On 1/28/26 3:01 AM, Abdullah DevRel of IPinfo via NANOG wrote:
The whole reason I am here is because of this. Please check your data with us. If it is incorrect, let us know, please. Active measurement by design should be better than geofeed, but if it is not due to network architecture or some reason, you need to tell us. We will investigate. ... inherently, the system is designed for evidence-based accuracy.
I appreciate you coming here to engage. If you're looking for evidence-based accuracy, I don't think you are going to beat trusting someone's geofeed. Assuming you are not using cellphone GPS data, which you haven't said you are, it is not EVEN IN THEORY possible for "active measurement" as you have described it to be better than geofeed in the case of non-mobile services provided by ISPs that take geofeeds seriously (and if they don't, why would they publish them?). Active measurement as described by you depends on having servers close enough to "triangulate" location by ping time, but: * Every microsecond of jitter amounts to approximately 100m of uncertainty. If typical jitter is 1 millisecond (which strikes me as LOW), that's 100km of uncertainty * Traffic doesn't following straight line paths to your servers--it goes from the customer to their ISP's head-end, from there to an exchange point, and that's about as close to you as it will likely get. There's no reason to believe any of those paths are even remotely straight so even the best case of triangulation will be only triangulating to the ISP's head-end Google suggests that you have approximately 1000 probe points in your probe network, and that gets you to a typical accuracy of 20-100 miles--better than the "Region" field of a geofeed in the middle of some large regions, but worse for small regions (eg, New England) and near region edges (what do you want to bet 1/3 of Indiana, Michigan, and Wisconsin IPs are classified as Chicago?) and worse than the City and (deprecated) Postal Code fields. A geofeed in an urban area that is accurate to the granularity of "zip code" will be accurate to less than 1 mile, and may be precise up to a few meters at the boundary of the ZCTA. In suburban and rural areas the geofeeds will be somewhat less precise, but so will your "active measurements" due to the longer backbone runs, lower probe density, etc. I think the message from the experts here is clear: if you want accurate data, you should be trusting the geofeeds, unless you have substantial other evidence that the traffic is not originating where the geofeed says it is, and that it's not a mistake the provider will remedy. Perhaps taking that good advice to heart will allow IPInfo to dramatically improve its accuracy, and take over the IP geolocation market! Peter p.s. You show my IP--probably all of our IPs--in Chicago. That's like 130 miles away. So what am I supposed to do now to correct your error? Perhaps you can see why this is frustrating given that we literally ALREADY TOLD YOU where we are via our geofeed, the standards-compliant method for conveying this information to you!