data-analyzer

Start by clarifying the decision, target metric, and time window, then profile the data for missing values, duplicates, and type issues. Compute counts, rates, and quantiles, compare relevant groups or periods, and report each finding with its denominator and supporting calculation.

Use robust statistics such as medians and quantiles when outliers make ordinary averages misleading. Report center, spread, and range together, and never silently drop or replace extreme values without making exclusions visible.

No, the Skill explicitly avoids claiming causation without a design that supports it. It separates observed facts from inference, tests alternative explanations, and states uncertainty and sample limitations for every finding.

The inspection covers missing values, duplicates, impossible values, inconsistent categories, parsing failures, and date coverage. The original data is preserved, and all exclusions are made visible in the final report.

The Skill states clearly when the data cannot support the requested conclusion rather than fabricating results. It never invents missing observations or silently replaces them with zeros.