hi hi
As a game designer myself, I can tell you that positive and negative feedback loops are things that get a lot of thought put into them from start to finish.
Andrzej Marczewski has a
good overview of positive and negative feedback. But basically, any particularly long running game is going to need both positive feedback loops and negative loops to kick in at different points in order to prevent players from getting bored or frustrated. One example of the interplay between different feedback loops in real life is how hunger propels a predator to hunt, and then eating food provides energy for it to hunt more, and failing to eat food reduces its energy for hunting; but having too much food also propels a predator to become lethargic or become overweight and its hunting capacity decreases.
Positive feedback loops are generally a lot easier to work with than negative feedback loops, and what Homeworld 2 really needed was a well designed negative feedback loop to prevent players from becoming overly successful and bored. Instead, they made a rather poorly designed positive loop for the AI, which for a lot of players, seems to magically counter their every move. If you can manage to hide a Psychic AI under the bed so that people don't realize it, then one might squeak by, but if people realize that the AI is blatantly cheating, they're generally not going to feel happy about it.
The exception to that is where this next point comes in. Different challenge types.
There are several basic kinds of challenge and even those can be broken down into subcategories in a game. A game where you fight against an AI with perfect aim might be enjoyable if the AI is deficient in other ways that allows a player to get the upper hand, like not noticing explosive mines affixed to doorways.
Part of Homeworld 2's problem is that there was only one real mode of challenge, optimizing production of rocks, paper, or scissors. Another part of the problem was that, in an effort to make the game more fast paced than its predecessor, the feedback loop was able to progress a lot faster as well, apparently beyond the ability of many people to respond to it.
If a game designer can implement more than one kind of challenge, they can allow a player to follow a feedback loop and become very strong in one regard, while making things harder in another way. In many real time strategy games, one way they do this is by increasing the strength of micromanagement as troop size decreases, and decreasing the strength of micromanagement when troop size increases. (Click's per minute on production, vs clicks per minute on unit actions.)
If I were to make a Homeworld style game, where resources are carried over from mission to mission, I might try to include a maintenance and optimization mechanic. (Something like this was used in Mechcommander, which is a favorite game of mine which also has resource carryover.) Similar to the unit cap per mission, but more of a soft cap, having the ability to upgrade surviving ships into veteran units with abilities optimized to one's gameplay style would help cut down on player power creep, so long as there is only so much maintenance capacity between missions to go around.
If players see that they can only keep X value of units in prime fighting condition between games, they may decide it is worth it to scrap a larger fleet in favor of a mechanically less powerful, but more easily controllable fleet. Or they might decide to keep some of the fleet in mothballs in reserve, in case they have a bad run on another mission. Or they might decide to bring all of them in for an important battle, knowing that many of them will be fighting at reduced capacity, but its ok because they just need to buy time for objective Y to happen.
That, I think, could work well to prevent cross-level creep, while still rewarding players who complete a level without many losses. But for in-level power feedback, which Homeworld 2 was also pretty brutal about, I think one way to add a negative feedback loop without strictly punishing the player for their success would be to have things happen that are not directly related to the victory conditions. Like in Ender's game, the victory condition was not defeating the enemy forces, the victory condition was reaching the goal. If you have ships with strategic utility that is not a function of destroying the enemy fleet better, you can include goals that don't just involve "rock, paper, scissoring," the enemy fleet better.