# From source data to a checked report

Allow approximately 20–30 minutes. All companies and transactions are fictional. No CRM or email connection is needed. Attach sales-demo.csv to a chat, or open it as text and paste its contents. Use a calculator or spreadsheet for an independent check. A text answer is enough; PDF export is optional.

## The task and the rules

A manager asks whether sales value increased in the second week of October 2026. For this exercise only, sales value means the sum of posted invoices and adjustments in PLN. This is a learning definition, not accounting advice.

- Compare 1–7 October with 8–14 October 2026, inclusive.
- Count identical repeated document IDs once. Conflicting duplicates need clarification.
- Exclude draft proposals.
- Report EUR separately; do not convert it or add it to PLN.
- Apply the negative adjustment in the period of its date.
- Percentage change = (new value − old value) / old value × 100%. If the old value is zero, a normal growth percentage is undefined.

## 1. Try a vague request

Ask: “Summarise sales from this file.” Save the answer and identify unstated assumptions. The answer is not necessarily wrong: the model may ask questions or identify the issues itself.

## 2. Give a clear request

Paste this request together with the rules above:

> Prepare a report for the sales manager using the attached CSV and the rules below. Include a table comparing the two periods, the difference in PLN, percentage change, exclusions with document IDs, and up to three sentences of commentary. Choose among available calculation tools; show the formula and included document IDs for each total. Do not invent transactions or causes absent from the data. Ask if necessary information is missing. Do not publish or send anything.

This specifies the output, source and counting rules. Listing included documents makes the totals auditable.

## 3. Check independently

Calculate each total with a calculator or spreadsheet. Check the duplicate ID, the draft, the EUR row, the sign of the adjustment and the percentage denominator. For each claimed cause, find evidence in the file.

## 4. Find the mistakes

This deliberately incorrect training example is authored for the exercise, not an observed output from any particular model:

> Sales rose from PLN 3000 to PLN 6500, or 116.67%. The marketing campaign performed very well. Increase its budget by 20%.

Find errors in the selected inputs and the interpretation. Explain each using a row or missing information.

## 5. Compare and transfer

Open ANSWER-AND-RUBRIC.md. Save your report and a sentence describing one mistake you can now spot. Appearance alone is not evidence of accuracy.

For an independent second trial, change FV-006 from 600 to 800 PLN. Predict which results change, then request an update without copying the full prompt. If you cannot edit CSV, paste a replacement table and clearly label it as replacing the earlier data.

Before using company data, agree the metric definition, allowed sources, file freshness and reviewer. Save the rules and checks as a draft procedure. Consider a skill and schedule only after repeatable, checked trials.