codidact
Abilities 2.0: proportions of most recent actions, not all-time Wilson scores
meta:288363
Our Wilson-score-based abilities system is too hard for users to break into -- one bad outcome three years ago can doom you for years to come. There's been lots of discussion over the years, not just that meta post. Reviewing those answers (and their votes) and after lots of discussion in chat, my original proposal was that ability criteria be determined by "at least N positive within T1 and no more than M negative within T2". So, for each case you'd have four values: positive #, positive timeframe, negative #, and negative timeframe. Raw numbers, not percentages.
After more discussion on Discord: let's simplify: X% of good events within the last N events (two values to set rather than four), defer time constraints. We're only looking at resolved events, so either they were successful or not, so we don't need two Ns (or %s). For each row, you set a percent and a number, like 80% of the last 10 flags or 90% of the last 20 edits. The UI could be similar to the current abilities UI:
We'll want to discuss what the values should be for existing abilities and then work out a transition. Meanwhile, https://github.com/codidact/qpixel/issues/1650 would be best implemented by adding a new ability, so I propose creating the Abilities 2.0 framework as described here, implementing a single ability to support comments (i.e. do not change current abilities at all yet), and using that as a test case for tweaking the new abilities system before changing any existing ones. So the table for this effort would look like this:
After we implement this, we can use it to address https://github.com/codidact/qpixel/issues/1650.
Requesting input from @luap42 , @ArtOfCode- , @trichoplax , @Oaphi . (Input welcome from anyone else!)
From discussion on Discord: let's simplify: X% of good events within the last N events (two values to set rather than four), defer time constraints. We're only looking at resolved events, so either they were successful or not, so we don't need two Ns (or %s). For each row, you set a percent and a number, like 80% of the last 10 flags or 90% of the last 20 edits.