ACME Scanned Receipts: OCR to Transactions and Item Rows

The challenge

Turn these receipt sheets into a sales ledger. Each page can hold several receipts, so keep each purchase together with its own items and check the result against the JSON answers.

Start here

Open the SharePoint folder linked above. ScannedOCRSheet.pdf has 480 pages. transactions_2025-09-06.json contains storeConfig, generatedAt, transactionCount, and a transactions array with 1,522 records.

Render the first, a middle, and the last PDF page. The first page shows receipts but has no extractable text layer. This can happen even when a PDF library reports no ordinary embedded images, so decide from the render and text extraction together. The first page contains three receipts.

Try it

1. OCR the sheets, segment receipts, and produce transaction and item records.

2. Match extracted receipts to the JSON without pretending its numeric IDs are printed.

3. Validate item arithmetic and reproduce total revenue, units sold, and leading products.

4. Report extraction coverage and error types separately from reference-derived aggregate answers.

What to build

Build nested transactions or two linked CSV tables, a source-page matching table, and an accuracy/reconciliation report. Start with a reviewed pilot batch before running all 480 pages.