The Architecture of Stealth: Notes from the greybox

The first stealth space I ever built got solved from a balcony I forgot I'd made.

It was a warehouse. Two floors, four patrols, a crane gantry running down one side that existed because I needed a way to get the player from the office up to the catwalk and the gantry was the cheapest thing to block out. I'd spent most of two weeks on the floor plan. Patrol timings tuned to the second, a nice little overlap where two guards cross and open a three-second window, a shortcut through the loading bay for anyone who

thought to look.

First playtester walks in, climbs the gantry, stands there for ninety seconds, and then walks the entire level without stopping. Doesn't use the window. Doesn't find the loading bay. Doesn't need to. He'd seen everything from up there, decided everything from up there, and the rest was just admin.

I watched the capture back that evening and I could see the exact moment my level died. It wasn't when he climbed the gantry. It was four seconds after he got to the top, when his mouse stopped moving, because he'd finished reading the room.

That's the whole discipline, really. Everything I know about stealth level design is a consequence of that four seconds.

The room is the opponent

Here's the thing that took me embarrassingly long to internalise. In most of the games I'd worked on before, the level was a container. The encounter happened inside it. You could pick the fight up, drop it in a different room, retune the cover, and it would survive the transplant. The room contributed atmosphere and a couple of angles.

Stealth doesn't work like that. The guards are furniture that moves. What actually beats a player is the building — where the sightlines fall, how many doors the room has, what the floor's made of, whether there's a second storey. Swap the room and you haven't retuned the encounter, you've deleted it. This is the only genre I've worked in where the level is the design rather than the stage the design happens on, and it's why stealth LDs are so insufferable in meetings. We're not decorating your fight. We are the fight.

Where the player stands to think?

I used to think about cover. Now I think about where the player stands to think, which turns out to be a completely different question and a much better one.

Steve Lee describes the loop as observe, plan, execute, react, and if you've watched enough footage you can see it happening in the input. There's a rhythm to a player who's engaged with a stealth space: movement, stop, camera sweep, movement. The stops are the game. If you record a session and mark every place the player stood still for more than about three seconds, you've mapped the actual structure of your level, and it usually isn't the structure you drew.

Ben Bauer calls the failure mode the vantage point problem, and my gantry is the textbook version: elevation, cover, and clean lines into the space, all in one place. Give a player that and they'll strip the encounter for parts from a chair. What's useful about Bauer's fixes is that every one of them is architectural rather than mechanical. Put glass in front of the vantage so they can look but not shoot. Take the cover away so standing there is expensive. Restrict which enemies can even be reached from it, as long as the geometry has an honest reason. And the big one: break the space up so no single position reads the whole floor.

That last one is the fix I actually use, and it pays twice. Split one big prospect into four small ones and the player has to run the observe-plan-execute loop four times instead of once, which is four times the game out of the same square footage. It also lets you stage your spawns, because the player can never see the whole room, so they can never audit your population. I've had tech leads thank me for a layout decision I made for entirely aesthetic reasons.

But you can't just delete the high ground either, and this is where I've watched people overcorrect. Take away every safe place to stand and you take away the observation phase, and without an observation phase there's no plan, only reaction. Which is a shooter with a slower walk speed. You want the player to have somewhere to be smug. You just don't want it to be one place with a view of everything.

Count your routes on paper !

Bauer's other named failure is the avoidance path: one route through the layout that nothing ever contests. It might be long. It might be clever. Doesn't matter — once anyone finds it, the level is over, and someone always finds it, and then it's on YouTube.

His minimum bar is that every path gets seen by somebody at some point. Not every corner swept. The main lanes contested, occasionally, enough that you have to stop and wait and consider.

His numbers are the most directly useful thing in the whole talk and I've written them on a whiteboard at three different studios. Three to five main routes. Two or three at ground level, roughly the same again above and below. Plus small connectors so players can hop between mains, because that's where the improvisation lives. I check this on paper before I open the editor: top-down, trace every viable line, count them. If I get two, it's a corridor with a detour. If I get eleven, I've built a park.

The counterpart is Dana Nightingale's rule from the Clockwork Mansion, which she mentions almost in passing: never more than three ways in and out of a room. It sounds like fire code. It's the same problem from the other side. A room with five exits is a room where a guard can't be cornered and a player can't be trapped, and nothing that happens in it can carry any weight at all. Space makes stakes by taking options away.

Three, not one.

The plan lies to you

Everyone designs in plan because plan is what the editor shows you when you hit the top-down key. Stealth lives in section.

I now sketch section first, and it changed my hit rate more than any other habit. The rooftops in Thief, Blink in Dishonored, the rafters and vents in the Arkham predator rooms — that verticality isn't traversal flavour, it's the third axis of the sightline problem and it's how players buy time. Lee is blunt about the causal chain and he's right: vertical layouts plus mobility gets players high, high gives them perspective, perspective gives them time, time is what makes deliberate play possible. Take the roof off and you've removed the planning phase without touching a single guard.

It also means stealth levels can be tiny, which is the best news in the discipline. Lee's Half-Life 2 map — the one he made in fifteen days for the Arkane application — packs about ten minutes of play into a square you could sprint across in thirty seconds, because the player crosses it four times at three heights. Dense folded circulation beats sprawl, costs less, and looks better in a screenshot. I have never once regretted making a stealth space smaller.

Floors are a system

The best purely architectural idea the genre has produced is in a game from 1998 and almost nobody has copied it.

Thief made building materials into mechanics. Carpet swallows footsteps, run all you like. Tile is a permanent liability: no matter how slowly Garrett walks, tile makes noise, always, and you route yourself through a mansion partly by reading what it's built out of. Then the water arrows, which put out torches and let you pour darkness into a room. You don't find shadow. You manufacture it. Light is a substance you can remove from a space, which makes the player something closer to a contractor than a burglar.

