Where address autofill actually works: OpenStreetMap coverage across 16 Texas neighbourhoods
Address coverage in OpenStreetMap follows jurisdiction, not density. Measuring 135,347 buildings across 16 Texas neighbourhoods, coverage ranged from 0.03 address points per building in Round Rock to 0.90 in Frisco. Houston's downtown scored 0.08, worse than rural Hill Country. Whether a door-to-door app can fill in a house number depends on whether that city ever imported addresses.
16
Neighbourhoods
135,347
Buildings counted
50,139
Address points
75.3%
Roads named
What we did
- 1Read the same Protomaps basemap extract KnockMode ships to the device, not a third-party geocoder or a general OpenStreetMap query. The point is to measure what a rep actually has on their phone.
- 2For each neighbourhood, every vector tile at zoom 15 covering roughly a 4km box was decoded and its features counted: building polygons in the buildings layer, points tagged kind=address in the same layer, and road features with and without a name.
- 3354 tiles across 16 neighbourhoods, chosen to span urban cores, dense and standard suburbs, small towns and rural areas within one state.
- 4Reproducible with scripts/measure-address-coverage.mjs in the KnockMode repository. Counts are of map features, not of houses that exist — a building missing from the map is missing from these totals too.
The measurements
Sorted by address coverage. Every figure is a count of features in the map data, reproducible from the script above.
| Area | Type | Buildings | Address points | Per building | Roads named |
|---|---|---|---|---|---|
| Frisco | Suburban | 7,251 | 6,539 | 0.90 | 74% |
| Austin — downtown | Urban core | 22,201 | 16,833 | 0.76 | 73% |
| Austin — Mueller | Dense suburban | 8,776 | 6,399 | 0.73 | 72% |
| Pflugerville | Suburban | 13,717 | 8,726 | 0.64 | 75% |
| Dallas — downtown | Urban core | 6,760 | 3,507 | 0.52 | 73% |
| San Antonio — downtown | Urban core | 7,746 | 1,595 | 0.21 | 78% |
| Sugar Land | Suburban | 2,896 | 518 | 0.18 | 76% |
| Katy | Suburban | 7,668 | 1,299 | 0.17 | 74% |
| New Braunfels | Small town | 8,068 | 1,387 | 0.17 | 75% |
| Fredericksburg | Rural town | 1,073 | 171 | 0.16 | 83.5% |
| El Paso — east | Suburban | 605 | 79 | 0.13 | 84% |
| Hill Country (Blanco Co.) | Rural | 470 | 54 | 0.12 | 73% |
| Lubbock | Small city | 12,678 | 1,198 | 0.10 | 71% |
| Houston — downtown | Urban core | 12,706 | 1,069 | 0.08 | 76% |
| Plano | Suburban | 9,758 | 419 | 0.04 | 74% |
| Round Rock | Suburban | 12,974 | 346 | 0.03 | 74% |
What it says
Coverage follows city limits, not density
The expected result was a clean gradient: dense urban cores well covered, suburbs thin, rural areas empty. That is not what the data shows.
Frisco, a Dallas suburb, has 0.90 address points per building — the best coverage in the sample. Plano, adjacent to it, has 0.04. Round Rock, a comparable Austin suburb, has 0.03 while Pflugerville twenty minutes away has 0.64.
Neighbouring suburbs of similar age, density and income differ by a factor of twenty. Density cannot explain that. What explains it is whether the city or county ever imported an address dataset into OpenStreetMap, which is a decision made in a municipal office rather than a property of the neighbourhood.
Two urban cores at opposite ends
Austin's downtown reads 0.76 and Houston's reads 0.08 — a gap of nearly ten times between the two largest urban cores in the sample. Houston's downtown has worse address coverage than rural Blanco County.
If density drove coverage, this comparison would be impossible. Both are dense, both are heavily mapped for buildings — Houston contributed 12,706 building polygons — and only one has the addresses attached to them.
The street half is reliable everywhere
Named roads ran between 71% and 83.5% of road features in every single area, urban and rural alike, averaging 75.3% across 19,042 roads.
That matters because composing an address needs both halves: a house number and a street name. The street half is dependable more or less everywhere. The house-number half is the variable one, and it is the one that decides whether autofill saves a rep any typing.
What this means if you knock doors
Ten of the sixteen areas sit below 0.2 address points per building, which in practice means typing the house number and letting the map supply the street. Five sit at or above 0.5, where a full address line is usually available.
You cannot predict which you are in from how the neighbourhood looks. The only reliable way to find out is to open a map in the area and see whether numbers appear.
Why we published a number that is bad for us
An overall figure of 0.37 address points per building across the sample is not a flattering statistic for a product whose map fills in addresses. The alternative was to keep quoting a cleaner urban-versus-suburban story that this measurement does not support.
It also corrects our own earlier framing. KnockMode previously described coverage as good downtown and thin in the suburbs, which the wider sample shows is the wrong axis.
Common questions
- Does address autofill work everywhere?
- No, and the pattern is not what most people expect. Across 16 Texas neighbourhoods, coverage ranged from 0.03 to 0.90 address points per building, and neighbouring suburbs differed by a factor of twenty. It depends on whether that jurisdiction imported addresses into OpenStreetMap.
- Is coverage better in cities than in the suburbs?
- Not reliably. Houston's downtown scored 0.08 in this sample, worse than rural Blanco County at 0.115, while the suburb of Frisco scored 0.90. Density is the wrong axis.
- Why does coverage vary between neighbouring towns?
- Because address data in OpenStreetMap largely arrives through bulk imports done at city or county level. One municipality does the import and its neighbour does not, and the map reflects that decision for years afterwards.
- What about street names?
- Far more consistent. Between 71% and 83.5% of road features carried a name in every area measured, averaging 75.3%. The street half of an address is dependable; the house number is the variable half.
- Does this cover states other than Texas?
- Not yet. KnockMode's map extract is Texas today, and this study measures that extract. The method applies to any region and the script is published, so it can be re-run as coverage expands.
- Can I reproduce this?
- Yes. scripts/measure-address-coverage.mjs in the KnockMode repository decodes the vector tiles and produces these counts. It uses the pmtiles and pbf packages and takes a PMTiles archive as input.