30 Shipping Labels: OCR with Clipped Fields

The challenge

Turn these shipping labels into a manifest. Some PDF pages leave out fields that appear in the JSON answers, so keep track of what you can read and what's missing.

Start here

Open the SharePoint folder linked above. shipping_labels_1757230798842.pdf has 30 pages, one label per page. shipping_labels_1757231335519.json has 30 answer records. Start with page one: it has no selectable text, so it needs OCR.

Render page one at full-page size. Its tracking number is printed near the top and again under the barcode. Its bottom is clipped before the weight/date area that appears in the related image-label collection. Inspect further pages to determine which fields are visible on each, rather than assuming every reference field can be OCRed.

Try it

1. Extract company, service, tracking number, sender, and receiver without mixing the address blocks.

2. Preserve leading-zero and ZIP+4 postal codes and handle wrapped city names.

3. Compare predictions against the reference by tracking number, separating visible-field accuracy from answer-assisted completion.

4. Summarize service mix and the clean reference's noise and package-weight distributions.

What to build

Build a JSON array or flattened manifest CSV, a page-to-tracking-number map, and a validation report identifying missing visible fields. Add source_page and review_status to every result.