People are mad about AfD. There are a lot of issues that get brought up every time.

Here are some of the commonly noted problems (and potential solutions) I'm pulling out of my hat. Some of them will be smart and some will be not so smart. Anyone who thinks of something is encouraged to add it. Everyone reading this is encouraged to go take a look at User:JPxG/Oracle/All and familiarize themselves with the general trends in AfD history. It's also worth looking through some of the entries at User:JPxG/Oracle/Largest AfDs to get an idea of what very long discussions look like.

Rapid-fire nominations and !votes

I think it's pretty safe to say that someone commenting on one AfD every thirty seconds is unlikely to be contributing a whole lot to the discussion, or doing WP:BEFORE. This has happened before, a few times.

A problem with potential solutions is that there are some exceptions, i.e. a string of articles that are all drawn from the same source about the same subject to which the exact same considerations apply, for which "once you've seen one, you've seen them all".

Potential solutions

• AfDstats.py is already capable of quickly summoning a (mostly) complete history of someone's AfD participation. It seems to me like it'd be fairly trivial to include some functionality to display intervals between !votes. This may solve the problem on its own. jp× g 23:47, 12 November 2021 (UTC)
• Some variant of the {{ canvassed}} template could be made, and (manually or automatically) added to !votes that were made a very short time after the user's last edit. jp× g 23:47, 12 November 2021 (UTC)

Voting blocs

A few people have said that groups of editors will follow each other around to different AfDs and make the same !votes, regardless of the merits of an article. WP:CANVASSING is a guideline that applies here. Even if the first !vote is constructive and cited guidelines, "me-too" !votes made after it are arguably disruptive.

A problem with potential solutions is that a group of people voting together is often not the result of collusion. For example, there are a number of people who show up at almost every GNIS stub cleanup AfD; some tend to !vote "delete" a lot, some tend to !vote "keep" a lot, and some vary based on the availability/quality of sources. People in all three categories are likely to agree with others in the same category. Moreover, they're often likely to agree with each other. If "Jerkwater, West Tacoma" is a dogshit article with no sources, most people (including "inclusionistas") will !vote to delete, and if it is a great article with lots of sources, most people (including "deletionistas") will !vote to keep. jp× g 23:47, 12 November 2021 (UTC)

Potential solutions

• It might be possible (through great effort) to write software capable of detecting voting blocs (i.e. a number of people who always !vote the same way on the same discussions). jp× g 23:47, 12 November 2021 (UTC)

• While the above caveats definitely apply here, some sophistication could be used to winnow down results. For example, "every pair of people who !vote together on more than 10 AfDs in a month" is uninformative (this will probably include all active AfD editors). "Every pair of people who !vote together on 70% of their AfD overlap" is a little better. There are probably additional data points that could be used.

AfDs run too quickly

AfDs run for seven days. This has been the case for nearly the entire history of the process. The idea is that it produces better results when some time is allowed for consensus to form; at the same time, discussions cannot simply drag out forever, and have to be closed at some point. Furthermore, the shorter an AfD runs, the less will be open at any given time.

However, in 2007 there were 46,000 AfDs (~127 per day). In 2019, there were 18,000 AfDs (~50 per day). If current rates are much lower than historic rates, assumptions that made sense at the time may need to be re-examined. Moreover, relisting is not a very well-defined process (whether or not an AfD is closed or relisted on the seventh day seems to be largely arbitrary, and there is no guarantee that on the seventh day a close-looking discussion will not simply be closed one way or the other).

One thing I've noticed is that AfDs tend to have a lot of fierce contention in them based on (perceived or actual) urgency. If you only have a few days to do research, collect sources and make your argument for each article — and there are fifty nominations per day — it seems inevitable that some of these arguments are going to be sloppy. jp× g 03:20, 13 November 2021 (UTC)

Potential solutions

• A trial could be run for assigning some AfDs to run for longer than 7 days (perhaps 10, or 14).
• A trial could be run where some AfDs were given a guaranteed relist at the 7-day mark (unless !votes were unanimous).

AfDs are structured in a way that's not conducive to consensus

At every big AfD, you will notice a certain pattern of repetition. It will look something like this:

• Keep. Clearly notable, as shown by ref 1, 2, and 3. – Prima ( talk)

• Those aren't reliable sources. – Secunda ( talk)

• They were considered reliable at this RSN discussion. – Prima ( talk)

• Delete. Nothing showed up on Google News. – Tertia ( talk)

• There are plenty of results on other searches (and offline sources are already in the article). – Prima ( talk)

• Keep. Refs 2 and 3 are SIGCOV. – Quarta ( talk)

• Those are self-published sources, see WP:SPS – Quinta ( talk)

• No they're not. – Prima ( talk)

• Delete. No search results that would indicate notability of this event. – Quinta ( talk)

• Events of this nature are rarely covered by online news organizations, . – Quarta ( talk)

In some other types of discussions, there are different sections for people to say they support or oppose, and for people to converse about different aspects of the subject of the discussion. Perhaps this would be a good idea to try at AfD.