The Heuristic Circle of Real-Time Strategy Process: A StarCraft: Brood War Case Study
by Simon DorAbstract
This article aims to describe competitive playing experience in StarCraft: Brood War. Strategy is defined as a process using game plans (strategies) and game states. By using cognitive psychology works, as well as their applications to chess and in film studies, the goal of this article is to summarize cognitive and perceptive processes in the heuristic circle of real-time strategy process. This model is based on three levels of strategic plans (operational, mobilized and projected plans) as well as on three levels of game states in the player’s mind (immediate, inferred and anticipated game states). This conceptualization of strategy as a process and its usefulness for the understanding of real-time strategy games is then illustrated by a specific StarCraft game session analysis [1].
Keywords: starcraft, brood war, real-time strategy, strategy, cognition, perception, gameplay, schema
Studying Real-Time Strategy games (RTS) is not commonplace in game studies today. Most articles on the genre focus on its audiovisual representation, either to question its political implications (Mauco, 2005) or its cultural bias (Dillon, 2008; Ghys, 2012), or to understand the subject’s perspective on the narrative (Voorhees, 2008). There are of course some contributions on the strategic aspects. Alexander R. Galloway (2007) suggests that strategies in StarCraft: Brood War (Blizzard Entertainment, 1998) are algorithmically inscribed in the game, thus arguing that “zerg rush” is inherent to this faction rather than having emerged from gaming practices. Christian McCrea’s article on StarCraft spectatorship (2009) is a clear and precise response to Galloway’s analysis, although its description of the rock-paper-scissors dynamic in-between the different factions is not representative of actual strategies. Gerald A. Voorhees (2008) did include strategies in his analysis of StarCraft in a relevant way, but using what he sees as common strategies as if they were fixed narratives rather than a changing ecosystem. Tony Fortin (2004) goes as far as to argue that RTS is a genre that contributes to reinforce the proletarian masses in their social position, saying that the RTS player is formed to re-enact a pre-existing optimal strategy as quickly as possible, therefore being the perfect employee that serves optimally and unquestioningly what they have to do (Fortin, 2004, p. 58).
This overview of RTS studies brings to light the necessity of having a deep and thorough comprehension of what this genre is all about. James Paul Gee reminds us that “if you play games, you know that even violent games require more attention to strategy - to finding patterns and solving problems - than to the images, violent or not, on the screen” (2007, p. 3). Game researchers need also to comprehend how the game is seen strategically before analyzing and interpreting it from a representational, narrative, social, or political perspective. If playing a game can be seen as a political act, we must understand how a game is played before trying to understand how a game is inscribed in a specific cultural frame. Proper criticism of political discourses in RTS needs to understand what is at stake when actually playing a RTS. As such, preparing the grounds for RTS study is one of the goals of this article. How can we figure out how the process of strategy works from the perspective of an individual player in real-time strategy games?
The close-reading approach as described by Jim Bizzocchi and Joshua Tanenbaum (2011) will be preconized here, while nevertheless bearing in mind that games do not follow a technological or formal determinism. As Julian Hochberg and Virginia Brooks noted about film studies, film analysis has a detached posture, exterior to the online perception which is central in film viewing (1996, p. 381). With this inexorably exterior posture, cognitive film theory represents an interesting avenue to understand the viewer’s position from an external perspective. I used cognitive psychology for the same reasons and with the same limits. Finding with exact precision what is in the players’ brain is not the point here [2]. Therefore, I have based this article’s demonstration on StarCraft firsthand play, as well as on a viewer’s perspective of competitive play. I have played the game for months with a competitive mind, but according to ICCup standards [3], I would still be qualified as beginner. In order to be more familiar with competitions and common strategies, I have watched hundreds of StarCraft game sessions online, broadcasted by GomTV or distributed on YouTube.
It seemed obvious that to understand this genre, we should not talk about real-time and strategy as opposed, but real-time as a quality of strategy. How does a player deal with strategies - here defined as game plans - in real-time during a game session? Trying to illustrate the role of the cognitive process involved in real-time strategic thinking is the main goal of this article. In order to show the importance of strategic play, I will elaborate a model of competitive StarCraft: Brood War one-versus-one ladder play. First, I will describe what StarCraft is and will show how a player sees the game strategically. Then, I will propose a model of strategy - here defined as a cognitive process - that I call the “heuristic circle of real-time strategy process,” based on Bernard Perron’s (2006) previous works on videogame experience. Finally, I will exemplify these conclusions by analyzing a specific StarCraft gameplay session.
A Brief Description of StarCraft: Brood War
In StarCraft, as in most RTS games, the player has to manage an army by building a base, collecting resources (minerals and gas), creating and controlling units to destroy all their opponent’s buildings. In common tournament play, two players fight against each other and each one has to choose between three factions: Terrans, humans with mechanical technology, Zergs, insect-like aliens, and Protoss, aliens with psionic-based technologies. Each player’s units and buildings are identified by a specific color and must fight on a pre-established map that determines different obstacles (bridges, ramps, etc.) and resource locations. The expression “real-time” in “real-time strategy” implies that each player performs their actions at the same time: when they give an order to a certain unit by a mouse click or by hitting a key, the order takes a certain time and is fulfilled independently by the unit, thus giving the player the opportunity to manage something else in the meantime. Optimizing units’ actions is thus at the core of RTS gameplay.
The game starts with a single building and four workers. Each player has to gather resources with their worker and bring them back to their main building: Command Center for Terrans, Hatchery for Zergs and Nexus for Protoss. In order to have a faster resource gathering flow, players will constantly build new workers and will eventually establish new bases, called expansions.
