[OSM-talk] OSM needs a measure for completeness

[Dair Grant]

Chris Morley wrote:

>I have started a new thread with a measure for completeness in the
>title because this is an important topic for OSM. But the response
>to the recent posts quoted above, and my raising of it last July,
>has been only luke-warm.

I also think completeness is a very important idea - it's false 
to think that a map is ever going to be "complete", but I don't 
think it's a good answer to say it can't be done.

There are places in OSM where there is no data; these are 
obviously incomplete.

Equally there are places in where the data is at least as good 
as any other map (bits of London say), subject to our data model 
(no house numbers, say).


>I wonder whether this is because "completeness" is associated in
>people's minds too closely with verification. As Andy has describes
>it in the (incomplete) quote above, verification involves individual
>accountability - "I personally accept responsibility for the
>accuracy of this data".

I don't think you accept responsibility for it in the sense of 
liability in case of mistakes, but as soon you as you enter some 
data into OSM you do accept some responsibility for it.

You're responsible for ensuring that it bears some relationship 
to reality, that it didn't come from an illegal source, etc.


>As is the case for all other mapping information, an assertion of
>completeness should only imply the best endeavours of the
>contributor, not a guarantee of 100% correctness. If you have ridden
>round a housing estate systematically and collected all the required
>information, you can reasonably say the area covered is complete.
>With this understanding, completeness would become part of routine
>mapping. It would encourage a systematic approach and the collection
>of any missed information.

Absolutely. I would be very happy if there was some way I could 
give a simple badge (or a score, 1-5, where 1 is empty and 5 is 
complete), to some area to indicate how "done" I thought it was.

Both to myself as a way to keep track of what's next, but also 
so that other mappers can see just how much there is to do.


>A possible detailed approach is as follows. A completeness boundary
>would be modelled on coastline: it would enclose an area, but would
>consist of many (ordinary) ways, with the completed area on the
>right. They would be tagged with one (or possibly more) definitions
>of completeness like "major roads", "public roads", "public paths",
>which would be defined on a wiki page. Boundary ways would be moved
>or added on a day-to-day basis by anybody with local knowledge. An
>area might even have holes in it.  Somebody would provide an
>overview map showing completed areas, and its animation would
>feature in most presentations on OSM.

I was thinking of a simpler model - each OSM account gets to 
define a list of bounding circles, and a 1-5 completeness rating 
for each circle.

Circles rather than boxes, because completeness is by definition 
a fuzzy subject.

Rather than trying to exactly cover the world with areas that 
are done/not done, it'd be better to just drop some approximate 
circles to cover smallish areas.


These will all overlap and generally look a bit of a mess, but I 
think could be rendered in a way that would give you a sense of 
completeness in some area and demonstrate progress.

E.g., at this point in time the areas that are complete should 
have more priority when rendering a zoomed out view - everyone 
knows that London will have lots of little holes in it, but in 
general most of the circles in there will be "complete".

When you zoom in then the incomplete areas become more 
important, so by the time you're at town/village level you want 
to be able to see which suburbs still need work to do.


The only difficult bit is setting up the database to manage this information.

I think it would be better done as part of the OSM user 
accounts, rather than in the OSM database, since I think putting 
it in the real data encourages us to try and over-specify 
something that's always going to be ambiguous.

Periodically some software pulls all that information out and 
renders a map of it, or sends a message to any users with 
obvious contradictions (two circles share more than 75% of their 
area and their ratings are too far apart, or similar).


-dair
___________________________________________________
dair at refnum.com              http://www.refnum.com/