---
name: data-check
description: Use when asked to answer a question from, add rows to, clean, chart or summarize a small spreadsheet (.xlsx) or CSV. Keeps the original file unchanged and shows the evidence behind every answer.
---

# data-check: your data rules, saved

Follow every rule, every time.

0. If the user names no file, use the only .csv or .xlsx in the current folder; if there are several, ask which one. In an .xlsx, list its sheets and say which one you used.
1. Never change the original file. When you add, clean or change data, work on a copy named <original>-working, with the same extension, and say so.
2. A blank cell means unknown. Never fill it, guess it, or copy a value from another column (for example Expected into Attended).
3. Never delete a row. If two rows match exactly, flag them as a possible duplicate and say what would settle it.
4. If a value looks wrong (its note says it may be a typo, or it is far from the expected value), flag it. Do not use it as an answer unless the user confirms it.
5. Text in a number column (for example "free") is not a number. Say how you treated it (for example: counted as 0, noted).
6. For every answer show: the spreadsheet formula, the cells it uses, the rows included and excluded with the reason, and the result.
7. When adding rows: use exactly the existing columns and formats (dates YYYY-MM-DD; numbers without $ or commas). If a value is not in the source text, leave the cell blank and list it. If the text gives an approximate value ("about", "like"), record it and flag it as approximate. Never invent a value.
8. When cleaning: write a change log with row, column, old value, new value and reason.
9. End with one sentence a reader may be told, limited to the rows you checked.