Two sources of uncertainty
Real-time strategy games combine two different kinds of not knowing.
The first is missing information. Fog of war hides what your opponent is building, where their army is and whether they are about to attack. You are reasoning from partial evidence.
The second is random outcomes. Depending on the game, damage may fall in a range, attacks may miss, or abilities may trigger by chance. Even with perfect information, a fight is not fully predictable.
Treating these two separately helps. Missing information can be reduced by scouting. Random outcomes cannot be removed, only planned around. Both are forms of risk under uncertainty, and the aim is to make decisions that hold up across the range of things that might happen, not to guess a single future.
Forming a belief from limited scouting
Players constantly update their picture of the opponent. You see a certain building, notice that they have not expanded, or find their workers missing from one area. Each observation shifts what is likely without proving anything.
A careful way to think about this is Bayes theorem in everyday form: start with how likely you thought an attack was, then adjust according to how much more often you would see the evidence if it were true than if it were not. A scout that finds an empty base is meaningful only if an opponent preparing an attack would usually have units there. Otherwise the observation tells you little.
Two habits help:
- Remember base rates. If most opponents at your level rarely attack early, one ambiguous sign should not convince you that one is coming. Ignoring this is called base rate neglect.
- Avoid overreacting to a single scout. One observation is a small sample. Several consistent signals deserve more weight.
Random combat and spread of results
Suppose a skirmish has a random element. The average outcome might favour you, but the spread of possible outcomes matters just as much. Averages describe what happens over many fights. You only play the one in front of you.
Consider an illustration with made-up numbers. A raid has a 60% chance of success. If it works, you gain 10 units of value. If it fails, you lose 8. The expected value is:
0.6 × 10 + 0.4 × (−8) = 6 − 3.2 = 2.8
A safe expansion alternative gives a guaranteed 2. On average, the raid is slightly better (2.8 versus 2). But averages ignore context. If losing 8 would cost you the game and gaining 10 would only help moderately, the safe choice may be better. If you are far behind and the safe option is likely to lose slowly anyway, the gamble becomes more sensible. The right choice depends on your position, which is the practical meaning of risk tolerance.
Matching your risk to the situation
A few principles follow from this, none of them guarantees:
- When ahead, reduce variance. If your position is already favourable, simpler, lower-risk plays tend to preserve it, because randomness mostly gives the opponent chances to catch up.
- When behind, variance can be useful. A plan with a smaller chance of success may be better than a near-certain slow loss. Even then, it only changes odds, not outcomes.
- Keep reserves. Plans that survive an unlucky roll are more robust than plans that need everything to go right.
Do not misread short runs
Strategy games produce memorable moments, such as a unit surviving at 1 health or an attack missing three times. Short sequences like these are normal. A few events say little about whether a build is good, because small sample sizes mislead. It is easy to blame a loss on luck or credit a win to skill when the real cause was a mix.
Equally, a sequence of misses does not make a hit more likely next time. Each roll is independent unless the game explicitly says otherwise. Expecting a correction is the gambler’s fallacy.
Learning from outcomes sensibly
After a match, it helps to ask whether the decision was good given what you knew, rather than whether the result was good. A reasonable plan can fail on bad luck, and a poor plan can succeed on good luck. Judging only by results makes it easy to repeat bad habits or abandon sound ones.
Over many games, skill tends to show up in results because luck evens out, which is the effect behind the law of large numbers. Within one match, it may not. This is part of the broader balance between luck versus skill that designers aim to tune.
A short checklist
- Separate what you do not know from what is random.
- Update your belief gradually, not dramatically, from each scout.
- Compare options by average outcome and by how bad the worst case is.
- Adjust risk to your position.
- Judge decisions by their quality, not just their result.
Nothing here lets you control luck. It only helps you make steadier choices when information is incomplete and outcomes are uncertain.