Other building types will be needed to train military units, to research upgrades and new technology for existing units or to defend bases. Most buildings’ locations are limited in space: Zergs must build them on creep, generated around Hatcheries and expanded by Creep Colonies, while Protoss must build most buildings around another building type called Pylon. Terrans, however, do not have such location restrictions. Each military unit and building needs prerequisites in term of buildings, organized in a tech tree: thus, if a Zerg player wants to build Mutalisks, they will need a Spire building, which needs a Hatchery upgraded to Lair. But in order to train units, the player needs to raise their population limits, either with a building type (Terran, Protoss) or with a specific unit (Zerg). During a fight, each unit’s properties (damage, range, attack type, cooldown, armor, etc.) will determine how their hit points are affected. A unit or building will die or be destroyed when their hit points score goes down to zero.
Figure 1. Minimap in the bottom-left corner represents the whole map.
What is seen on screen is always only a part of the whole map, which is represented by a minimap on the bottom-left corner (Figure 1). But the screen will not show every element on the map: the player can only see information available to their units’ field of vision. Therefore, at the beginning of the game, the opponent’s decisions are hidden. A player has to send a scout in order to know the spawning location and moves of their opponent. Information is given to the player using the fog of war principle: an explored area is represented in normal light, while a location explored by a unit that is not there anymore is represented darker. The player can still see the topographical information (rivers, rocks, etc.) and the buildings location as they were last seen, but units’ locations are unavailable. An unexplored location is shown completely in black (Figure 2). The process of strategic thinking is very dependent upon this information dynamic.
Figure 2. Respectively, an unexplored, seen, and explored location.
Strategic Perception of the Game
When a player enters the game with a strategic perspective, they perceive the game with certain preconceptions and expectations which can be illustrated by the concept of schema. For psychologist Frederic C. Bartlett, schema plays a part in the whole cycle of perception: it is an internal “plan” that will guide a subject’s exploratory activity and movements. It will determine perception in a given situation, and this same perception can eventually change the schema itself (Bartlett, [1932] 1954, p. 54). This notion was adapted by Marvin Minsky under the name of frame. Minsky introduces frames as a structure of data that represents a stereotyped situation. Each frame will contain information on the situation itself and on what is possible after or what could replace it if needed; in short, the frame is not only the situation, but different contexts in which it can be placed (Minsky, 1975, p. 212). Some elements are more important than others for a specific frame; other elements may have to be detailed following a specific situation.
Each frame is inscribed within a frame-system, that is, a set of frames that could work under similar circumstances. Consider the frame “to order something in a café.” When entering an unknown café, I can presuppose that I will either be served by a waiter, have to order by waiting in line, or pour my own coffee before having to pay. The same specific situation - entering a café - gathers different schemata that will guide my perception regarding a real situation. In short, schema structures knowledge as well as prepares a subject to certain perceptions and not to others.
For the cognitive psychologist Ulric Neisser, when we perceive an object which we are familiar with, we immediately perceive a range of possible actions involving this object: ways to use it, different contexts it could be part of, and so on. Those meanings are directly perceived, before details could prove to us that they actually exist (Neisser, 1976, p. 71). These possible actions are a matter of perception, just like the shape or the colour of an object. Neisser gives the example of a pencil: just by seeing it, its “writing” function automatically comes to mind. This is what Neisser, after J. J. Gibson’s concept, names affordance. These perceived functions are of course not universal, but depend on the schemata of the perceiver.
Every natural object has a vast number of uses and potential meanings, and every optic array specifies an indefinite variety of possible properties. The perceiver selects among these properties and affordances, by virtue of specific readinesses for some or not for others. Perception of meaning, like the perception of other aspects of the environment, depends on schematic control of information pickup (Neisser, 1976, p. 72).
Neisser uses chess as another example. Looking at a chessboard, a baby only sees wooden pieces; we must first know chess in order to perceive the chessboard concept and the possibility of chess as a game. Information is theoretically available for every perceiver, but it will be perceived differently following each perceiver’s background (1976, p. 181), that is, what they are ready to see. During chess play, a good player “quite literally sees the position differently - more adequately and comprehensively - than a novice or a nonplayer would” (p. 180, emphasis preserved). They already perceive potential moves, which do not exist without game knowledge (p. 181). In the same way, the player’s strategy (defined here as game plan) will make them ready to perceive some game states rather than others.
In order to predict the moves of an opponent in a given game, the player needs to know the game in a similar way. One needs to grasp the same information from the same game states (p. 183). In the RTS context, this perception occurs rather fast. Actions and decisions a player makes are already directed by a schema that sometimes includes the anticipation of future actions (p. 182). As De Groot stated, experience plays a key role in understanding chess strategies: “one recognizes the opening type from the whole array of pieces, one sees immediately what is going on and what should, in principle, happen from typical configurations of pieces - known by experience” (quoted in Holding, 1985, p. 74). This is what, following Charles S. Peirce’s terminology, I shall call strategic habits.
According to Peirce, a “habit arises, when, having had the sensation of performing a certain act, m, on several occasions, a, b, c, we come to do it upon every occurrence of the general event, l, of which a, b, and c are special cases” (Peirce, [1868] 1991, p. 76). Habits work with things - paper will burn every time I put it in a burning fireplace - as well as with people and thoughts - a Protoss player will usually perform a Forge Wall-in against Zerg in order to defend against early Zerglings aggressions. Laws of nature would be best described as habits that almost every object on Earth tends to have, such as being attracted to the ground when we let them fall (Lefebvre, 2007, p. 147).
