I once had a UPS driver deliver a package to me where my entire name and address were so badly mangled that I had no idea how he did it. It was almost random, looked like a base64 encoding, but wasn't. But he got it in one try.
It was about the same size as a few other recent deliveries I'd had, so maybe that explained it. But I decided it was UPS ESP.
No idea if relevant in your situation, but postal companies are crazy good at mapping indicators to physical addresses.
F.ex. high-throughput OCR in the 1960s! [1]
To look at it another way, they were one of the first companies with access to big data-scale training sets (all mail flow!), albeit due to electronic storage limitations, persisted in the form of physical mail.
When you're running at a volume where a < 1% error rate means opening a new floor of manual sorters, you get pretty good at things. And it's a perfect environment for continuous improvement (customer abstracted from actual implementation, well specified process input and output interfaces, fallback plans for exception handling, and internal authority to experiment).
Perhaps the UPS logistics backend was fed a photo of the recipient address, and did a likelihood analysis of what the text in the image could potentially be OCRed to, given the places the sender usually sends stuff, the mass+volume of the box, the content-type customs declaration (if any), etc.; and it ranked your address as the first one to try.
Is that within the current state of the art—metadata-assisted OCR like that?
Who knows. I recently ordered a 12 foot long rug. It got delivered by fedex to a lady across town. Neither the address nor the lady had anything to do with me. Wasn’t like a one letter off thing.
Called fedex and they were like that’s weird. Let me call the driver. Like 4 days later the driver went to the ladies house, got it back, and brought it to me.
Postal magic is weird. I feel like we all remember the examples when something went awry, but rarely stop to think about the times it surprisingly goes right.
I had an Amazon package divert across the country, before falling into a black hole in a Milwaukee post office. After waiting a few days, Amazon CS shipped a replacement.
A few weeks later, after the replacement arrived, the original package (looking like it had been dragged behind a truck) surprisingly showed up at my address.
I think most postal routing does something akin to (initial read) -> (write machine readable code) -> (internal mapping of code to location) -> (physical routing).
And if it falls off the fast-path, remote human transcription is used at the origination tagging step.
Expect even with error-tolerant encoding, sometimes the codes themselves get corrupted (smudged, misread, etc). In which case mail ends up at (corrupted code location) instead of expected. E.g. one bit-flip, which may or may not be physically adjacent.
Where I am it seems like online shops already have an API to work with the post (surely UPS/FedEx too?). They would say using the API "I would like to register a package addressed to [netsharc's address]", and the endpoint would probably do that and respond with a barcode that the shop would print and stick to the package. Then the postman would just scan the barcode when picking up the package, take it to the sorting office where there'd be more barcode scanning and automatic sorting based on the ZIP code stored in the database entry identified by the barcode...
So the only reading of the actual typed address would be when the mailman double-checks to confirm he's delivering the correct packages for my address.
It was about the same size as a few other recent deliveries I'd had, so maybe that explained it. But I decided it was UPS ESP.