User:Vaticidalprophet/GA reform

GA sweeps
The history of GA, to the best of my ability to backfill, is that of an originally embarrassing Brilliant Prose-style 'process' that got whipped into shape c. 2008-2009. A major part of this shape-whipping was a systematic sweep of all extant GAs to assess their compliance with the standards. This was a vitally important process that permitted massive strides and maintained the quality of a broad swathe of articles.

It also took nearly three years, and we have over ten times as many GAs now as we did then.

The current situation is unsustainable. Many GAs have not been looked at in well over a decade, and the varying standards in the Reviewer Roulette process make it impossible to tell if even recent reviews are of any kind of quality. However, we can't really afford to spend thirty years sweeping. In all likelihood, we're reaching a turning point which will demand either a radical realignment or a complete deprecation of the GA process, as the backlog of likely-out-of-compliance GAs grows so high as to be completely unmanageable.

The solution is to break the process into chunks. Rather than sweeping every GA, we sweep GAs in high-risk categories that are more likely than average to be noncompliant, and high-view GAs that make up a disproportionate share of the reader experience. My current thought is to sweep all GAs shorter than 800 words; in my opinion, relatively few articles below a thousand words are able to give sufficiently broad content coverage. (Opinions on this front vary quite significantly, and many short GAs are perfectly within standards -- 'relatively' is a powerful word -- but I am confident the given length range will contain a far higher proportion of articles needing further review than average.) This makes up [shit I need to run a quarry or something] GAs.

A sweep of high-view GAs will also be necessary at some point. Details can be worked out once I figure out how many GAs are below that length benchmark.

How to handle length benchmarks? Getting numbers for prose size is exceptionally difficult. Numbers can be gathered for absolute character length; 8000 bytes or shorter gets you 660 GAs, the longest of which are a bit over 800. This is a good base to work from, but a lot of short GAs are much longer than this in absolute bytes and may make up a disproportionate share of substandard GAs. I've noticed a lot of poor-quality (including BLP vio) GAs for entertainers in the sub-800 range that are much longer than that on absolute bytes because of the inclusion of filmography, discography, etc tables. Will need to run multiple queries broken down by e.g. individual project to get a full list.
 * Difficulty determining a good break point for the first list -- there seem to be a lot of GAs just above 8000 such that when you bump it to just 9000 the list expands to 1092 articles; this may be a better starting point, as there seem to be a lot of borderline ones in the higher reaches of the expanded list, but it gets more false positives (which may as well be looked at too)

One consideration about the rapidly rising number of GAs is that it represents not just a lack of reassessing old GAs, but also a lack of people bringing articles to FAC. Sweeps shouldn't focus exclusively on substandard articles, but also on high-quality ones with FA potential. Potential FAs should have plans developed to bring them to that standard, and their authors encouraged to work towards the goal.

Crackpot scratchpad
A repository for half-baked ideas. Don't rely on these or assume any sanity from them.
 * A-class is a good concept, but only a couple projects do it -- what would a functioning system do?
 * Many articles are far better than the average GA, but will need more work for FAC than possible at this time (Prehistoric religion needs three whole sub-articles written to prevent people complaining about the length before I take it there)
 * An intermediate between GA and FA, rolled out over the whole project?
 * Unforeseen consequences -- "everything not mandatory is forbidden", would this raise the risk of A-class becoming expected/mandatory? Avoid overcomplexity and bureaucracy wherever possible
 * Would there be a need for this if GA standards didn't suck?
 * Is there a need for GA at all?
 * "Approved Article" rating -- an intermediate (or a deprecation of GA?) across the project that gets a main page slot
 * Deprecate ITN, it's clogging up valuable real estate with one of the single most dysfunctional processes on the project
 * Replace ITN slot with "Today's approved article" ("Today's good article" is ambiguous to readers unfamiliar with our ridiculous internal language), which run 12-hour slots rather than TFA's 24
 * To avoid running out of AAs, GA sweep should classify as much stuff as it directly as possible
 * "Sweep coordinators"/"AA coordinators" can promote with agreement between the group?
 * Clear "hierarchy of reward" -- DYK slot for GA, 12-hour slot for AA, 24-hour slot for FA
 * Or just have the guts to replace GA and re-rate everything as AA or B
 * Pop A-class down to an intermediate between B and GA (or AA, whatever) assigned by individuals rather than through a quality-assessment process? (Advantage of less stupid naming scheme)
 * The projects that currently use A-class quite like it, but have inconsistent standards between them as to what it is, complicating its use
 * Any idea like this that actually came to fruition would need extensive hashing-out with MILHIST, WPTC, et al for a solution they can work with