A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "Shipping: A Bundle For The Shape Of The Answer," Kin Lane, November 12, 2026, https://apievangelist.com/2026/11/12/shipping-a-bundle-for-the-shape-of-the-answer/
Marisol read the post on her lunch break, at her desk in the receiving office, where she works as a systems analyst. She had opened it because of the word shipping. She stayed because of the second paragraph, where the author says he has to walk back a claim from his earlier post.
That earlier claim sounded a lot like something she had once said in a meeting. A pattern that demands eighteen digits, she had told people, means a malformed pallet ID can't get into the system. Here was someone saying that the length holds and the check digit doesn't. The pattern ^[0-9]{18}$ gets you eighteen digits and nothing more. A model guessing at the last digit would, the post says, produce an invalid SSCC of the right length about nine times out of ten.
She felt a little defensive. Her own intake validation was built on exactly that pattern, and it had never once complained. She wondered whether that meant it was working or only that it had never been asked a hard question. She didn't like not knowing.
She read on with her arms crossed. The part she liked was the status list. The post says a decoder cannot produce a tenth status, and the fixture where Delayed got refused backed that up. That matched her experience of closed lists. Normalizing carrier language into nine values was most of her problem too, and she didn't need a bundle to tell her so.
Then she hit the part about severity, and she disagreed. The author makes a missing reason on an Exception an error but only a warning on a failed attempt, because a failed attempt "mostly resolves itself on the next try." Marisol had seen failed attempts at a dock that never resolved. The consignee had moved, and nobody noticed for a week. To her, a warning that flags something for review was a quiet way to lose a pallet. The author did call these his own judgment calls and easy to change, which softened her a bit, but she still thought he had picked wrong for her kind of freight.
The passage that stayed with her was the one about the second fixture. "A package marked delivered when it was not is about the worst thing a tracking system can say." The rules caught the invented quote. But if the email says delivery was attempted and a model answers Delivered, every check passes. She read the last sentence of that paragraph twice, then closed the laptop feeling mildly annoyed, mostly at herself.
She didn't plan to adopt anything. She had no model in her pipeline and wasn't about to add one. But the post had put a question in her head that she couldn't unask.
After lunch she exported the pallet IDs from last month's receiving log, about two hundred rows. She wrote a short script that ran the mod-10 weighting on the last digit of each one, and she ran it before anyone could talk her out of it. Then she put the output in a file and left it for the afternoon, because she wanted to read the results with a clear head, whatever they turned out to be.