As Martin Lefebvre (2007, p. 148) puts it, there is no fundamental distinction between mind and matter for Peirce. Lefebvre calls cultural habits (habitudes culturelles) the category of behaviour that goes beyond the “physical” properties of an object (p. 174). For example, a diamond has the property of being hard; thus, it is used in surgical tools. But it also has a great value and is associated with conflicts in Africa. Culturally, a diamond has properties beyond its hardness or its visual characteristics. Humanities and social sciences are usually more concerned with these cultural habits (p. 178).
In a corresponding manner, I shall therefore call strategic habits the tendency for human players to react to specific gaming situations in certain ways. In addition to a player knowing that a Marine can damage flying units, that Dark Templar has the “permanent cloak” ability and that Reaver is not a flying unit, s/he knows that a certain game state will lead to certain game possibilities, and perceives them directly, even though it is not a question of exactitude. Let me suppose that a Terran player opens with a two Factories build. If it is scouted effectively, the opponent will almost immediately infer that Terran is going for a mechanical build. This inference comes from a habit of mind, “[t]hat which determines us, from given premisses, to draw one inference rather than another” (Peirce, [1877] 1991, p. 147). Of course, it could be a trap and the Terran player could prefer to opt for infantry units rather than mechanical units. But still, the co-occurrence of the two phenomenon - two early Factories and a mechanical army - is seen so often that it could automatically create a habit in the player’s mind, thus conducting them to react accordingly even without figuring out the exact build order of their opponent. It works the same way in chess play: “patterns are directly associated with appropriate plausible moves” (Charness, 1977, p. 42). In order to think fast in a RTS, forging this kind of habit has to be done. In psychological terms, seeing habits in objects corresponds to building a frame where they are included.
Even if habits and frames created by a player are a central tool for gameplay, it could at the same time be a handicap in some regards. As Minsky writes, “[p]roperly chosen, such stereotypes could serve as a storehouse of valuable heuristic plan-skeletons; badly selected, they could form paralyzing collections of irrational biases” (1975, p. 228-229). A given strategy can be a functional basis for a game, but at the same time, can be a way to discriminate their gaze according to initial misconceptions. During play, a player is not thinking of every possible situation according to the game rules: they are thinking of the habits and strategies which they have to adopt in order to react accordingly to their opponent. As a counterpart of the quickness of play these habits give, they can limit the possibilities of a player, who responds directly with a specific action to a precise situation.
The game aesthetics favours habit-taking and frame formation. In chess, pawns are identical, thus more easily identifiable and most likely to become part of a habit or affordance. StarCraft uses a similar principle, a “tile-based” aesthetic (Rollings and Adams, 2003, p. 340). For example, each Firebat is identical to another and each player knows what they might expect from a Sunken Colony. If a gaming situation involves Hydralisks, a player can easily associate this situation with another one involving other Hydralisks, because they are identical and have the same functions. In the mind of the perceiver, every Hydralisk is part of a single “continuum” in order to be compared (Peirce, [1863] 1986, p. 103).
For Neil Charness, when a chess player sees a game state, they regroup pieces in different chunks, that is, a unit in long-term memory (1977, p. 40). Chunking is a way to understand different singular elements within the same concept in order to mobilize them easily in short-term memory. In chess, experienced players will regroup pieces in different chunks, while a novice player must rely on chunks made of single pieces (p. 42). A similar principle is at work in StarCraft: an end-game army made of Ultralisks and Zerglings, even supported by Defilers, is usually called “Ultraling.” The whole game state will be grasped more easily on the grand scale rather than in the details of its composition. As short-term memory can store seven chunks, plus or minus two (Charness, 1977, p. 40), storing more than one unit or building per chunk makes habit-taking easier.
Different types of knowledge
Jacques Tardif ([1992] 1997, p. 47) claims that cognitive psychology divides knowledge into three types: declarative, procedural and conditional. Declarative knowledge is theoretical and empirical, made of facts, rules, laws, and so forth. In order to be used for action, it has to be transformed in procedural or conditional knowledge. Procedural knowledge implies the acquisition of the action itself; of its realisation. For Tardif, this kind of knowledge needs a retroaction (p. 51), which is often the case when learning in videogames (Gee, 2008, p. 21; Sauvé, Renaud & Gauvin, 2007, p. 97). Conditional knowledge, also known as “strategic knowledge,” adds the moment and the context in which a specific procedure should be used. These are best learnt when the learner is in a situation where they have a choice of procedures (p. 54). Each of these types of knowledge can be considered a schema, as Neisser explains: “The schema is not only the plan but also the executor of the plan. It is a pattern of action as well as a pattern for action” (1976, p. 56). As Bernard Perron (2006, p. 67) puts it, schemata are a question of “what” to do and of “how” to do it.
Procedural skills in videogames also involve sensory-motor skills. When a player wants to learn a new build order, three steps which correspond to the three types of knowledge are ahead of them. First, there is the idea of what is to be done, which could be borrowed from a strategy guide or built up by the player. Then, it is learnt as a procedural knowledge by practicing its execution. As Gee explains, “[w]hen learners learn a new skill set/strategy, they need to practice it over and over in varied contexts in order to make it operate at an almost unconscious routinized level” (2004, p. 71). It is only in the third phase, when the execution is mastered and the game mechanics well known, that the player will learn conditional knowledge: knowing what to do, how to do it, as well as when to do it. In StarCraft, it is one thing to know that your opponent builds an early Cybernetics Core; it is another one to know what to do as an efficient response. Similarly, sending scouts is an important habit to inherit (procedural knowledge), but the key is to know what to do with this information (conditional knowledge).
