Wikipedia talk:Requests for comment/Metrics for measuring article reliability

What articles? In the Wikipedia? On the Web? Maybe I'm being dense, but I don't understand this at all. --  Donald Albury ( Talk )  13:28, 1 April 2006 (UTC)
 * Of course it's about Wikipedia. Maybe the proposal is now easier to understand --85.25.136.135 19:04, 1 April 2006 (UTC)
 * You want a hit counter on the article pages. My understanding is that this is a problem, as most hits are on cache servers rather than the main database. --  Donald Albury ( Talk )  19:15, 1 April 2006 (UTC)
 * Maybe the cache servers keep logs so the cached hits could be added retroactively. At least it would be interesting how big the overhead for such a solution would be. It would be nice if it could work, however, as it would be step into a more community-based direction of quality assurance opposing to recent movements to hierarchical solutions which are a bit against the initial momentum that made Wikipedia that successful. --85.178.36.219 21:56, 1 April 2006 (UTC)
 * My suggestion is to ask about this over at Village pump (technical). --  Donald Albury ( Talk )  23:34, 1 April 2006 (UTC)
 * It's now also discussed there. Additionally, there is a discussion about the more theoretical considerations of this mechanism at . --128.42.6.133 10:08, 3 April 2006 (UTC)

There could be different views of the Wikipedia built on this facility. A similar approach is used in the Debian Linux distribution. According to the metric, there could be stable and unstable versions of the pages, perhaps with steps between them, so users could decide how much actuality they want to trade for stability/reliability of the articles. This comparison also shows that approaches like this have been successfully used in reality. --84.141.191.184 13:55, 4 April 2006 (UTC)


 * Logging is minimal for performance reasons. If I remember correctly, neither the cache servers nor the Apache servers keep logs with the required detail. -- Jitse Niesen (talk) 07:48, 6 April 2006 (UTC)

A metric for stability
It occurred to me in the course of an afd discussion that a metric for stability of an article ("Thi s article is .... and unstable"; "no it isn't is is .... and stable") would be useful. i see it mentioned in passing here. How might one go about that? Stability is probably a component of reliability, and one that makes less demands on the servers than counting hits. Midgley 17:59, 13 April 2006 (UTC)

Better methods
There are better methods than a hit counter. Hit counters are rather coarse indicators, only weakly linked to an article's reliability. (And by the way, for the nay-sayers on technical grounds, it would be easy to take a random but representative sample of page hits, and extrapolate from there.)

For instance, one could use simply time since last edit. Or, one could use a "stability" metric- if too much of an article changes within a particular time period, then it is marked unstable (the major changes possibly being vandalism or a major rewriting or addition which is still not quite tweaked). Or one could work on the basis of reverts- the fewer reverts in a time period the better, or possibly whether the article is currently in a state that it is often reverted to (ie. two separate vandalism-versions would be favored less than the original version the article was reverted back to twice; edits could be seen as votes for a particular version); you could go with more elaborate versions which use trust metrics, in which a highly-rated editor "endorses" versions of articles (or possibly just sections of an article).... This has all be discussed a lot before, on Meta if nowhere else. This policy needs vastly more work. --maru  (talk)  contribs 07:09, 23 April 2006 (UTC)