[dwyerm]

You are fortunate that lat/long is sufficient for your use-case.

In my experience, the problem is that it isn't as exact as you'd think, especially when your measurement passes back and forth between reality, the idealized cartesian lat/long, and node/distance understandings of road networks. At a previous company, our map provider moved the entire United States three meters to the west and we spent the next week re-adjusting endpoints that were suddenly in medians, on overpasses, etc.

tl;dr: Nothing's easy.

[LarryMade2]

Yeah had that happen on google maps API a few years back where the earth adjusted southeast or something like that... I wonder if the offset you experienced was able to be re-adjusted by calculation or was the new geometry inconsistent?

Ive realized we can't ever get exact - best we can hope for is finding something that's constant enough for a long period. So far lat/lon has done best.

[dwyerm]

In retrospect I also wonder why we didn't just do a bulk update. I think the problem was that we discovered the geocoding problem and started correcting for it before we understood where the problem was actually coming from. By the time we discovered it was a simple constant offset, we'd already damaged our own data with the one-off fixes.

Worse, the offset was right there in black-and-white in the release notes that we failed to read.

By the way, I apologize for being flippant about it. Location is just one of those things -- like time -- that too many people think is simple, but actually has devilish details that bite the unwary. You are fortunate that lat/long is sufficient and I hope you remain wary.