So the weighted antag picker has been out of commission since mid 2020 (technically, it's around but set to roll "pick random player" 100% of the time)
But randomness is a fickle mistress and frankly I don't like it much. I barely trust the RNG to make a decent mining level, let alone distribute antag rounds in a way that feels fair.
I'm looking at the code and while I don't know how the API stuff works, here's how the picker works:
Code: Select all
For a single antag role and a pool of ckeys with the role enabled,
the picker calculates the percentage of antag picks versus rounds
connected to the lobby ("rounds_seen", not "rounds_participated")
The ckey's weight is that + some more math that I think is the same for all?
It'll add each ckey into a list sorted by those weights, descending
Players that have never been selected but are eligible get top priority
Then, for however many antags of that role need filling, it'll take
the first/next ckey in the sorted list and roll a chance to pick randomly
(currently 100%, but 10% prior to disabling)
if it doesn't pick randomly, it'll add the ckey it took instead.
Mixed will repeat this process for every antag type it's determined randomly.The (potential) problems are as follows (thanks in part to past pali on the PR disabling it):
-Because the system picks the first (highest) n ckeys sorted by weights, all that matters is whether you're at the top of the list. The only randomness involved was the 10% fuck you chance.
-Just making it to the lobby is enough to count?? The hell??
-The system doesn't account for how many rounds the player was actually eligible. Turning off antag roles for a while will count rounds towards your totals, so then turning them on again will land you several picks in a row. I've experienced this effect firsthand.
-As someone plays more rounds, the influence of one antag pick has less effect on their percentage
So the question is, do we want antag picking to accomplish?
-Some folks are going to play more rounds than others, and naturally they'll get more antag rounds as a result. Do we want to compensate for this at all?
-Do we want to cap the amount of antag rolls someone can get in a row?
-How far into the past do we want to care? Total history looks to be flawed, but do we just care about rounds since last pick or should we look farther back than that?
-Do we want round performance to influence future weights? Like completing objectives gives you a tiny nudge up in the future. (this is probably a great or a terrible idea, no in between)
-I can't tell if the database records picks per antag type or if the selected/seen percentage is based on all rounds and antags, but would we want to track those separately?
-Should some game modes affect your ongoing weights differently than others?
