Rank Sitting:
This implies this player is playing regularly compared to historic activity. If a player isn’t playing much Dota at all, they are not adjusted. If a player plays regularly and substantially more unranked/competitive Dota relative to ranked (if ranked is ~<15% of their games within the last 6 months) adjust using the following formula:
25x - 1250
x = lifetime ranked winrate.
This answers the question: How much mmr would this player gain if we asked them to play 50 ranked games?
Using the orange line here, plug lifetime win rate for x, find the Y coordinate, and dividing the result by 2 (for us halving adjustment impact) : https://www.desmos.com/calculator/brkussnwrh
Sandbagging:
If recent (within 1 month or 1 week, or whenever low win rate started occurring) ranked win rates are below 2 standard deviations away from lifetime ranked win rate adjust using the following formula:
(X((100-2Y)/100)x25)
X = games Y = ranked winrate
Divide this answer by 2 for us halving adjustment impact
This answers the question: What mmr was this player before they started losing games outside of normal deviation (95% of all probabilities)?
If players do not qualify for either, they are not adjusted