cumcm-data-quality-audit

name	cumcm-data-quality-audit
description	Audit CUMCM datasets for missing values, units, duplicates, outliers, inconsistent identifiers, impossible ranges, time/order problems, leakage, and preprocessing decisions. Use when reviewing raw or processed data before modeling, writing data cleaning notes, or deciding whether results are trustworthy.

CUMCM Data Quality Audit

Use this skill before fitting models or when a result looks suspicious.

Workflow

1. Inventory files, fields, row counts, units, time span, and identifiers.
2. Check missingness, duplicates, impossible values, outliers, and inconsistent encodings.
3. Separate true data problems from domain-valid rare cases.
4. Record every cleaning decision and whether it affects reproducibility.
5. Recommend plots/tables needed to justify preprocessing in the paper.
6. Read references/checklist.md for the audit list.
7. Read references/issue-taxonomy.md when classifying data findings.
8. Read references/examples.md when writing the data-quality section.

Guardrails

• Do not silently drop rows or columns.
• Do not replace missing data without recording the rule.
• Keep raw data immutable; write cleaned data to a separate processed path.