User:LEvalyn/GA process analysis

I love Wikipedia's Good Articles. I do not love that the backlog of unreviewed GA nominations broke 1,000 for the first time ever at the end of May. In this article (and likely future articles), I aim to understand the history, current state, and trajectory of the GA process. I've saved the details of my data-gathering and analysis methods for the end. If you have additional questions about the GA system/pipeline, I would be delighted to do more investigation in future articles.

• When did the backlog start to grow? Was it a gradual change or did something happen?
• What is the "net flow" over time?
• Is the backlog growing because there are more nominations, fewer reviews, or both?

The backlog is dramatic, but I also want to give serious consideration to something that is much harder to measure: typical wait time. In theory, a large well-functioning system could have a backlog of 1,000 while also having high throughput and turnover such that nobody waited more than a few days. I argue that 'median time to review-start' is the metric most closely tied to the 'quality of life' of reviewers.

• What is the typical time-to-pickup over time? (Should this be mean, median, something else...? I partly want to understand how outlier fast/slow pickups are distributed)
• Over time, how much of the backlog consists of very old nominations?
• Do nominations get smoothly more likely to be picked up as they age? Or do very new or very old nominations get treated differently?

• Does mean time-to-pickup differ by subject category? Outlier fast / slow pickups? Backlog size?

AI use disclosure: All words on this page are purely human (me). All data and visualizations are generated by Python scripts (ie, their origin is deterministic and inspectable, and the results are as accurate as I can make them). Those Python scripts were written with substantial assistance from Claude Code.