Daniel Koć daniel at xn--ko-wla.pl
Tue Aug 7 11:56:55 UTC 2018

```A continents discussion spin-off:

W dniu 07.08.2018 o 11:31, Christoph Hormann pisze:
> A word regarding tolerance of coordinates and the implication that they
> should be or have to be within the tolerance of measuring devices - i
> don't think this is or should be the case.  The point of verifiability
> in OSM is not a tolerance threshold, it is if multiple independent
> determinations converge to a single data point.  Standard deviation of
> that might be 50m or it might be 500km.  If you ask a thousand mappers
> to position a place node for Africa and 90 percent of them place it in
> or around the Central African Republic you have a verifiable mapping
> IMO.

The placing of points is always inaccurate this way or another, just as
all the approximations. Well, for flagpoles that might be good enough
from human point of view, but most of the objects have easily visible
shape and size that one can measure (with GPS or with a plain measure
tape). Building, country, pitch, bench...

For example nobody would say that a city is a point - it only makes
sense to see it like that when we see on the large scale and we can
generalize it a lot. But simply saying that it is a point in medium (and
even more in small) scale is much more inaccurate than even rough hand
drawn area. It just takes away some information. Well - this is the
whole point of generalization, but generalization is always related to
scale and visualization. From the database point of view there is no
such thing as "too much data", it may contain all the details in best case.

Another problem with point placing - there are multiple algorithms for
placing them within a shape. If the shape is very simple, like circle or
rectangle, it's trivial, but for anything more complicated there can be
multiple positions to place it.

When people put a node on a small scale, people do the same as machines
- first they take the shape they know (!) and then they try to find
centroid. How much more verifiable is this? For me it's less - first you
take some spacial object and then you either simply draw it, or you just
pretend you didn't know it and put a point, which location is based on
this shape, of course. If you don't consider the (approximate) shape,
how do you know where is the centre of it? Well - you don't. So the
shape is always there, but with points it's just hidden.

With small entities points might be nice approximation on big scale, but
for continents or anything like that there is not such scale that you
don't see it as an area. The problem of how much continents there is and
how are they named, is just parallel - it's equally valid problem for
points and areas.

--
"My method is uncertain/ It's a mess but it's working" [F. Apple]

```