Balance is a process that generally should come at the end of a game-building process, since any changes along the way could shift the balance significantly. That principle, however, comes with several caveats that should be taken into consideration much earlier in gameplay. The first is that all decisions made in the game design process will affect the eventual balancing process. Specifically, they will make it easier or more difficult. Choosing to make asymmetrical races will make the game more difficult to balance; making the game more complex will make balance even more difficult. Of course, this does not mean that the game should not have complexities. Instead, it means that a studio should be prepared to invest the time and money into balancing a complexity properly before they put it into the game. In the case of our project Replaceable Parts, the game will probably never make it past the theoretical stage, but we should still consider the balance implications of any game decisions we make.
Some complexities will make the game particularly difficult to balance. In particular, free win possibilities fall in this category. Examples of possibilities like this are dark templar in SC:BW or SC2; a dark templar in a base without detection is a free ticket to victory. Air unit rushes are a similar species. You can guarantee that these units were looked at very closely by Blizzard to ensure that they were stoppable. It is not necessarily bad to include such strategies in a game, but they are nightmares for balancing a game, and should be used lightly and with caution. A game that has too many free win buttons quickly becomes a game of coin flips and rock-paper-scissors, which is bad game design whether or not it creates a statistical imbalance in win rates.
Design decisions can also make balance easier. Scouting makes balance considerably easier, since the game can be balanced on the assumption that any strategy that approaches becoming too strong will also have telltale signs that indicate to the opposing player that the strategy might be coming. And, of course, there must be a way to beat that strategy once you pin your opponent on using it, so "counters" are a good way to ensure that the game remains relatively balanced. But "counter charts" are a slightly unsophisticated solution to the problem, and there are better ways to deal with problems.
For example, consider the infestor broodlord composition used in Starcraft 2 by Zergs. for some time it was considered imbalanced, and it is certainly a powerful army. One reason for this is that the units so neatly cover each others' weaknesses. Broodlords tend to be weak to either air units or clumps of ground units darting in to pick them off; infestors are fairly effective against air units, and destroy clumps that dart in. Infestors, meanwhile, are vulnerable to very small numbers of units darting in, and generally to large ground units; broodlords can easily kill any small numbers of units that dart in, and large ground armies are simply ineffective against broodlords. A group of marines large enough to dart in and kill some broodlords before dying to broodlings is large enough to fungal growth to great effect; a group small enough to render fungal growth impractical will be killed by broodlings before they can kill anything.
So many of the obvious "counters" that might appear on a table of counters are ineffective against the composition. Many called the composition imbalanced, and I'm not going to take a position on whether they were right or wrong. But the best response, at least, could not be found on any counters list. The weakness that infestors and broodlords share is immobility; the answer was (and is) not in another unit, but in hitting the Zerg where their army wasn't. So rather than designing units with counters from other races, it is better to design each unit with inherent weaknesses. Then enemy units will have potential answers to these weaknesses, but the "counter" terminology is not terribly useful, since not all weaknesses are answered by a specific unit.
These considerations are especially important for Replaceable Parts because it's based on a system in which the gameplay situations are not clear to the designer. A game could be designed with relatively few gameplay options, and designers could evaluate the strategic implications of all of these to balance such a game, but this wouldn't be a very good game. A better game is one complex enough that the strategies players will come up with may or may not be something the developers foresaw, but if the developers don't know what strategies will be used, they can't be certain that those strategies are balanced against each other. Given that this is the case, measures must be taken to ensure that there are responses to most strategies so that no strategy becomes too powerful.
There are some forces that help balance happen naturally. Players tend to evaluate their strategies based on win rates. So while they are always looking for the best strategy, they generally do not innovate greatly when a matchup is going in their favor (choosing instead to focus on other matchups). When they are losing, on the other hand, they search around for strategies that can swing the game more in their favor, and the more often they lose, the more creative they get. So the longer a matchup favors one race, the more and more likely that a strategy will be discovered that will swing things in the other direction, while the race that has been winning has not been developing their side of the matchup as much.
Of course, the majority of balancing goes on in the playtesting stage of a game. While Replaceable Parts will likely never make it to that stage, it is worth discussing how that can help the balance of a game and, more specifically, how it can not. Playtesting is a beautiful thing, and it can reveal facts about the game that no other design practice can. That said, everything discovered in this stage should be taken with a grain of salt. Players in this stage will find some strategy or set of strategies that they consider imbalanced, and they may be right, but more likely than not that strategy is just the best that someone could come up with in the brief playtesting stage. The strategies of your game will endure a great deal more development than could ever be simulated in the playtesting stage, and whatever strategy is being criticized will undoubtedly be figured out and beaten by the time the game makes it onto shelves. This doesn't mean playtesting is worthless, but it does mean that it cannot be trusted to find all issues before the game is released. Considering the balance implications of game decisions during the development stage is important to designing a well-balanced game, and measures can be taken that make balance easier.