Lesson: The Problems of Modern Stealth Design, and How Invisible Inc. Solves Them

Recently, I played through Shadow Tactics: Blades of the Shogun. Since then, I've been contemplating a problem I noticed during my play through. I didn't enjoy my time with the game. In fact, I haven't really been having fun with stealth games for a while now. But I love stealth. The thrill of being aware of more than others because they don't know you're there is a core gameplay fantasy of almost every stealth game, and it appeals to me on a fundamentally personal level, and the idea that I can't have fun with my favorite gameplay fantasy is a big problem. I started to ask questions, because I shouldn't love stealth games and not enjoy them at the same time. Those two thoughts should be mutually exclusive. I even started drawing up a design document for a stealth game of my own to see if I could design a stealth game I would enjoy (I think it's pretty great). Then I remembered Invisible Inc which I remembered enjoying a surprising amount. So I went back and played it, then I reflected. Over the course of my pondering and research on stealth I have learned many things. Foremost, that the stealth genre is quite stagnant, and it hasn't really changed since The Class of 98' (Tenchu, Thief, and Metal Gear Solid) introduced the modern stealth formula. The formula has been refined and cleaned up in games like Klei's Mark of the Ninja, but the issues haven't been fixed, only mitigated or ignored, despite the existence of games like Invisible inc. which demonstrate solid solutions to all of the problems. Stagnation is a problem in it's own right, but more fundamentally, I don't think modern developers should simply accept the "wisdom" of games created twenty years ago without asking some serious questions about their design. Preparing for this article, I had two driving questions. First, what is it about modern stealth games that I don't enjoy, and second, How would I fix them?

Stealth's problems can generally be traced back to the lack of a strong fail state for the players. This lack of a strong fail state leads to the players time being wasted and frustrating experiences with the AI. All three aspects are inextricably tied into one another. As such I'll begin with fail states and how they cause the other two problems, then return to them to close out the section.

Failure States in stealth games, are the primary source of much of the problems that are present in the genre. Most stealth games have a three step fail gradient. They begin in "Hidden mode", once players are spotted "search mode" starts, and they end in "dead or captured". Generally speaking, the searching mode is a soft fail state where you're given an opportunity to return to being "Hidden" before you get caught or die. In most cases, your only recourse during the "search mode" is to hide for a specified amount of time. This is where the problem begins. Most fail states lack verisimilitude and create either frustration in the player because they are boring and waste the players time, or are immersion breaking because the guards don't behave in a realistic manner. The current wisdom in regard to stealth games, which causes these problems, is that they're fundamentally a "traversal puzzle" where the player tries to get from A to B by manipulating guards with various tools (throwing rocks, etc), but this is where the problems really begin.

The idea that stealth games are actually just puzzle games with a "stealth" aesthetic sounds good on paper, but in practice, it is impressively restrictive. It really only applies to a linear experience, where you're at A and must get to B. It only allows one outcome, and it limits how a developer and player are allowed to think about their stealth. It makes it a much less elegant fantasy, and just turns it into a puzzle game where once you've found the solution, the only barrier is execution, which is problematic on two main levels. First, it creates an illusion that players must be given all the information in order to understand the problem and devise a solution (like a puzzle game). Generally this means a player gets to sit on a hill with a set of binoculars marking enemy positions so that when they try and execute their plan, they have all the information necessary to do so. Secondly, it is difficult to balance. If the player is not particularly skilled, execution becomes frustrating. If a player is skilled, gameplay feels tedious and boring. Even games that are seen as examples of outstanding stealth design such as Mark of the ninja adopt this mentality. Despite the developers' careful consideration of these problems, Mark, for me at least, was honestly quite a boring eight hours of play. The game was so desperate to make me feel like a ninja, that it failed to make me feel like I was really in any sort of danger. Especially once I'd unlocked the "stealth suit" and was able to literally sprint through levels without fear of being heard. Running from A to B isn't really any more fun than walking. In a more challenging game, like Shadow Tactics, where executing a plan is quite difficult, frustration becomes the core experience rather than boredom. Ideally there's a middle ground that is fun, but I've found that more often than not, it's like trying to balance on the edge of a knife. You're mostly on one side or the other, and either way you're going to get cut. The idea that stealth is just traversal is at best reductionist, and at worst, anti fun.

The other consequence of this design philosophy is in the AI. Anyone who's played a modern stealth game is likely familiar with the short sighted enemies who can't bend at the waist. The standard justification for this is that players want "predictable" rather than "realistic" AI. The problem with this rationale is that it assumes that "predictable", and "realistic" are mutually exclusive terms. People can be predictably manipulated with distractions, but they also learn. The only reason a player would predict a guard to have short term memory loss is if they'd become accustomed to it from previous experience. The idea of "predictability" in AI design should align with a players perception of reality. The worst examples of this are after a player has entered the "spotted" state and the only way to continue progressing (in pure stealth games where fighting back is not an option) is to either restart from a checkpoint, run, or hide. Which sound reasonable on paper, but there are two big caveats. The first, is if every enemy on the map instantly knows about the players position. It makes no sense, and is simply no fun and frustrating. The second is when the act of hiding brings with it no tension because guards are programed to be unable to look under a desk. These two occurrences can shatter a games world. The player is immediately made aware that they are playing a game and that they aren't actually a sneaky ninja. When immersion is broken, the players fantasy is broken, and they become aware of the fact that the game is cheating on their behalf; any sense of challenge is lost, and the fantasy is ruined. That's not even mentioning the suspension of disbelief required to accept that every guard is basically shortsighted and unable to distinguish basic shapes at a distance of twenty feet.

