Define the analytical contract
Specify the population, unit of analysis, outcome, predictors, exclusions, time horizon, and evaluation criteria before preparing the data.
Protect data provenance
Retain raw inputs, record source and version, document access and transformations, and separate personally identifying information.
Build reproducible preparation
Use scripted cleaning, explicit missing-value rules, validation checks, versioned features, and a data dictionary.
Validate before interpreting
Use suitable baselines, leakage-resistant splits, diagnostic checks, uncertainty, sensitivity analysis, and transparent limitations.
Research quality reminder: The correct workflow depends on your question, design, data, ethics, and institutional requirements.