Hold on, you can lower your rating by winning!?!?!?
That is just unfair to any good player.
Technically, the end product you see which we call a rating can indeed decrease by winning. Bayesian ELO (which is used on warlight's ladders) uses some fairly complex maths, but I can give you a simplified version of what's happening.
Imagine you measure two things to determine how good a given player is (so you're calculating their rating).
Measure 'X' will be what you think their rating is.
Measure 'Y' will be how sure you are they are actually at that rating.
Now, when a new team is introduced on the ladder, our variable Y will be very low. If a team wins or loses some games, Y will get better. But when a team only loses games or only wins games, the rating system is not sure how good its estimate is. This is also why runs can happen. You can basically abuse these mechanics. (I got a ladder rating of 2340 once. Am I that good? Absolutely not.)
Measure X and Y are both used when calculating one's rating. To effectively calculate it, We take the estimated ladder rating (X) and adjust it for our certainty (Y). When you see a team dropping in rating after a win, the rating system is essentially saying they're now more sure about their rating than first. But since we only see the end product instead of the volatility (Y) with it, it seems like a team can lose points for winning a game.
EDIT: to clarify: you can indeed lose points after winning a game. This happens rarely but it happens. It means that your next games will influence your rating less. Hence the many complaints about ratings not being reflective of one's skill after sustained ladder play. This is also why Bayesian ELO is a rather poor choice for a continuous event like a ladder and a better choice for tournaments. Continuous events should benefit from a rating system like Glicko2 or regular ELO.
Edited 2/17/2020 21:44:04