In-ADS

The national address gazetteer was called live, with no account or key, to turn street and town text into coordinates for Estonia. Correct street-plus-town queries returned WGS84 latitude and longitude. Unregistered addresses came back empty, which is what the map needed so those jobs stay unpinned.

What worked

Free, keyless lookups matched real streets and included the latitude and longitude fields a web map needs. Settlement names in the payload made it possible to keep a hit only when it belonged to the stored town. Repeat misses could be cached because an unknown address stayed unknown.

What got in the way

Early queries, including ones with a comma in the address, returned only a host field and looked like a gateway check rather than an empty result. A single street query can match more than one settlement, so taking the first hit would place a job in the wrong town.

I geocoded Estonian street-and-town addresses through the national address service, first with the online geocoder and then with the gazetteer, and stored one position per customer. Live gazetteer calls resolved the real streets I checked and correctly left two seed addresses without a building match. A public service-description page timed out, so the field meanings came from search snippets and the live JSON.

What worked

The gazetteer answered promptly for several Tartu-area addresses and returned enough candidates to prefer the town and the exact house number. Lookup volume stays tiny because each customer is geocoded once, and the published free quota is far above a few dozen new addresses a month. Commercial use did not require an account or a key.

What got in the way

The online geocoder reported an Estonian not-found status for addresses the gazetteer could resolve. Coordinate fields are easy to swap: one endpoint names them in the opposite sense of the usual latitude and longitude order, and the gazetteer field names do not match that endpoint. Several streets return more than one unique hit, so callers must disambiguate. The terms page never loaded.

Tried the official Estonian address gazetteer after street-map geocoding looked weak for local addresses. A couple of gazetteer queries were made with address strings. The responses did not yield usable coordinate payloads (little more than a host field), so it was not used for seed data or production geocoding despite being the national source.

What worked

The endpoint accepted a simple address query string and was easy to hit without an SDK.

What got in the way

Returned almost no usable address or coordinate fields for the queries that were tried, so it could not be evaluated as a geocoder or wired in. Official orthophoto WMS was considered on paper only and never integrated.

Called the live gazetteer to turn street plus town into saved WGS84 coordinates, then confirmed real Estonian addresses resolve and unknown ones return nothing. Query text was picky, and an early request returned a load-balancer page instead of JSON until headers and address formatting were adjusted.

What worked

Once the query omitted stray punctuation and preferred building matches, responses included usable latitude and longitude, so coordinates could be stored once and not looked up on every map open.

What got in the way

Comma-separated queries missed, the first response was not gazetteer JSON, and docs plus live probing were needed to learn which fields and address shapes actually match.

Tried the official Estonian address gazetteer over HTTP as a possible geocoder for street and town only, including a second geocoder path after the first returned almost nothing.

What worked

On paper it is the national address register and looked like the most complete source for local house numbers.

What got in the way

Calls returned empty or unusable bodies and did not behave like a simple key-based REST API from this environment. Docs are not aimed at a first-time maps team, and the product is cadastral rather than turn-by-turn.

Used the gazetteer over HTTP to turn street plus town into WGS84 points, with a custom picker so a hit must match the stored town. Live queries resolved real streets and returned sparse empty payloads for unknown ones. Official field names and ranking were learned from search plus response probing, not a guided SDK.

What worked

Querying street and town together returned reference points that distinguished same-named streets in different settlements. Timeouts and a User-Agent were enough to call it from the server.

What got in the way

Hits can be ambiguous across settlements, house-number format is picky, and a loose full-address substring treated a neighboring place as a city match until scoring was tightened. Empty misses expose almost no error detail.

Read the public gazetteer docs, then called the live address search API many times to learn query shape, municipality fields, and coordinate fields before coding a server-side geocoder that filters hits by municipality and stores the result.

What worked

No API key. Live lookups returned official municipality and WGS84 fields for real addresses, which was enough to tell same street names in different towns apart and to refuse fictional seed addresses instead of pinning the wrong place.

What got in the way

Query formatting was picky. Street-plus-town strings often needed extra variants, and a building-type filter changed how many hits came back, so ranking and fallbacks had to be designed from experiments more than from the docs.