I've tried to get material-as-system into two projects and lost the argument both times, and both times for the same reason: it's an audio and materials dependency that has to be locked early, and it doesn't demo. You can't show it in a vertical slice. You can only feel it in hour three.

Set it against tall grass. Grass is a material with no properties. Nothing you can do to it, nothing to exploit, nothing to ruin, and standing in it renders you invisible to a man eight feet away. It's the anti-Thief: architecture reduced to a boolean. Watch any stream of any open-world game with a stealth mode and count how long people spend crouched in shrubbery looking at nothing. That's the design working exactly as specified. It's just that the specification is sad.

Buildings that fight back

The Clockwork Mansion is the maximalist version: make the house the antagonist. Walls slide, rooms recombine, the ground under you isn't yours.

The detail from Nightingale's interview I keep coming back to is that the clockwork soldiers were originally hostile from the start, and it was too much, so the level now begins neutral. The building is polite before it turns. That's a pacing decision made at the level of the architecture itself, and I don't know anywhere else that's true.

The other thing worth stealing is that she prototyped the transformations before there was any layout at all. Which is why slipping through the walls mid-transformation was never a bug — every prototype was an actual space with an inside and an outside, so the gap between states was a real place the whole time. Prototype the verb, then build the house around it. Do it the other way and you spend six months papering over the seams.

Set your metrics before you draw anything

Architecture is calibrated to a body. Door heights, stair risers, ceilings, handrails, the whole discipline is sized to a person, and you only notice when it's violated.

Same for us, and the mistake I see most in junior work is drawing the space before locking the metrics. How high can they jump, how far can they see, how fast do they walk crouched, how wide is the cone. Every one of those numbers is a dimension in your building. Change the walk speed by fifteen percent and every room you've built has silently retuned.

Styx is the interesting case because they changed the body. The goblin is half a man's height, deliberately, so the weakness reads in silhouette before he's done anything — but the second-order effect is what it does to the architecture. A pot becomes a room. A grate becomes a door. All the void space that a human-scaled building doesn't consider habitable turns into circulation that the guards don't have access to. One decision about a character model rewrote every level in the game for free.

Then there's the Arkham gargoyle. Who commissioned these? Who bolted them up, at that spacing, at that height, in a chemical plant? It's a joke but it's also the real critique: when a building's features exist only to service the mechanic, the building stops being a place and the infiltration fiction quietly becomes a jungle gym. It works in Arkham because those games are honest about being playgrounds. Put a gargoyle in Thief and the whole thing falls over.

Watch other people play, because you can't see your own level any more

The single most useful thing I do is watch strangers play. Not for the bugs. For the four seconds.

Bauer names the reason and it's the thing every LD refuses to believe about themselves: you play your own level so many times that you go myopic on it,. and then you defend asetup that nobody outside the building can beat while insisting it's obvious. He's blunt that you shouldn't let your team dismiss playtest results and call the testers idiots. I have been the person calling the testers idiots. I was wrong every time.

Streams are the cheat code here, and specifically streams rather than telemetry, because streamers talk. They narrate the plan out loud before they execute it. That is free intentionality data and there's nothing else like it. If a player can say "right, I'm going up the pipe, waiting for the smoker to turn, then through the window" — your affordances landed. If they can only say "I dunno, let's try this" — they landed nothing, and no amount of heatmap is going to tell you that, because the heatmap shows the same path either way.

The corollary is the old principle, show the door before the key. Doom's coloured locks work because you meet the lock first and the lock writes the objective. Lee does it three times in that Half-Life map: shows you the gravity gun through a doorway you can't reach in the first minute, and that memory is what turns a route into a discovery. It sounds obvious right up until you build a non-linear level and accidentally hand out the key on route B, and then watch someone open the door on route A with no idea they'd solved anything.

And Bauer's fair-warning rule, which I'd tattoo on people: if there's a dog behind the right-hand door, the player hears it barking. Then if they open it anyway, that's on them, and they know it's on them, and they'll laugh. A blind choice where one option is death is just mean, and worse than mean, it teaches the player to stop choosing.

Why we don't build these any more

Everyone frames the decline as an AI problem. Guards who forget a corpse in thirty seconds, patrols that don't exist, men studying a wall while you jog past.

It isn't AI. Look at what has to be true for one of these rooms to work. Three to five contested routes across three elevations. Materials with behaviour. Prospect chopped into pieces so no position reads the floor. Sightlines audited by hand, in engine, by a person, repeatedly. Metrics locked before anyone draws. Affordances that never lie, verified against strangers, because the author went blind on it in week two.

None of that is visible in a trailer, all of it is slow, and a room built for a firefight and a room built for an infiltration are genuinely different buildings. You can't get both by adding a crouch button and some shrubs. Which is exactly what an open-world camp is: a space generated at scale, asked to serve every play style, serving none.

The instructions aren't lost. Bauer published a complete, practical, free set of them, from a man who shipped these levels for a living. It's just that they describe a very slow, very unglamorous way to build a house, in an industry that would much rather generate a landscape.

Three doors, though. Three doors is free.

-----

Why Go through this research?

Koel: spy Cholan SEA is one of my Passion projects set too debut on IGDC 2026 this october. As a spy sent to do biddings of prirate crew, you shape his island sitting on edge of the cholan sea serving as gate waya to South East Asia (SEA) and Easstern world. Inspired by Hitman and eailer Assassin's creed games i wanted to build a world that reacts to the players action, providing resistance and ta wind as the same. Even Playing all those games i inspied me is not enought to create something that is truthful to the genre and its auidence. This deep dive is important to achieve even afravtion of it.