong answers silently: duplicated refund events inflate revenue, null join keys fabricate matches, averages of averages distort weighting, and timezone shortcuts corrupt calendar windows. This Skill enforces data contracts, join-grain discipline, missingness accounting, and reproducible execution so reported numbers survive independent reconciliation. ## Core Features & Use Cases - Data contract and reconciliation workflows: Define population, identity, grain, and missingness before aggregating, then gate conclusions on executed checks like eligible = observed + missing. - Engine-specific guidance: Dedicated references for pandas ingestion and calendar endpoints, Polars lazy semantics and streaming, and DuckDB local SQL, joins, and spilling. - Statistical design discipline: Match inference to independent units, distinguish descriptive findings from causal claims, and report coverage alongside every incomplete measure. - Use Case: Finance needs net booked revenue by region from CSV exports containing retransmitted refund events and blank amounts. The Skill deduplicates by event identity, aggregates refunds at order grain before joining, keeps unknown amounts separate from zero, and delivers a rerunnable SQL script with reconciliation evidence. ## Quick Start Ask the agent to analyze your local CSV or Parquet files with this skill, for example to compute revenue by region from orders and refunds exports while preserving unknown amounts and delivering a rerunnable script.
