New Naurt Feature: Accuracy of Doors and Parking Spots

We’ve received lots of feedback from the initial release of the final destination API. By far the most common feature request was an accuracy metric. Many customers noticed that sometimes the quality of Naurt data was variable. This gets right to the heart of an important decision that has to be made when making a geocoder: how good does data have to be to be included?

Unfortunately, the quality of data across the world is variable. In an ideal world, we’d like every address to have our most accurate parking spots and building entrances available, but this just isn’t feasible. Sometimes addresses are available, but without parking spots or building entrances. This leaves us with a decision: to include or not include this data? If we chose to exclude it, then we won’t have as complete coverage and our users would still need to fall back on some other geocoder. However, if we include them, sometimes users wouldn’t get the accuracy expected from Naurt could cause confusion.

The feedback from developers was pretty clear: including this data is desirable, but only if it comes with a metric for how accurate it is.

When it comes to accuracy, there’s actually two different things which we could be sensibly talking about. There’s search accuracy which is whether the correct record was found. There’s also data accuracy which is whether the data itself is accurate. For example, imagine you’re looking up troubleshooting instructions for your car on the internet. If instructions for a different car model show up, then the search accuracy is low, if the troubleshooting instructions for your car are wrong, then the data accuracy is low.

In this update, we are addressing the data accuracy - these accuracy levels refer to how accurate the doors and building entrances are for the address that was found. This says nothing about how accurate the search was.

We’re pleased to announce that we have added this feature! There are three possible accuracy levels

• high - Entrance(s) is guaranteed to be on building perimeter and known to high accuracy
• medium - Entrance(s) detected near building perimeter and position adjusted accordingly
• low - Entrance detection of low quality with possible fallback to standard address based geocode

In other words, “low” accuracy means that Naurt is performing as well as any other geocoder.

These will appear in the properties of the GeoJSON, for example

1{
2  "features": [
3    {
4      "geometry": {
5        "coordinates": [
6          // ...
7        ],
8        "type": "MultiPoint"
9      },
10      "properties": {
11        "accuracy": {
12          "quality": "high"
13        },
14        "naurt_type": "naurt_door"
15      },
16      "type": "Feature"
17    },
18    {
19      "geometry": {
20        "coordinates": [
21          // ...
22        ],
23        "type": "Polygon"
24      },
25      "properties": {
26        "accuracy": {
27          "quality": "high"
28        },
29        "minimum_parking_to_door_distance": 7.62,
30        "naurt_type": "naurt_parking"
31      },
32      "type": "Feature"
33    }
34  ],
35  "type": "FeatureCollection"
36}