A circular conception of knowledge
By introducing schemata, Bartlett was clear that they have as much a directing function towards perception as they are modified by perception itself. Neisser adds that when an act of perception is made in a performance requiring skills, the performer “… acts, perceives the consequences of his actions, develops a more precise notion of what is to be done, acts again, perceives again, and so on until the final product is achieved” (1976, p. 51). Learning a set of actions in physical sports as well as in videogames is a heuristic principle. David Bordwell describes two different processes that operate simultaneously during film viewing:
“Bottom-up” processing refers to those fast, mandatory activities, usually sensory ones, that are “data-driven.” “Top-down” processes are concept-driven; they are more deliberative, volitional activities like problem-solving and abstract judgment (Bordwell, 1989, p. 18).
Bottom-up processing organises perception quickly, without a great collaboration from memory (Branigan, 1992, p. 37). Top-down processing will move from concepts towards perceptions: a general conception of a thing will be mobilized to direct perception in a specific direction that corresponds to that existing schema. Without pre-existing appropriate schemata, bottom-up will be prominent. In a situation where schemata are slowly confirmed by perception, top-down processes will be at work.
Bernard Perron adapted these processes to horror videogames analysis (2006, p. 66) [4]. For Perron, perception and cognition processes are best illustrated as circles, thus insisting on the retroaction part of spectatorship and gameplay activities (Figure 3). Existing schemata will guide the actions of the player, and these actions implemented on the interface will modify the actual game state. From the perception of a new game state, new schemata will be forged, or existing ones will be changed, and this “heuristic circle of gameplay” goes on again.
Figure 3. The heuristic circle of gameplay (Perron, 2006, p. 66).
But this circle works on another level, illustrated as an outer circle that goes on simultaneously. When an actual game state changes, the player can infer a potential game state, which also influences schemata and cognitive processes. These schemata will influence execution on the inner circle, as well as sensory-motor skills on the outer level. These execution skills help a player to think about potential new game states, thus operating the circle on two different levels. The heuristic circle of the real-time strategy process (Figure 4) model I suggest here came from this idea of different layers circling at the same time.
The Heuristic Circle of the Real-Time Strategy Process
The operational plan is the actual strategy effective at a given time. When a game session starts, the player has their build order in mind, which becomes the operational plan. This plan guides the sensory-motor skills to execute actions that change the immediate game state. Because a player knows beforehand the usual result of the order they gives to their units, this game state is modified in their mind before the action is implemented in the game. Still, if needed, the audio-visual response can give a confirmation to the player that their operational strategy is working normally.
Figure 4. The heuristic circle of real-time strategy process.
When encountering their opponent, or by trying to predict their actions, a player will infer a possible game state, from which they will be able to anticipate a future game state. Let me assume that the player scouts their opponent and sees that an expansion is setting up. With this in mind, a certain number of plans are mobilized in short-term memory. From the moment a game state is inferred, a certain number of plans are mobilized automatically, as a habit. Mobilized plans are short sequences of actions to implement on short- or mid-term, so that the projected plan works. Anticipating their opponent, the player will be able to create this projected plan, which will contribute to further anticipation of their opponent. The projected plan does not necessarily try to anticipate every action until the end of the game: it plays a functional role to choose mobilized strategies. Facing a fast expansion, a player could anticipate that the opponent’s army will be delayed and therefore plan an early aggression or, on the contrary, secure an expansion of their own. It is clear to us that strategies are not the only kind of knowledge that the player uses while playing: we therefore identify them as within the schemata zone of our figure, which could include “real-life” knowledge, the rules of the game and general knowledge about computer games.
A set of strategies could be mobilized following a projected plan. But these precise plans will change as needed by specific game circumstances. A mobilized plan corresponds more or less to a chunk and could be as short as securing an expansion, building a Starport, or having a Zealot army with their speed upgrade. The multitasking skills of the player will consider only one mobilized plan at a time, giving it for a short amount of time the status of the operational plan, depending on the action needed following the projected plan priorities or the specific circumstances. But the other mobilized plans still operate in parallel in their mind. The player knows that each mobilized plan is a potential set of actions in the actual circumstances, but is not necessarily what is leading their moves right now. If a player projects to gain a short term economic advantage, they can have three precise mobilized plans: quickly build two expansions, build Dragoons continuously and harass the opponent’s army to keep them on guard. Each of these three mobilized plans will be promoted to the operational plan - even if only for less than a second - and a player will usually want to optimize each of them, but according priority to some rather than others - which is guided by the projected plan.
When the projected plan is considered obsolete, whether it is because it has been countered, considered inefficient, or fulfilled, the player will select another projected plan. They can select one using every game state they perceive, infer, or anticipate, but the “real-time” aspect of the game suggests that a player usually chooses a plan by habit or by feeling, without necessarily judging its precise applicability, and changes it or refines it depending on new game states. Mobilized plans themselves can also contribute to the inferred game state: the fact that at a given time in the game, a player will want to build an expansion or move their army implies that they assume or anticipate an opponent’s precise or possible reaction. For example, an aggressive plan makes the player infer directly that their opponent will be reacting by training military units instead of building up their economy. The operational plan can also be modified directly by the bottom-up process. If a player is surprised by a backdoor attack, they will try to react as quickly as possible, but this reaction is not necessarily part of mobilized plans. Therefore, in this kind of situation, the operational plan was not necessarily a mobilized plan beforehand. Operational plans can be made up only through game habits. The player can get rid of the backdoor attack quickly, without interrupting their initial projected plan. However, it can, on the other hand, be so much more confusing and demanding all their attention that it disrupts the immediate game state. A new game state will have to be made up all over again in order for the cycle to work again.
