Improve position accuracy definition

[marcoscaceres]

@martinthomson wrote:

Bonus question: What is a "position accuracy"? In metrology, the area this described might be the 2d analogue of a confidence interval at a given confidence level (95%, 67%, something else). But the word "accuracy" implies something else.

We should improve the definition of that.

[reillyeon]

The Android API documentation uses 68% confidence: https://developer.android.com/reference/android/location/Location#getAccuracy()

The macOS/iOS API documentation does not specify what accuracy means other than that it is measured in meters: https://developer.apple.com/documentation/corelocation/cllocation/horizontalaccuracy?language=objc

However, I suspect that the 68% number from the Android documentation corresponds to the 1-sigma error values provided in a NMEA-0183 GST message. It seems safe to assume that any GPS-based device would provide similarly derived error information and so standardizing on 68% (or 1-sigma) for accuracy values.

[marcoscaceres]

I wonder why we have 95% in the spec? And yeah, re-reading what we have in the spec it is (to me) still a bit confusing. I wonder if we can do better there but also check what each user agent is providing and how accuracy is used in practice.

[marcoscaceres]

Ok, did some digging, yeah... we should drop the 95%. We should define them as “radius of uncertainty (horizontal)” and “vertical uncertainty interval” as provided by the platform.

Do we want to surface the accuracy confidence for horizontal and vertical as provided by the platform? Would be a static value (e.g., "0.68" for Android and null for iOS/macOS, for horizontal). There's no certainty I think on various devices... a laptop will be different from mobile device.

We could then provide some guidance to developers for how to interpret that values.