Instruction file imported from kelvinlim/localexpert (.github/instructions/data-handling.instructions.md). Copyright stays with the author.

Data handling

• Keep data local. Never upload the dataset, or suggest uploading it, to any cloud or web service. This workflow exists so sensitive data stays on the machine.
• Report missingness before modelling. Show missing counts per variable (and per row when relevant) and consider the mechanism (MCAR / MAR / MNAR) before deciding how to handle it.
• Never silently drop rows. If you exclude cases (listwise deletion, filtering), state how many and why, and prefer principled handling (e.g. imputation) when missingness is substantial.
• Inspect outliers, don't quietly delete them. Flag extreme values, look at them, and decide explicitly (cap / transform / exclude / keep) — reporting the choice and its effect.
• Note the data's shape and provenance (rows, columns, source) when you load it, so the record is self-explanatory.