With a dropping revert rate we can think of a few different causes, not mutually exclusive, with very different implications on our efforts:
- Quality of edits is getting better
- Quality hasn't changed, but oversight is getting worse-->leading to lower overall quality
- Quality hasn't changed, but oversight is getting more forgiving, leading to no change in overall quality
Questions around the scope and vector
- is it project specific or universal?
- is it correlated to edit increases?
- it it correlated to anon/bot/user tenure status?
Most direct quality question
- Can we look at average ORES scores of un-reverted edits over time? That might get us what we need to know. But there might be modeling artifacts that confound the results. We'd need to run it by Aaron @Halfak who created and maintains ORES.