A guard must show what he knows

Stealth games sell a clean fantasy. Stay unseen, then strike. The fantasy collapses the moment a guard reacts to something you could not have known he noticed. Vision and hearing are not flavour. They are the rules of the puzzle, and a puzzle whose rules are invisible is just a punishment.

A readable guard has a cone you can learn, a sound that means suspicion rather than instant failure, and a recovery when you break line of sight. Teach those before you make them harsh. A game that opens on an instant-fail eyeline has skipped the lesson and kept the exam. Players do not feel outsmarted. They feel lied to. The later, crueller rooms only work if the first rooms made the senses legible.

Stealth games open from a position of concealment, and everything after that is a question of whether the concealment holds. Few genres carry a fantasy this clean: stay unseen, then strike from the dark.

The promise is that one person can outmaneuver an entire garrison without any of them ever learning the person was there. Spies, assassins, and now and then Batman.

Delivering that promise means running several fiddly systems at once. Enemy awareness, the information the player is allowed to gather, detection that holds up under scrutiny. One of them out of tune and the whole structure folds.

So the real design question is not how the player hides. It is what the guards are able to notice, and how they notice it.

A guard is built out of two fake senses: sight and hearing. Those are the human faculties worth modeling, and almost every stealth game models them in roughly the same way.

Sight is usually a viewcone, an invisible wedge of cheese fixed to the enemy's face. Step into it and you become detectable. In practice a single cone is too crude, since it lets a player stand shoulder to shoulder with a guard and remain invisible, so most games build something lumpier.

Splinter Cell Blacklist stacks three shapes. A narrow cone covers where the guard is looking, a much wider box stands in for peripheral vision, and a small pocket behind the guard reproduces that prickling sense that somebody is right over your shoulder.

Height matters too, and the answer depends on whether the game wants you perching above enemies. Cover is resolved with a raycast, a line drawn invisibly between two points to test whether anything blocks it. Splinter Cell refines this by firing rays at eight separate bones in Sam Fisher's model and only calling him spotted once enough of them come back visible.

Entering the cone without cover rarely means instant discovery. What happens instead is that a guard's awareness begins to climb, and only a full meter means the guard knows precisely where you are.

How fast it climbs depends on the situation. Distance slows it. So does sitting in peripheral vision, standing in dim light, crouching, or simply not moving at all.

Guards can register more than the player, too. An open door, a disturbed object, a body on the floor. That widens the toolkit for traps and distractions, and it sells the idea that these are people paying attention to a world rather than tracking a single target.

Hearing is the harder sense, because sound does not travel in straight lines through architecture. The usual starting point is simple enough: a noise gets a radius scaled to its volume, and any guard inside the radius is told to go investigate the source.

A gunshot, a creaking floorboard, a thrown stone. Each one carries its own reach. But drawing a straight line from the noise to the guard breaks the player's expectations immediately, since players assume walls muffle things.

The common fix is to reuse the pathfinding graph, the same data that lets enemies walk around the level without walking into furniture. Push the sound along those paths and it bends around corners and dies out through rooms, which is much closer to how sound behaves.

Perception can be layered further. Thief lets guards pick up secondhand knowledge of the player from what their colleagues are doing. Hitman 2 makes its enforcers far worse at recognizing Agent 47 when he has his back to them, which gives a disguise real value.

A perception system only works if the player can reason about it, so games surface it through interface, animation, audio, and guaranteed safe spaces. Done well, the simulation lines up with what people already know about light, shadow, and sound, and they can judge for themselves where it is safe to stand.

Some ambiguity always survives, and ambiguity causes arguments. Nearly everyone who plays these games has felt certain they were invisible right up to the moment a guard turned and saw them.

Interface feedback is the first lever. Thief's developers understood that a first-person view gives you no way to tell how brightly lit you are, so they put a light gem at the bottom of the screen showing current visibility. Splinter Cell reports how much noise Sam is making on his heads-up display. Most games also carry a detection indicator that mirrors the guard's awareness meter, sometimes pointing toward whoever spotted you.

Bodies and voices carry the same information. A guard slouching through an idle animation reads as switched off. One with a weapon up reads as hunting. Audio barks do the same job out loud, announcing the moment suspicion starts to build.

Then there are refuge spaces, spots where concealment is not a judgment call at all. The gargoyles overhead in Batman, the long grass in Assassin's Creed, the crates and cupboards in Hitman. They hand the player at least one place to stop, watch, and plan with nothing at risk.

