Run a Disciplined Clean-Up Pass Before Analysing a Dataset

You are a meticulous data analyst guiding a hands-on clean-up of a real business dataset before anyone draws conclusions from it.

<context>
[THE DATASET — source, rough row count, and the column names with a few example values each — e.g. "Square sales export, ~12,000 rows: Date, Item, Qty, Gross, Customer ID"]
[THE ANALYSIS I WANT AFTERWARDS — e.g. "which products are actually profitable by month"]
[MY TOOLS — e.g. "Excel", "Google Sheets", "Python/pandas"]
[KNOWN MESSES — e.g. "staff typed product names by hand", "refunds appear as negative rows, sometimes"]
</context>

Before prescribing steps, work out which quality problems could genuinely corrupt MY stated analysis (duplicates inflate revenue; inconsistent product names split a bestseller into five) and rank them — the checklist must be ordered by damage to my conclusion, not by textbook order.

<task>
1. Produce the ordered checklist. Each item: the check — how to run it in MY tool (exact formula, menu path or pandas snippet) — what a bad result looks like — the fix and its risk.
2. Cover, as relevant to my columns: duplicates, date parsing and format traps, inconsistent categories and names, missing values (and whether missing means zero, unknown or not-applicable, column by column), impossible values, refunds and negatives, and unit or currency consistency.
3. For hand-typed category columns, give the concrete standardisation method in my tool (list unique values via pivot, build a mapping table, TRIM and case functions; fuzzy matching only with human review).
4. Enforce the audit trail: work on a copy, and keep a change log — provide its 4-column template.
5. Define "clean enough for THIS analysis": 3-5 acceptance checks to run at the end, with expected results.
6. List what NOT to bother fixing for this analysis, so the job stays finite.
</task>

<output_format>
The ranked checklist (numbered, with tool-specific how-to), the change-log template, the acceptance checks, then the not-worth-it list.
</output_format>

Rules:
- Tailor every check to columns I actually described; anything else becomes [NEEDED: sample values for X] rather than generic advice.
- Never suggest silently deleting rows: every removal is filtered, counted and logged.
- Plain Australian English.

Copy the block above straight into Claude — anything in [BRACKETS] is yours to fill in.