ACM MegaMarket: OCR a Partial Receipt Collection and Repair Its CSV

The challenge

Build sales records from the PDF receipts, then find out how much of the full dataset they cover. You'll also fix a CSV export and use the complete JSON to explore the store's sales.

Start here

Open the SharePoint folder linked above. MegaMarketSalesReceipts_1.pdf, _2.pdf, and _3.pdf each have 500 pages; AggregateMegaMarketSales.zip packages those three PDFs. SimulationDataACMMegaMarket.pdf is a two-page generator summary/preview. transactions_2025-09-06.json is the complete structured reference, and transactions_2025-09-06.csv is another export with broken quoting inside the items column.

Render the first and last pages of each receipt PDF. The first page of each part has no extractable text layer. Many have two receipts, so neither PDF page count nor archive member count equals transaction count.

Try it

1. OCR receipt regions, recover item rows, and match them to the structured reference.

2. Establish the observed PDF coverage and explain why a missing reference match is not automatically an OCR failure.

3. Produce valid CSV from the nested data and show what is wrong with the included CSV.

4. Calculate revenue for the full dataset, units, and product rankings while keeping partial-PDF results separate.

What to build

Build an extracted receipt subset, a coverage/matching ledger, normalized transaction and item schemas, a CSV repair walkthrough, and answers for the full dataset. State clearly whether each result came from OCR or the answer export.