The top-down process concerns neither execution nor procedural skills. Having an operational strategy directs game state perception. A player’s gaze will be more or less limited to units they consider useful according to this operational plan [5]. For example, this plan could require every bit of attention from the player to the point that they will not see other important aspects. Harassing strategies rely on this seeing principle: when a player sends a few units towards the mineral lines of their opponent, the latter almost has to react, otherwise their resource gathering will suffer so much in the long term that it could cost them the game. However, this switch to a defensive mode - and defensive operational plan - is most likely to make a player forget about their original mobilized plans, or become too much focused on this crucial defense so as to miss a simultaneous attack. “A highly probable (and thus highly guessable) event will contribute very little information when it occurs, whereas an improbable event will contribute a great deal” (Holding, 1985, p. 78). The more the game states are in line with strategic plans - projected, mobilized or operational -, the less they will be disturbed. If a new immediate game state was not planned, the player will have to integrate these new events to their inferred and anticipated states, thus losing more time to strategizing.
At the beginning of a game, a player will sometimes have a projected plan based on an anticipated game state. They can know their opponent and predict some possible actions. The player can force their opponent’s actions in the direction they want with an “all-in” plan, that is, a build order that could force their own defeat if their opponent can defend themselves against it. In normal circumstances, a player will begin with the same operational, mobilized and projected strategies as usual, and will try to gain in-game information before embarking on a specific road. Before using these concepts in a game analysis, I will summarize them and add a few remarks.
Precisions on Game States
The immediate game state is what is seen at a specific moment. But, this “seen” game space also includes its affordance: part of this game state is assumed as a result of a habit. For Umberto Eco ([1979] 1985, p. 63), a text is made of blank spaces that the reader can fill by at least two means: its encyclopedia and its “inferential work.” Encyclopedia includes every possibility permitted by the game rules known by experience (p. 148). It is made of every habit a player knows. By inference, the player will fill the spaces they cannot perceive. For example, by seeing Mutalisks, a player knows their opponent has built every building needed for the construction of this air unit. Depending on the moment when these Mutalisks are deployed, and following other clues such as the time an expansion has been built, the player will suppose that their opponent has a certain number of other units they did not see. This encyclopedia thus works with game rules themselves - prerequisites for a certain unit - as well as strategic habits - how Mutalisks are usually used. These signs will let a player fill the blank spaces to create a set of possible worlds (p. 165). In a game of StarCraft, a player will build a set of possible worlds following their encyclopedia and inferential work. But this world is not necessarily analogous to the “real world”: when the player knows the game sufficiently, using the real world to infer game elements is less relevant. In this regard, the set of possible worlds does not expand with experience; rather, it tends to shrink because these worlds are reinforced by habits. The player has to create those worlds really fast if they want to adapt their play accordingly. A conditional knowledge - what to do in a specific situation - is sometimes based on an inferred game state rather than an immediate state. A complete and precise inferred world can lead to making an efficient decision if it is close to the real game state, but the more precise it is, the more difficult it will be for the player to adapt if the inferred world failed to figure out the actual game state.
Precisions on Strategic Plans
I described strategic plans as three different levels, but we could also describe the operational plan as a mobilized plan used in real-time. Mobilized plans are plural because they are part of the arsenal of possible operational plans at any given moment. Projected plans are slightly different because they imply a longer-term situation. However, each plan is selected from within a greater set of plans from the player’s encyclopedia of possible strategic plans found in the player’s long-term memory.
As mentioned earlier, in chess as in StarCraft, choosing the right game plans can be a question of habits. But the choice of strategies is not only a question of automatism. For Charness, if a chess player must compare different possible moves, the player “must search selectively among the exponentially exploding number of positions as he attempts to look ahead” (1977, p. 36, emphasis preserved). From a specific situation (immediate, inferred, or anticipated), how can a player know which plans are more suited than others? How does a player choose a new strategy within a set of possibilities without having to search the whole set of possible worlds permitted by the game rules?
The frame method can illustrate this optimization process in a strategy game (Minsky, 1975, p. 258). A precise situation can be attached to different frames that are different possible actions already experienced and embedded in the player’s long-term memory. Each of these frames must be evaluated quickly so that the appropriate one can be chosen.
As noted previously, a gaming situation can sometimes be summarized in a single expression, a single chunk, which can regroup a lot of possibilities with a common point. For example, there are an infinite number of “containment” situations, that is, a moment in the game where your opponent is trying to block the entrance of your base. Players will have a summary in mind of every plan attached to a containment in order to navigate through them, but still have deeper information about every plan if needed (Minsky, 1975, p. 260). For example, if a player judges during a game that their strategy is inefficient in a precise situation, they can decide to keep the reasons in mind why, to make sure the mistakes are not repeated in a subsequent game. To act quickly, the player only needs a summary of each frame in a specific frame-system, which is each frame suited for an immediate, inferred, or anticipated game state. Starting with an immediate game state, if no operational plan seems natural, a player will select an appropriate strategic plan following the specific situation, with frame summaries. Mobilized and projected plans are selected in a similar way from inferred and anticipated states.
Game session analysis: WhiteRa (Protoss) vs yhnujmik (Terran)
In order to better demonstrate the utility of this model, I will analyze a game session between two competitive players. The match I chose was between WhiteRa (Protoss) and yhnujmik (Terran), on “Heartbreak Ridge (v1.2)” (Figure 5), December 2nd, 2009, during the ladder of the TeamLiquid StarLeague (TSL) [6]. This game was accessible in “replay” format, on the TeamLiquid website [7]. I will cover the whole game, but will detail key moments to illustrate how strategic plans and game states work.
WhiteRa is a Ukrainian Protoss player named Oleksiy Krupnyk, who was later eliminated in the quarter finals. The Terran player yhnujmik was banned from the competition because the IP used for this account was shared by a player named Yosh (TL.net Bot, 2009). Yosh is an American Terran player named Sherwin Mahbod. I cannot be a hundred percent sure that yhnujmik was Yosh trying to gain an unfair advantage of the rules of the tournament, so I will refer to him as “yhnujmik”, or simply “the Terran Player.”
