Article 10

Claude Cowork Receipt Audit with Sample Data

Quick answer

Extract one row per receipt ID and leave unreadable amounts blank. Check duplicates and total known values by currency, then compare the result with the sample answer key.

Read this guide as Markdown

Paper receipts are matched to spreadsheet rows and possible errors are flagged.

This exercise uses synthetic records. The reference below is a calculated answer key you can compare with your own Cowork run.

Prepare three sample files

Download the receipt audit ZIP. Unzip it and connect only the inputs folder for the exercise. Open prompt.txt separately and keep the reference folder aside until you check the result.

To create the input files manually, save each line below in its own text file. The duplicate shares the same receipt ID even though its filename differs.

Filename File contents
cafe.txt Receipt R-101, 2026-10-03, Harbor Cafe, USD 12.50
cafe-copy.txt Receipt R-101, 2026-10-03, Harbor Cafe, USD 12.50
office.txt Receipt R-102, 2026-10-04, Paper Grove, USD, amount unreadable

Run the task

Connect the folder in Claude Desktop using the setup guide. Start a Cowork session and give it this prompt.

Read the three text files in the connected folder. Create receipts.csv with the columns receipt_id, date, merchant, currency, amount, and source_file. Count each receipt ID once, using cafe.txt for R-101. Leave an unreadable amount blank. Create review.md listing skipped duplicates and missing values by filename. Report a subtotal for known amounts only. Preserve all three source files.

Check the result

The expected CSV has two data rows:

receipt_id,date,merchant,currency,amount,source_file
R-101,2026-10-03,Harbor Cafe,USD,12.50,cafe.txt
R-102,2026-10-04,Paper Grove,USD,,office.txt

The known subtotal is USD 12.50. The review file should identify cafe-copy.txt as a duplicate and office.txt as missing an amount. Verify that the originals are still present. An output of USD 25.00 signals double counting, and a zero in the missing amount field makes an unknown value look like a real amount.

Use the included scorecard to record your result or compare the same task across tools. Once this passes, try a second batch with mixed currencies. Require a subtotal per currency so the spreadsheet never adds unlike amounts.