PT4 calculates ICM-adjusted winnings: for every pre-river all-in, it calculates the difference between the tourney equity in $ (according to the ICM) you've actually gained or dropped and the tourney equity you'd gain or lose on average if if you 'ran it infinitely many times', and adds that difference to the total tourney prize to get the adjusted winnings.
For example: assume that there are three players left with equal stacks in a 6-man $109 SnG, the prize structure is 65%/35%, and your equity is 55% at the point where you and the other blind go all-in (the button has folded). To simplify, assume that you win the entire pot 55% of the time and bust out in third the other 45% of the time.
Then before this hand your net tourney equity is $200 - $109 = $91 (1/3 of the prize pool, as stacks are equal, minus the BI; trackers ignore the influence of posting blinds and don't run future game simulation (FGS) as it would be too slow to compute). When you go AI, your average net tourney equity becomes 45% * $0 + 55% * (2/3 * $390 + 1/3 * 210) - $109 = 55% * $330 - $109 = $72.5, because if you win this AI, you'll have a 2:1 chip lead in the HU and be supposed to win it in 2/3 of cases (ignoring skill).
So, even though you were a slight favourite, you've dropped $18.5 ICM-adjusted, no matter what the outcome of the hand was.
Assuming that you won the hand and then the entire SnG without any further (nor prior) pre-river AIs (stack your opponent on the river in its last hand), you'll have won $281 net, but $281 - $18.5 = $262.5 net adjusted.
ICM-adjusted winnings are a better (faster-convergent) and unbiased (up to the imperfection of the ICM) measure of skill because they exclude the pure RNG luck factor but keep everything that's strategy-related. The acceleration of convergence depends on the tourney type, e.g. in turbos you need a twice smaller sample (number of games) to confirm or reject a winrate hypothesis (e.g. that you're a longterm post-cashback winner) basing on the ICM EV than on actual winnings; for hypers, this acceleration factor may be 3-4.
But beware of trackers' tourney detection mistakes. E.g. I chronically 'run below EV' in iPoker jackpot SnGs just because their prize structures are detected incorrectly and the ICM EV is miscalculated, so I have to switch the orange line off altogether. The EV line should be correct for Stars turbos (and, I hope, all Stars SnGs), but every time you pick up a new SnG type, check if PT4 detects its prize structure correctly (by right-clicking on a tourney where you got HU in a report and choosing to edit its prize structure manually).