Sorry, was't meaning to be critical. As I say, I see where your getting that data from. You have to do a very fine grained house by house check to get the full story.
I know you're not, and I agree (the response was a bit blunt, sorry
). I have just considered removing it so it just says live/not-live.
Where are you getting your postcode data from? It's slightly off for my address. I'm sure I was looking at mobile coverage maps the other day and it was spot on. Oh yeah, the mobile coverage is very handy, thanks! I didn't know there was any 5G in my neck of the woods!
The postcode data comes from maxbox, it sort of centres on the postcode itself (annoyingly, despite them passing a number and postcode to geocode), I have tried to get it more specific but I think I'll need to maintain a database to do that. The UPRN database would be great here.
The importance of accurate pinpointing has only just become a problem since I didn't have cellular coverage checks on there before.
Edit: bit more context to address matching:
Upon house number and postcode entering, the tool will geocode (turn an address into long/lat). Once that request has a result, it'll fire off all the checker queries (none of the fibre checks needs long/lat so that could probably start earlier, but the geocode could be seen as some basic validation too).
The fibre checks all require both the number and postcode and in almost all cases (excluding Openreach, I'll talk about that in a moment), the provider will return a list of addresses from a postcode search, it's then my APIs responsibility to choose the right address for the user (this is the part I have described as dumb). If the API can't make a perfect match, it'll take the first from the list and add a "looseMatch" entry (if you search with a fake number on a street, you'll see triangles next to the results, that's this logic happening). The query then takes place on that property it chooses.
The cellular checks all require a long/lat, and just return a rating from 0-4 (0 being no service, 4 being best). These queries depend on the providers and I have noticed they do lag quite far behind reality. It's good for a first glance look, if it says you have service, you probably still have service.
Regarding address matching improvements, I have been thinking of using Openreach's checker more aggressively for this. They take the number/postcode and make this address matching decision themselves. As part of the result, there's a UPRN. A few other providers (definitely not all) can query on this UPRN. It's also possible to get a perfect long/lat mapping from UPRN with a free database from OS. Instead of using Mapbox, the flow could change to use that first and take advantage of their data to geocode. It'd also improve the efficiency of the checkers that can utilise a UPRN.