Figure 5. Heartbreak Ridge version 1.2. White lines (added by myself) delimitates starting positions. The map features nine expansions, easily identified by the presence of the minerals, in light blue. Six crossable ridges create a spiral around the middle expansion and block the vision of the units.
Heartbreak Ridge [8] is a map with only two possible starting positions. As such, each player knows beforehand where his opponent spawned, which can delay the moment when they send a scout. The natural expansions are at the same height level as the main base and there is a narrow choke point between both bases. This expansion’s entrance is larger but can still be protected with some buildings. Additional expansions are more difficult to secure. The second expansion will usually be built on the high ground near the natural expansion (7 o'clock and 1 o'clock positions). A Terran player can defend both his second and third bases with the same Siege Tanks in Siege Mode and a Protoss player can block the access to the third base with three Pylons. The ridges on the map block the vision of units and let ranged high damage dealing units such as Tanks to take advantage of this when they have vision of the other side.
In this match, WhiteRa spawns at the 10 o’clock position and yhnujmik starts at the opposite of the map. The Protoss player already has the “12 Nexus” as his operational plan. This build order consists of an early expansion, by building the second Nexus at “12” food, between the moment when the 12th Probe started being trained and the 13th began training. If WhiteRa can defend this expansion, he will have an economic advantage.
The Terran player has a more standard opening and decides to delay his scout, to optimize his resource gathering. However, yhnujmik knows from experience that chances are that his opponent goes for a “proxy,” that is, hiding a production building (a Gateway in the case of Protoss) near the opponent’s base. With this building, Protoss could go for an early aggression and receive quick reinforcements. Therefore, Terran will add the early exploration of usual spots where the Protoss could hide a building to his mobilized strategic plans.
At this point, Protoss has the projected plan to secure his expansion, while yhnujmik’s projected plan is to be prepared for an early attack. Protoss has three mobilized plans: 1) to build Probes constantly and order them to gather minerals in both bases; 2) to build a Gateway and a few military units to be able to resist an early attack; 3) to send a scouting Probe after the Nexus construction is started. Terran has four mobilized plans: 1) to build SCV’s constantly and gather minerals in the main base; 2) to build a Barracks and then a Factory (a standard build, which leads to mechanical units); 3) to send an early scouting SCV where a Gateway could be hidden; 4) to send a scouting SCV in the Protoss base just before the Barracks is done, while still scouting around for a proxy.
Each scouting worker arrived in the opponent’s base relatively late, which is a Protoss advantage considering his build order leads to an economic advantage. Information gathering for the Protoss was denied as the Probe was stopped at the main base entrance, blocked by a wall of buildings. Terran, however, will realize that a second Nexus is already up and operational (Figure 6). He can then infer a game state where WhiteRa has few or no military units (B) because he knows that the resources were invested in the expansion instead of on a standard early Gateway. He can easily anticipate that this expansion will lead to an economic advantage for his opponent if he cannot counter it or have an expansion of his own (C). He then decides to change his projected plan. He tries to compensate for the economic lead of his opponent by denying the mining potential of the new base and by taking an expansion of his own (D). With his scouting SCV, yhnujmik blocks the narrow choke point between the two Protoss bases by building a Supply Depot, which he will repair with SCV’s (E). This plan will alternate with at least three other mobilized plans as an operational plan: bring reinforcements, efficiently control the units, and build a Command Center (F). Blocking the choke points could delay the arrival of Protoss reinforcements to defend the expansion from the Terran attack.
Figure 6. An illustration of the heuristic circle of real-time strategy process in play.
The Terran player has already built a Factory and so uses it to create a Vulture. The first attack squad consists of a Marine and two SCVs. Seeing the attack, WhiteRa changes his mobilized plans and defends: a Zealot will try to destroy the Supply Depot from the main base, while Probes from the expansion will try to neutralize the units. In addition to this defensive operational plan, WhiteRa also has a reconnaissance plan. The Terran player, being focused on his operational plan, missed the scouting Probe which snuck near his base. WhiteRa loses a few Probes in the attack, but manages to destroy the building with the Zealot and a Dragoon, before a Vulture and a third Marine come to support the attack. But the Probe WhiteRa snuck near the Terran base will be able to see the moment where yhnujmik builds his expansion (Figures 7 and 8). Even if the Terran quickly sends a Vulture to make sure no unit stealthily moved in, WhiteRa already has this precious information.
Figure 7. On the high ground, WhiteRa can see the Command Center under construction in the Terran base.
Figure 8. From the Terran’s vision, the high ground is hidden in the fog of war.
Keeping up with the spirit of his initial goal, the Protoss player will adopt a new projected plan: preserving the economic lead. At least two plans will be mobilized under this projected strategy: 1) to build and defend a second expansion; 2) to put pressure on the Terran player weakened by his failed skirmish and trying to gain his first expansion. This pressure is not so much a way of gaining a military advantage as a way of drawing the Terran player’s attention away from his own plans in order to diminish his resource gathering. WhiteRa nevertheless manages to destroy the Barracks that protected his opponent’s expansion and sneak in to the Terran main base with three Dragoons, which were killed after taking out a Vulture and a Tank. After the Terran Siege Mode upgrade for the Tanks, this pressure stops since Dragoons are outranged.
Near 7:50, Terran has the projected plan to start a “timing push” (TeamLiquid, 2010) with all his troops. This common Terran manoeuvre consists of an attack with Vultures and Siege Tanks aimed to be effective at the moment where Protoss has taken a second expansion, but has not made it fully operational yet. The Terran will try to slowly move with Siege Tanks, changing their modes from Siege Mode (which has a great attack force and range but no movement at all), to Tank Mode (which diminishes their strength but allows them to actually move) in order to keep a strong attack rate. Bunkers can help to absorb some damage in the army: the Terran player starts building one near the entrance of the Protoss natural expansion.