Now back to fail states. Some people may say that hiding spots are necessary for the player so that they can return from a "searching" game state to a "hidden" one, but the only reason that that would be true is if you insist that stealth is traversal. There is no real reason a player has to return to a totally hidden state, as if the AI forgot they'd spotted an intruder a minute ago. Mark of the Ninja's own Nels Anderson has stated that there should be "limited consequences for failure", but that only works if your fail state works. If you're punishing a player by wasting their time, then it is bad design. It doesn't matter how much time you waste. It isn't fun. Stop it. If the player made a mistake, they should have to deal with it and learn to adapt to the new state of enemy awareness. They really shouldn't even have the option to restart from a checkpoint. Just as you shouldn't waste a players time as a developer, you should also stop the player from wasting their own time, or they will. Players can easily ruin their own game experience. If you insist that they learn from their mistakes and grow as a player, they will experience a sense of accomplishment and enjoy the game that much more. The easiest way to avoid degenerative strategies is to simply make them impossible. But your fail state needs to be improved in order for that to work.

Which brings us to the second part of this article where I explain, by using Invisible Inc. (Invisible) as a proof of concept, how to solve the aforementioned issues. Invisible, while not a perfect game, may just be the best stealth game made in quite a long time. By incorporating a new style of fail state, never wasting the players time, and implementing realistic AI, Invisible Inc. creates a stealth gameplay experience like none other.

First off, Invisible's fail state is represented by the "security level". It doesn't even really function as a scale. The compound becomes aware of the players the second they enter, and from there, it slowly becomes more and more aware of them over time. No going back. No hiding. Small mistakes such as getting spotted or killing guards bump the "security level" up a bit faster than normal. That's it. This has several effects, but I'm going to focus on the problems discussed previously and how it solves them. First, it allows the game to be perfectly balanced. Second, it allows the AI to be realistic.

In Invisible, difficulty scales with time. The player can stay in a level as long as they want, or leave the second they find the exit. The longer they stay, the harder it gets, and they run the risk of being killed. On top of this, Invisible offers a fully customizable campaign difficulty. For players who want less of a challenge, they can tune the game to suit their preferences, and those who want a grueling experience can do likewise. By default, the game never wastes the players time, and doesn't even allow players to replay failed missions on the harder difficulties. All of these features keep the game feeling fresh, and because the difficulty always increases, the challenge is always present, regardless of skill. To add to that, Invisible is a turn based strategy game. It isn't about traversal, it's about resource management. The players ability to execute a strategy is limited only by the resources available to them. As a mission progresses, resources become increasingly scarce, making escape more difficult. But because the player controls how much challenge they desire, they're never really frustrated by the game.

The AI further helps the game by never breaking player immersion. Once a guard has been made suspicious, by whatever means, they stay suspicious. They no longer patrol along their predetermined route, they begin wandering around to seek out the player. The player's only option is to avoid them. They can hide behind cover, but the guard has only to walk over to the cover and look behind it to find the player. At which point, the player can take more aggressive measures, or find some other means of avoiding detection. There are many tools at their disposal, and the AI will constantly pressure the player to use them in the most efficient way possible. Along with that, guards never become aware of the player without a reason. Once you're spotted by a guard, they'll shout and nearby guards will turn to investigate the shout. But the range on this is very limited and doesn't break the illusion of reality that the game creates, since the effect will never affect a guard beyond the current room. On top of this, guard behaviors are always predictable. They will always investigate sounds, and resources can be spent to predict a guards movement so that you're never caught off guard. The AI is realistic and predictable.

All that said, the fail state is what allows all of this wonderful design to shine through. Invisible Inc's "security level" while functionally replacing the "soft" fail state of other stealth games, doesn't actually ever end in "player death". The true fail state of a campaign is a "team death", which is permanent. Invisible gets around this rather severe punishment by giving you more than one character to control at the start of the campaign. If one dies, you have a limited set of options to save them, but you can always leave them behind. If you do, you'll likely have the opportunity to recover them or someone new later on. You have to lose all of your characters to truly fail the campaign, which requires a significant amount of player neglect to do. Invisible is a game that never wastes the player's time. It has alternate consequences that are far more engaging. They force the player to think and adapt to the changing situation, instead of just turning back time. Which is what a stealth game should always aim to do.

I love stealth. Being a sneak, whether an evil assassin or humble thief, is among the best fantasies that games can create. Recently, most developers seem to have lost sight of how to make that fantasy come to life. I hope that this essay has shed some light on the problems and solutions to many of the stealth genre's most pressing issues. I know the solutions Invisible inc. presents will not serve every design space perfectly, but I encourage developers to explore solutions to these problems instead of falling back on old standards. Most importantly, I hope it will encourage developers to think about their design choices more critically rather than simply accept industry "wisdom" when it is no longer wise. My name is Brock, and this has been a Lesson on the problems of modern stealth design and how Invisible Inc. finds solutions to them. Thank you for reading.