Wiki quality / stable version update

[Erik Moeller]

For those not on internal, here's the mail which explains the
parameters of the implementation that is currently underway.

---------- Forwarded message ----------
From: Erik Moeller <erik-***@public.gmane.org>
Date: Feb 17, 2007 7:04 AM
Subject: Wiki quality / stable version update


Dear all -

after a couple of failed attempts, Philipp Birken and I have tried to
make sure that the stable version project is firmly on track. For the
members of the Advisory Board: this project is meant to ensure that we
can distinguish between different versions of an article: those
reviewed for vandalism, those reviewed for quality, and those not
reviewed at all.

We have now reached an agreement with Joerg Baach, an experienced
freelance developer, to implement the project for us. We have met
Joerg in person, are confident in his abilities, and have given him
reasonably precise instructions how the first implementation should
operate.

Under the specs we agreed upon, Philipp Birken from Wikimedia Germany
will be Joerg's point of contact. We have also set up a mailing list
for a group of invited people to discuss questions and issues with the
implementation as they arise (not so much to debate, once more, the
entire proposal that has been agreed upon).

For this purpose, I'm happy to hear suggestions of people who would be
useful contributors to this list. These should be individuals who have
made notable contributions to the quality discussion in the past.
Members of the Advisory Board who are interested are also invited to
contact me, but please do read the full proposal below.

After implementation is complete, I intend to open up the list, and
make it the primary forum for brainstorming on quality improvement
initiatives.

Joerg was officially authorized to begin work last Wednesday. We hope to
have something presentable within 4-6 weeks, to be installed on
test.wikipedia.org first, and then tested on de.wikipedia.org once
Brion gives the technical go-ahead and the main kinks are worked out.

Once implementation is making visible progress, I will also discuss
separately with ComCom about the best way to communicate it to the
community and the press. For now, please do keep this reasonably
confidential, as there's always a risk that the project will fail or
be delayed once more for unforeseen reasons (and posting to a public
mailing list almost equates to posting to the New York Times these
days).

== The implementation ==

The specs to implement are a variation of the proposal by Philipp and
other German Wikipedians, found at:
http://de.wikipedia.org/wiki/Wikipedia:Gesichtete_Versionen
http://de.wikipedia.org/wiki/Wikipedia:Gepr%C3%BCfte_Versionen

The changes are aimed mostly at making the feature more flexibly
configurable, and streamlining some of the UI tasks.

The feature is to be implemented as an extension, if at all possible.
It must be GPL-licensed and will be committed to the Wikimedia
Subversion server.

The wiki will allow the configuration of "revision tags", which can
then be associated with any revision of an article. Tags are organized
in tag array, where each array represents a set of tags that describe
a similar attribute, e.g., accuracy-related tags would be organized in
one array, while those used for flagging materials for export might be
organized in another. This is to ensure that "levels" of quality
(unvandalized -> reviewed for accuracy -> featured article etc.) can
be represented correctly.

Each tag has certain attributes:
- tag comments: Should the tag support a comment field which explains
why the revision was tagged in a certain way?
- permissions: Which user groups have permission to set the tag?
- stylesheet and UI: not requiring explicit configuration, each tag
should have a stylesheet class and MediaWiki: namespace message(s)
associated with it, to allow for differences in visual and textual
representation
- implicit tagging: should this tag be implicitly set by any user
within the associated group when editing? (this will be used for the
"non-vandalized" tag)

A generic change is required to allow for automatic membership in a
user group when a user has more than X edits and is older than Y days.

There should be a configuration option which associates a namespace with
- a tag-array
- a minimum level in that array.

This option determines that, by default, the revision shown from this
namespace will be the one from that array which is also >= the minimum
level. So, for instance, one could determine that pages in the article
namespace need to be at least checked for vandalism, or at least
reviewed for accuracy.

As a high priority wishlist item, these view options should also be
applicable on a per-page basis. The existing protection UI should be
expanded for this purpose. Implementation (and schedule) of this item
will depend on overall implementation progress.

Whichever view one ends up with, we expect that the top of the page
indicates this, and allows you to switch & get diffs to other views.

Because MediaWiki currently does not show templates and revisions in
time synchronization, this behavior has to be fixed for old revisions.
When one has expressed a preference for a revision with a specific
tag/level (e.g. "unvandalized") AND this is the most recent revision,
it will be shown together with the most recent equally tagged
templates if they exist, otherwise with the most recent ones.

Example: On a page like the Main Page, which includes many templates,
one would typically want to have an unvandalized view of the entire
construction. The Main Page itself rarely changes, but because the
most recent revision is flagged as "unvandalized", it would be
synchronized with templates for which this is also true. When viewing
an older version, however, the templates would be shown as they were
at that date&time.

It is crucial that queries for revision lookup are highly performant;
we should aim for a performance impact of less than 10% on uncached
pageviews with a revision tag preference. Needless to say, the feature
needs to interact correctly with Squid proxy caching.

Tagging revisions should be possible from three places:
- when editing (with the help of a collapsible diff)
- when looking at diffs
- when looking at revisions without any prior or later tags.

A tag can only be set with reference to a diff to the last version
that has the same tag. The Special:Recentchanges tool should be
customizable to have such a diff link to the last version with a given
tag. It is desirable that this view also includes an icon that
indicates the state of the logged revision (derived from the tag
stylesheets).

Wishlist items for the future include things like mass vandalism
review and advanced queries. I also have some ideas for phase II of
the project, which I would love to see implemented before the tool is
switched on on the English Wikipedia, but this will wait until phase I
and any adjustments needed for its operations are successfully
completed.
--
Peace & Love,
Erik

"An old, rigid civilization is reluctantly dying. Something new, open,
free and exciting is waking up." -- Ming the Mechanic