At that moment, both of the armies are similar in strength (three Tanks, four Vulture and three Marines against nine Dragoons), their workers’ number is comparable, but Protoss already has a third base operational. Both players also have the same production buildings (two Factories and one Barracks against three Gateways). WhiteRa takes advantage of quick reinforcements and army positioning and takes down the push as it starts. After the attack, Protoss is at 75 food, while Terran is only at 50.
Terran is still not aware of its economic lack, and changes his projected plan to denying Protoss expansions by having the map control. One of his mobilized plans is to send Vultures to place mines at expansion spots. At the same time, he will build a third base. This base will be in place when Protoss builds his fourth and fifth bases, having a population of 131 against 85. At 15:00, High Templars debark successfully at Terran expansions to cast Psionic Storm and kill workers, thus strongly damaging the already weak Terran economy, while the Terran responds with aerial landings of his own with Tanks and Vulture to keep WhiteRa on guard. WhiteRa then builds a sixth base. Both armies reach the maximum 200 food around 18:00. The Terran manages to take advantage of an engagement and carry on with his plan of limiting the Protoss economy by taking out two bases. Even though he loses some skirmishes, WhiteRa is still one base ahead. By using its mobility advantage and attacking with smaller squads on both top and bottom parts of the map at the same time, WhiteRa denies the Terran from taking any other expansions. Eventually, every potential new base for Terran is defended by WhiteRa’s squads and Terran cannot reinforce his army. WhiteRa takes this game.
Conclusion
The heuristic circle of real-time strategy processes is an attempt to summarize key ideas about the cognitive and perceptive process in StarCraft competitive play. This model is based on three levels of strategic plans: operational, mobilized, and projected plans. But these plans must be conceived simultaneously with three levels of game states in the player’s mind: immediate, inferred, and anticipated game states. The key idea here is that players must conceptualize different layers of strategy and different constructions of the game state. Thinking about future potential states is part of the strategic process and occurs at the same time as having to struggle between different mini-goals at any given moment. Therefore, the typical definition of “strategy” as game plans is not enough to describe what a “real-time” strategy game actually is: strategy is a process in which a player has to deal with strategies and game states.
Having a general knowledge about videogames is not necessarily enough to adequately see strategy at play in RTS; the cognitive frame used here is a good start for the understanding of this genre. Micro-analyses such as the game session between WhiteRa and yhnujmik is also an avenue I encourage: general statements on any video game risk being unaware of playing considerations. Being able to understand what is at stake in StarCraft and to understand players’ perception of gameplay is fundamental to any deep study of real-time strategy, and I encourage future investigations on this genre to consider the concerns I highlighted here. Criticism and political considerations in video games clearly necessitates a comprehension of the game actions beforehand, since this comprehension provides the context in which the player operates and makes decisions.
Notes
1 This article could not have been made without support and advice from Bernard Perron, and I deeply thank him. Thanks also to Kelly Boudreau for revision and proofreading.
2 Such an approach would probably bring interesting results, like the Skillcraft project lead by Mark Blair from Simon Fraser University is aiming to do (www.skillcraft.ca).
3 ICCup, short for International Cyber Cup, is a gaming server for StarCraft: Brood War, Warcraft III (Blizzard Entertainment, 2003), DOTA (a Warcraft III mod), and World of Warcraft (Blizzard Entertainment, 2004) that keep track of players’ score (www.iccup.com).
4 I will not discuss here the model Perron developed later with Dominic Arsenault (Arsenault & Perron 2009) which is more suited to the analysis of games of progression.
5 That is approximately what Dennis Holding stated about chess: “Whatever form of representation he uses it is clear, or at least strongly suggested by the evidence, that the player only considers a selected part of the board at any time. His imagined, remembered, or viewed selection - or zone of orientation - seems to be limited to the most active pieces or squares” (Holding, 1985, p. 65).
6 The ladder of the TSL scores each player according to the score of the opponents they beat. Each individual game is less important than in the elimination brackets.
7 A replay saves each action in a game session so that a player can watch it later. This replay was made available by TeamLiquid (www.teamliquid.net/forum/viewmessage.php?topic_id=115850). In the WhiteRa compressed file, the specific game is numbered #0228.
8 Information on this map and its image were taken from Liquipedia (wiki.teamliquid.net/starcraft/Heartbreak_Ridge).
References
Arsenault, Dominic & Perron, Bernard. (2009). In the Frame of the Magic Cycle. The Circle(s) of Gameplay. In Bernard Perron & Mark J. P. Wolf (eds.), The Video Game Theory Reader 2 (pp. 109-131). New York: Routledge.
Bartlett, Frederic Charles. ([1932] 1954). Remembering: A Study in Experimental and Social Psychology. Cambridge: The University Press.
Blizzard Entertainment. (1998). StarCraft: Brood War [PC]. Blizzard Entertainment.
Blizzard Entertainment. (2003). Warcraft III: The Frozen Throne [PC]. Blizzard Entertainment.
Blizzard Entertainment. (2004). World of Warcraft [PC]. Blizzard Entertainment.
Bizzocchi, Jim & Tanenbaum, Joshua. (2011). Well Read: Applying Close Reading Techniques to Gameplay Experiences. In Drew Davidson, et al. (ed.), Well Played 3.0: Video Games, Value and Meaning (pp. 262-290). Retrieved July 8, 2013, from www.etc.cmu.edu/etcpress/content/well-played-30-video-games-value-and-meaning.
