[Imports] How good can an import be?

[Andy Allan]

On Wed, Apr 6, 2011 at 6:23 PM, Paul Norman <penorman at mac.com> wrote:
>> From: Andy Allan [mailto:gravitystorm at gmail.com]
>> Subject: Re: [Imports] How good can an import be?
>>
>> On Wed, Apr 6, 2011 at 1:28 PM, Steve Singer <ssinger_pg at sympatico.ca>
>> wrote:
>>
>> > I think a big part of the problem is that import guidelines set a
>> > fairly low bar for an import and most imports don't even follow those
>> > and 95% of the time there are no real concequences for not following
>> > the import guidelines if you can 'complete' your import before anyone
>> notices.
>>
>> Sadly, very true.
>
> What do you think the options are for fixing this? I'm aware of an import
> that was done without consultation with the local community, with
> questionable tagging and bad data quality. To top it off, the import had
> technical problems which lead to some of the objects being fixed by
> xybot. Do you think the best option for imports like that is for them to be
> reverted?

I would concentrate on the pre-import stages first. There's a lot of
the instructions on the wiki that are quite vague about what is
required, and I think they could be firmed up a bit. It would be good,
in amongst the import documentation, to explain the other non-import
options for using datasets, since a lot of the time I think people get
confused between the general use of external datasets and the specific
solution of importing them.

I also think it would be interesting to "review" imports a few months
later, especially those that have gone especially badly or especially
well, so that we can all learn from our mistakes and successes. It
might also help to make people revisit their imports and deal with the
problems that they didn't notice up front.

I was pondering a review system where each import starts with 100
points, and then points are deducted for each problem they have (e.g.
not using a separate account => -10 points. No listing on the imports
catalog => -10 points. No reference to the wiki information from the
changesets/account details => -5 points. Duplicated geometries => -15
points etc). If anyone builds up a track record of poor-quality
imports then they could get some advice on how best to improve them in
future - especially giving consideration to using the data in other
ways than just uploading it.

A little part of me wants to go overboard and instantly revert any
import that doesn't follow the guidelines, but I'm not sure it's
practical or particularly effective. We should aim to work together to
improve things though, and I'd hate to thinkg we've reached a kind of
plateau where the guidelines are routinely ignored and nobody seems to
care.

The imports working group (this list, effectively) was started with
the aim of making imports better, not necessarily just tracking which
ones are going on. There hasn't been as much progress on that front as
I would have hoped a few years back during our first meetings.

Cheers,
Andy