Merge pull request #41 from pelias/describe-notable-response-features

Describe notable response features

response.md

@@ -19,7 +19,7 @@ The top-level structure to every response looks like this:
 
 The `features` property of the result is where you will find the list of results that best matched your input parameters.
 
-Each item in this list will contain all the information needed to identify it in human-readable format in the `properties` block, as well as computer friendly coordinates in the `geometry` property. Note the `label` property, which is a human-friendly representation of the place, ready to be displayed to an end-user.
+Each item in this list will contain all the information needed to identify it in human-readable format in the `properties` block, as well as computer friendly coordinates in the `geometry` property.
 
 ```json
 {
@@ -53,6 +53,23 @@ Each item in this list will contain all the information needed to identify it in
 }
 ```
 
+## Notable features
+
+### GID
+This is a "global id" that can be used to reference a result with the [/place](/place.md) endpoint. It consists of an identifier for the dataset, a layer, and finally an `id` for the individual record. This `id` corresponds to stable ids from datasets wherever possible (such as the ID of an OpenStreetMap Node or Way), but not all datasets have them.
+
+### Label
+The `label` is a human-friendly representation of the place, ready to be displayed to an end user.  The label field attempts to use a format that is appropriate for the region the result is in, although Mapzen Search only supports a few countries at the moment.
+
+### Confidence
+The confidence score is an estimation of how accurately this result matches the query.
+
+For the `/reverse` endpoint, the confidence score is determined solely by its distance from the coordinate specified. Closer results get a higher score.
+
+For the `/search` endpoint, it primarily takes into account how well properties in the result match what was expected from parsing the input text. For example, if the input text looks like an address, but the house number of the result doesn't match the house number that was parsed from the input text, the confidence score will be lower.
+
+Additionally, the confidence score can optionally be biased relative to other results, just like test scores in a classroom might be graded on a curve. This takes into account both the property matches described above and the distance between results. This relative scoring is enabled on Mapzen Search, but can be disabled when hosting your own Pelias instance.
+
 ## Result count
 
 By default, Mapzen Search results 10 places, unless otherwise specified. If you want a different number of results, set the `size` parameter to the desired number. This example shows returning only the first result.