Bordwell, David. (1989). A Case for Cognitivism. Iris, 9, 11-40.
Branigan, Edward. (1992). Narrative Comprehension and Film. London/ New York: Routledge.
Charness, Neil. (1977). Human Chess Skill. In Peter W. Frey (ed.), Chess Skill in Man and Machine (pp. 34-53). New York: Springer-Verlag.
Dillon, Beth A. (2008). Signifying the West: Colonialist Design in Age of Empires III: The WarChiefs. Eludamos, 2, 1. Retrieved July 8, 2013, from www.eludamos.org/index.php/eludamos/article/view/vol2no1-10.
Eco, Umberto. ([1979] 1985). Lector in fabula. Le role du lecteur. Paris: Grasset.
Fortin, Tony. (2004). L’idéologie des jeux vidéo. In Nicolas Santolaria & Laurent Trémel (eds.), Le grand jeu: débats autour de quelques avatars médiatiques (pp. 45-73). Paris: Presses universitaires de France.
Galloway, Alexander R. (2007). StarCraft, or, Balance. Grey Room, 28. Retrieved July 8, 2013, from www.mitpressjournals.org/doi/abs/10.1162/grey.2007.1.28.86.
Gee, James Paul. (2004). Situated Language and Learning: A Critique of Traditional Schooling. New York/ London: Routledge.
Gee, James Paul. (2007). Good Video Games + Good Learning: Collected Essays on Video Games, Learning and Literacy. New York: P. Lang.
Gee, James Paul. (2008). Learning and Games. In Katie Salen (ed.), The Ecology of Games: Connecting Youth, Games, and Learning (pp. 21-40). Cambridge, MA: MIT Press.
Ghys, Tuur. (2012). Technology Trees: Freedom and Determinism in Historical Strategy Games. Game Studies, 12, 1. www.gamestudies.org/1201/articles/tuur_ghys.
Hochberg, Julian & Brooks, Virginia. (1996). Movies in the Mind’s Eye. In David Bordwell & Noël Carroll (ed.), Post-Theory: Reconstructing Film Studies (pp. 368-387). Madison: University of Wisconsin Press.
Holding, Dennis Harry. (1985). The Psychology of Chess Skill. Hillsdale: Lawrence Earlbaum Associates.
Lefebvre, Martin. (2007). Théorie, mon beau souci. Cinemas, 17, 2-3, 143-192.
McCrea, Christian. (2009). Watching StarCraft, strategy and South Korea. In Larissa Hjorth & Dean Chan (eds.), Gaming Cultures and Place in Asia-Pacific (pp. 179-193). New York/London: Routledge. Retrieved July 8, 2013, from www.routledge.com/books/Gaming-Cultures-and-Place-in-Asia-Pacific-isbn9780415996273.
Mauco, Olivier. (2005). Les représentations et les logiques politiques des jeux vidéo. L'intériorisation des logiques collectives dans la décision individuelle. In Sébastien Genvo (ed.), Le game design de jeux vidéo : approches de l'expression vidéoludique (pp. 117-135). Paris: Harmattan.
Minsky, Marvin. (1975). A Framework for Representing Knowledge. In Patrick Henry Winston (ed.), The Psychology of Computer Vision (pp. 211-277). New York: McGraw Hill.
Neisser, Ulric. (1976). Cognition and Reality: Principles and Implications of Cognitive Psychology. San Francisco: W. H. Freeman and Company.
Peirce, Charles Sanders. ([1863] 1986). Chapter IV. The Conception of Time Essential in Logic. In Writings of Charles S. Peirce. A Chronological Edition. Volume 3. 1872-1878 (pp. 102-105). Bloomington: Indiana University Press.
Peirce, Charles Sanders. ([1868] 1991). Some Consequences of Four Incapacities. In James Hoopes (ed.), Peirce on Signs. Writings on Semiotic by Charles Sanders Peirce (pp. 54-84). Chapel Hill/ London: The University of North Carolina Press.
Peirce, Charles Sanders. ([1877] 1991). The Fixation of Belief. In James Hoopes (ed.), Peirce on Signs. Writings on Semiotic by Charles Sanders Peirce (pp. 144-159). Chapel Hill/ London: The University of North Carolina Press.
Perron, Bernard. (2006). The Heuristic Circle of Gameplay. The Case of Survival Horror. In M. Santorineos (ed.), Gaming Realities: A Challenge of Digital Culture (pp. 62-69). Athens: Fournos.
Rollings, Andrew & Adams, Ernest. (2003). Andrew Rollings and Ernest Adams on Game Design. Indianapolis: New Riders.
Sauvé, Louise, Renaud, Lise & Gauvin, Mathieu. (2007). Une analyse des écrits sur les impacts du jeu sur l’apprentissage. Revue des sciences de l’éducation, 33 (1), 89-107.
Sirlin, David. (2005). Playing to Win. Becoming the Champion. David Sirlin. 134 p.
Tardif, Jacques. ([1992] 1997). Pour un enseignement stratégique. L’apport de la psychologie cognitive. Montréal: Logiques.
TeamLiquid [wiki]. (2010, June 22). Terran Timing Push vs. Protoss. Liquipedia. Retrieved February 8, 2012 from wiki.teamliquid.net/starcraft/Terran_Timing_Push_vs._Protoss.
TL.net Bot. (2009, December 13). DQs: Yosh, RiboRibo, Scan, Mercury. TeamLiquid Forums. Retrieved February 8, 2012 from http://www.teamliquid.net/forum/viewmessage.php?topic_id=107728.
Voorhees, Gerald A. 2008. “Simulations of the self: Rhetoric, argument, and computer game criticism,” unpublished Ph.D. thesis, The University of Iowa.