Data Analysis Evidence and Review Checklist
Use this checklist before reporting a metric, comparison, or causal interpretation.
- Use
Globto locate schemas and data notes,Grepto trace metric definitions, andReadto inspect the authoritative query or documentation. - Define the unit of analysis, population, time window, exclusions, and denominator before calculating a result.
- Keep raw observations, transformations, and interpretation separate. Show the query or formula when the user can safely reproduce it.
- Check missingness, duplicate records, timezone boundaries, sample size, seasonality, and selection effects. Treat correlation as non-causal unless the design supports causality.
- Report the result, uncertainty, decision implication, and one validation or sensitivity check.
Use only the user's existing authenticated data environment. Never request, print, or embed credentials. If authentication is unavailable, stop at a reproducible query plan and explain the required access without inventing results.