The other major answer is deliberate bias: handicap the simulation so it agrees with the player's intuition rather than with physics. Splinter Cell Blacklist programmer Martin Walsh framed it directly. "It doesn't matter what the NPC can see or hear from a simulation perspective. It's what the player thinks the NPC should be able to see or hear."

That game halves a guard's hearing while the guard is offscreen, on the grounds that being caught by someone you cannot even see reads as a cheat. The Last of Us raycasts to Joel's head for line of sight, then drops the target to his chest when he crouches, so peeking over cover does not give him away.

Fuzzy detection is the largest concession of all. Being spotted the instant you brush a vision cone would feel arbitrary, and a short delay before full alert feels like a person registering something rather than a trigger firing.

The boldest option is to stop hiding the system and draw it. Mark of the Ninja, a side-scrolling sneak 'em up, leaves almost nothing to inference.

Vision cones are painted on the screen. The ninja is in shadow or in light, and the sprite itself tells you which. Noises bloom outward as fat circular pulses, and the game will even preview the radius before you commit, so you know in advance whether a distraction or an escape route is going to reach.

With everything visible there is nothing left to dispute. You were inside the cone or you were not. The sound reached the guard or it stopped short. That clarity is what lets the game get away with detection that is essentially binary, instant awareness with only a little analog softness at the edges of a cone.

Shadow Tactics: Blades of the Shogun grades it more finely, splitting each cone into three bands. Bright green near the face is the danger zone and detects fast. Dark green is survivable while crouched and fatal while standing. A dotted band marks refuge like bushes and high grass, where you stay invisible regardless. Trip a cone and it floods with yellow, and when the yellow reaches you, you have been seen.

Full 3D makes this presentation much harder, though not impossible. Metal Gear Solid sidesteps the problem by duplicating the level into a top-down radar and drawing the cones there instead. Deus Ex: Mankind Divided abstracts along similar lines. Sly Cooper goes physical, arming guards with torches that throw obvious yellow pools of light: inside the pool you are caught, outside it you are fine. Cartoonish, and instantly readable.

Hiding the system produces tension, because a player who cannot verify their safety has to keep estimating it. An analog, ambiguous perception model forces them to read the room the way they would read a real one, weighing light, shadow, distance, and noise.

Nobody ever reaches certainty, and that suits the fantasy. Stealth heroes are not powerful because they can win a fight. They are powerful because they can stay unfound, and keeping that status fragile is a constant reminder of how thin the advantage is.

Thief programmer Tom Leonard described the goal as "getting the player's heart pounding by holding them on the cusp" of discovery. Survival horror that borrows stealth mechanics depends on the same doubt. Alien: Isolation would be far less frightening if you could read exactly where the Xenomorph was looking, because most of the dread lives in not knowing what it can sense.

Showing the system produces confidence, and confidence turns the player from prey into predator. Once the rules are on the table, plans get elaborate: bait a guard into a trap, line up a route, close in for a silent kill.

Mark of the Ninja producer Jamie Cheng put the appeal plainly. "I wasn't nearly as interested in guessing whether a guard would hear me or not, and way more interested in creating an elaborate death trap."

Refuge zones, gadgets, and outright superpowers can deliver some of that feeling as well. But prediction is the shortest path to it. The more reliably a player can forecast what an enemy will perceive, the faster they stop guessing and start scheming.

Every guard sees and hears through the same kind of simulated senses, so the design choice is not really about the senses. It is about how much of them you expose.

Conceal the system and stealth becomes tense, brittle, and uncertain. Expose it and stealth becomes tactical, deliberate, and precise. Neither is the correct answer in the abstract. It depends on whether you want the player to feel hunted, dangerous, cautious, clever, or some mixture that shifts as the level goes on.

The hard part is not making guards perceive things. It is choosing what the player gets to know, what they are left to infer, and how fair it feels the moment a head finally turns. That is not a decision you can reason your way to on paper. Build the guard, get the level running, walk past it, and see what the moment feels like from inside the game.

More on this: Hitman's Sapienza expects the first run to fail

In the Flockbay app, ask for one guard, a vision cone you can see, and a noise that alerts him without ending the scene. Play until you can predict the cone. Only then ask for a second guard. If the first cone is still a surprise when it catches you, it is not on screen enough.

A stealth-action template and the AI stealth game maker are the right scale. One corridor.