[OSM-talk] Quality assurance strategy?
frederik at remote.org
Thu Jul 31 12:46:01 BST 2008
> Some time ago there was discussion about roads which really
> do not have names vs. roads which have names but they are not yet
A problem we'll see in many other places as well - e.g. a junction
that has no turn restrictions vs. a junction that just doesn't have
them entered yet. Which is the harder case since the majority of
junctions don't have turn restrictions (whereas the majority of roads
do have names).
> is also a separate OpenStreetBugs service for informing on any
> errors on the
> map. Tagwatch is another tool that can be used for QA. However, I
> think that at
> the moment it is very hard to find out, let's say, all the features
> in one city
> that needs verification.
Well, the ITO tools at least offer you a graphical representation of
the "age" of mapping in a user-defined area so you can check whether
it's fresh or not. Of course it would also be possible to generate
lists of objects coming up for "revalidation" - restaurants once a
year, roads once every 5 years, and pot joints once a week.
> Any thoughts about how to make it easier? Should
> there be an annexed quolity control report that is linked to OSM
> feautures by
> feature id? Or could each tag has an quality assurance tag like
> True='I know
> this information is OK', False='This information is uncertain'?
I would like to have that information separate from the rest because
it is not data about the real world but data about our data. If I
were to tag a way with the information "this way needs verification
becaus it has not been edted in the last 2 years" then I would edit
the way by doing so...
What you're speaking of is a huge amount of work if one wants to do
it properly. I have no doubts that we'll manage to map the world
once, but will we be able to map it again every 5 years, essentially,
to verify? I don't know.
I have been thinking about a system where you do not wait for people
to take responsibility and say "this way is correct as of <date>,
signed <name>", instead turn this around: The quality of an area is
not defined by the number of quality assertions, but by the absence
of error reports!
Suppose you have a really easy error reporting mechanism right on our
central map page, a big fat "report error here" button. Suppose that
you cleverly analyse log files (and those of mirrors/caches
obviously), so you know how many people have looked at a certain area
within the last month (and perhaps even how long they have looked).
Then you can say: "150 people have looked at the map you're seeing
without reporting an error". Of course only a fraction of viewers
will notice an error and only a fraction of those will report it, but
if the numbers are high enough, you should get reliable statistical
> OSM quality supervisors
Frederik Ramm ## eMail frederik at remote.org ## N49°00'09" E008°23'33"
More information about the talk