Therefore a "reality-based" or even "RTS-based" autocalculation of a "1 trooper vs 5 troopers" battle had to lead to a 100% chance to lose the battle and a very, very, very tiny chance to even kill one single enemy unit.
While I'm not so sure about Sinister's calculations either, I think you are potentially wrong with that quote as well (sorry).
At least given the internal logic of the game.
If we would play the battle in RTS mode instead of using autoresolve, we wouldn't see our troops directly facing of against the lone trooper.
We would have our troops start at their base. Likewise would the enemy trooper.
While we would directly march towards the enemy base, the enemy would instead start building more troops. Of course it wouldn't be a lot of troops, but let's assume that by the time we reach the base, he has produced ~5 more troopers. Suddenly the odds are totally different.
While the enemy still has no realistic way of winning the battle, he will still inflict higher losses than a single trooper could.
Now, we are talking about autoresolve here, not the RTS mode, but autoresolve should still be balanced somewhat to the RTS part (otherwise autoresolve would be vastly superior to RTS mode for such battles, which is strange considering that the RTS part should probably give better results due to it being controlled by yourself, in form of a friggin dragon).
My personal suggestion would be to first adjust the RTS part in a way that what I described is less likely to occur (e.g. increase building time/costs, put in a timer before you could start building new units in RTS, etc.). That would probably result in lower losses in a RTS match, in turn allowing for more sensible losses in autoresolve as well.
