What problem does it solve? Analysts routinely build reports and models on datasets they have never verified, so grain bugs, sentinel values, and missing time ranges silently corrupt every downstream number. This Skill enforces a structured first pass over any new dataset so defects are caught in exploration rather than in the executive review. ## Core Features & Use Cases - Grain Verification: Confirms what one row actually represents using uniqueness checks before any aggregation, catching versioned rows and join fan-out. - Systematic Profiling: Covers column types, null rates, disguised sentinel values, distributions, time coverage, cross-column invariants, and outlier characterization. - Written Deliverables: Produces a one-page findings note plus a cleaning worklist for downstream pipelines instead of an unstructured notebook of plots. - Use Case: Given a subscriptions table for churn analysis, the workflow reveals the grain is status-changes rather than subscriptions, finds a two-week pipeline outage, and flags internal test accounts — before the churn rate is ever computed. ## Quick Start Ask the agent to run exploratory data analysis on your dataset, for example: "Profile this subscriptions table and tell me what one row represents, what's missing, and what looks suspicious."
