Really cool and thorough work! Thanks for sharing!

My own first intuition is that it's not actually a data problem at all, and that "Google Maps has no concept of.." might simply reflect the ongoing, enshittening, transition from structured "concepts" to ML "vibes" for products like Google's.

It's not that the underlying maps data store has the addresses wrong, but there's a layer between the input field and the result generator that's statistically deciding you mean something besides what you explicitly enterred and is giving you a route to a silently "corrected" address.

We've seen that happening more and more in Search for years, silently ignoring keywords and directives without a "did you mean" callout, and it would seem natural for some product owner to be pushing an equivalent initiative in Maps. Aggregate metrics move the right way, so the company is happy, but of course the actual product experience sees a fractal failure pattern that nobody can quite address but makes results less and less reliable.

I'd love to be wrong, though, as I think this would be a terrible advent for something that can be life-critical like a popular mapping tool.

While Google is no stranger to enshittification, I think Maps (and Contacts) are two services where it's just damn hard.

In particular it was developed in the western world from a startup, so the first approach probably was minimal and adapted to local streets, without even covering the edge cases.

Going from that to mapping every single address in the world is a huge leap and the underlying data system must be an incredible mess, also creating regressions on what used to be reliable.

I don't know if they got to it, but a few years ago you couldn't have vertical stacking (e.g. floors), shops with the same address needed workarounds to have a different entry.

Then some places have multiple addresses. The article talks about street numbers making sense, but in most places in the world they don't, disappear at random points, some countries don't have street names. We could talk about it years.