User:Chriscf/AfD evaluation

Evaluating an AfD debate is something of a black art. In some cases, there is the inevitable treading on toes and breaking of eggs. Since AfD is not a straight vote, but a discussion, it is not worth counting merely numbers. I have seen a few too many articles presented to WP:DRV with the appeal being "debate closed at 10-8, should have been no consensus". This is how I weigh them up when asked whether a given debate reached an appropriate outcome. Since these scores are not directly tied to individual comments, there is no "magic ratio". The rules for evaluating an expected result follow at the end.

Weighing up arguments

 * A strong argument rooted in policy scores 1.
 * A strong argument rooted in policy independently stated by a second person scores 1½.
 * A weak argument rooted in proposed policies, or contested guidelines, or essays, scores ½.
 * An argument based on opinion or fallacious reasoning (including, but not limited to, "I like it", "popular", "more important than", "does no harm", etc.), which doesn't address the article or any relevant policy, scores 0.
 * An attempt by an article creator, or someone heavily involved in the subject to garner support by posting to external websites or blogs scores -1.

Measuring support

 * People following up a strong argument ("X, per Y") score the binary logarithm of the number of third parties supporting (2 scores 1 point, 4 scores 2, 8 scores 3, etc.). This is logarithmic because the weight of each additional voice becomes less.
 * People following up a weak argument score the binary logarithm the number of third parties supporting, halved.

The following is an approximate indication of the number of supporters required for a recommendation to score a given number of points. Note that with weak arguments, a pile-on will not score highly. Given that the score for weak arguments is based on halving the logarithm, you'd think that te bottom row would simply be squares of the top. The reason this does not happen is that the entries are the ceiling values of the antilogarithms. For example, the values given under 2¾ are $$\lceil 2^{2\frac{3}{4}} \rceil$$ and $$\lceil 2^{5\frac{1}{2}} \rceil$$ respectively.

Interpretation
For convenience, points are only ever counted in quarters. Binary logarithms are rounded down to the previous full quarter, e.g. lg 5 ≈ 2.32, which is rounded down to 2¼. Again, when halving the numbers, we round down to full quarters, so were this a weak argument, 1⅛ is rounded down to 1.

A score of zero for keeping the article suggests that the result should be delete, even with any number of recommendations to the otherwise not preventing the closing administrator from exercising their discretion.

A low score on both sides suggests that the result should be the winner. Remember, a debate scoring 3-2 on this scale gets whichever argument scores 3, since it does not require a supermajority.

A high score on both sides suggests a no consensus close.

The one and only exception is that an unmitigated failure on any of the non-negotiable core policies, or an uncontested valid argument based on them, suggests a delete by default, regardless of any support for the article.

Applicability
This is my personal evaluation criteria. They are not binding upon anyone else, least of all the administrators that have to carry out the deletions.

Rationale
Some arguments are better than others, so a strong argument rates more highly, double that of a weak argument, triple if is it proposed by two people independently. Arguments should be relevant to the policies and guidelines. People often question the "validity" of certain recommendations. Hence, arguments don't score if they are pointless, non-issues, straight votes, or perhaps are just wide of the mark.

Canvassing for votes is frowned upon. Canvassing for votes to save your own personal article especially so. Canvassing votes to save your own personal article outside of Wikipedia is something that cannot be tolerated. As a result, if someone has engaged in an active campaign to save or delete an article, by canvassing members at large, then a point is deducted from total score of arguments to that side. Additionally, those that appear to have responded to canvassing, be they new or established members, will not be counted in support.