There is a concept called “volatility” used by codechef, which is in play here. Like in probability, we express “Expected value and variance”. Volatility corresponds to variance, or in other words, “How sure are we that he will be able to get the expected rank.”
Of course, if we arent sure about that, then him getting expected rank should lead to a bigger increase in rating (and our expectation next time :3). This was the reason I suffered a ~80 loss of rating in lunchtime despite <200 rank.
Further, I think it was made by a friendly user of our community, not codechef, so I am not sure if @admin can do anything about it.
it is to accelerate the convergence to the 'correct' rating. So when a person starts out, her volatility will be high, so that if she's like petr, her rating can jump up a lot in the first few contests itself. And then it tends to stabilize.