Guide

Stealth game design explained

You memorize a guard’s turn, slip through a vent, and reach the objective without anyone knowing you were there. That quiet triumph — not the body count — is what stealth games sell. Titles from Metal Gear Solid and Hitman to Dishonored and Alien: Isolation share a spine of information asymmetry: the player plans with more knowledge than NPCs have about the player, until a mistake collapses the gap. This guide covers genre contracts and subgenres, the observe-plan-execute-exfiltrate loop, multi-route level architecture, ghost versus lethal scoring, alert escalation and recovery windows, social disguise layers, gadget economies, a Harbor Dockyard worked example, subgenre decision tables, common pitfalls, and a production checklist. For the underlying visibility, sound, and detection systems, start with our stealth mechanics primer; for systemic emergent play, see immersive sim design and tactical shooter fundamentals.

What defines a stealth game

A stealth game makes avoidance the primary verb. Combat may exist as a failure state, a costly escape, or a deliberate loud path — but the fantasy is infiltration, not domination. Three promises anchor the genre:

Readable risk — players can predict detection before committing; surprises come from their choices, not hidden dice.
Route multiplicity — at least two viable paths to the objective reward different play styles (roof, sewer, disguise, distraction).
Meaningful silence — staying unseen changes outcomes: score, story flags, loot, or world state — not just a cosmetic medal.

Stealth-first vs stealth-optional

Stealth-first titles gate progression on infiltration skill: getting spotted often fails the mission or triggers overwhelming force. Stealth-optional games — many action-adventures and immersive sims — offer ghost routes alongside loud ones. Design differs: optional stealth must not punish players who never learned the systems, while stealth-first can teach through failure. Document which camp you are building before tuning alert responses.

Subgenres: infiltration sandbox, survival hide, and squad breach

Pick your subgenre early; it drives camera, lethality, and level scale.

Infiltration sandbox (assassination puzzles)

Large, clockwork levels with social stealth, disguises, and environmental kills. Players replay for ghost ratings and creative solutions. Examples: Hitman, Mark of the Ninja (2D variant). Success metric: undetected completion plus style bonuses.

Cinematic infiltration (story missions)

Linear or hub-based missions with gadgets, codec hints, and set-piece boss arenas that allow stealth or combat. Examples: Metal Gear Solid, Splinter Cell. Success metric: narrative completion with optional rank tiers.

Survival horror hide

One overpowering pursuer; stealth is terror management, not puzzle optimization. Limited saves, narrow hide spots, and audio cues dominate. Examples: Alien: Isolation, Outlast. Success metric: survive traversal segments with scarce resources.

Tactical squad stealth

Top-down or first-person planning phases followed by simultaneous execution. Lethal and non-lethal loadouts, breaching tools, and synchronized takedowns. Examples: Desperados III, Shadow Tactics, Door Kickers. Success metric: clean plan execution with optional challenges.

Immersive sim stealth routes

Stealth as one systemic path through a simulation-heavy world — powers, physics, and faction reactions all interact. Examples: Dishonored, Deus Ex, Prey. Success metric: mission success with chaos or body-count scoring affecting later levels.

The core loop: observe, plan, execute, exfiltrate

Every stealth mission is a compressed heist screenplay:

Observe — gather intel: patrol routes, camera arcs, disguise rules, objective location, exfiltration points.
Plan — choose a route and tool loadout; mark waypoints mentally or via in-game planning mode.
Execute — move through the level, spending gadgets and timing distractions.
Exfiltrate — leave the space without raising permanent alarm; extraction is part of the puzzle, not an autoplay cutscene.

Pacing lives in the ratio between observe and execute. Horror hide games stretch observation through dread; sandbox assassinations reward long recon with bigger payoff moments. If execution exceeds 70% of mission time without new information entering the loop, players feel they are walking simulators with guard RNG.

Intel delivery without hand-holding

Surface intel through diegetic channels: overheard conversations, readable schedules on clipboards, map screens in guard rooms, camera feeds on security desks. A codec or radio can supplement but should not replace spatial learning — players who study the level should outperform players who skip to waypoint markers.

Level design for multiple routes

Stealth levels are clockwork dioramas. Guards patrol on loops; doors connect zones with different security postures; vertical layers (floor, ceiling crawlspace, roof) multiply paths without multiplying NPC count.

Zone security tiers

Partition maps into public, restricted, and high-security zones. Each tier changes disguise validity, camera density, and alert response. Players should feel escalation as they penetrate inward — not a flat difficulty number.

Chokepoints and bypasses

Every chokepoint needs at least one bypass: vent, keycard stolen from a patroller, distraction lure, or timed gap in a paired guard rotation. Chokepoints without bypasses become combat gates in disguise and break ghost runs.

Verticality and occlusion

Rafters, catwalks, and sewer tunnels let designers add routes without widening the horizontal footprint. Occlusion matters more than polygon count: if players cannot tell whether a guard sees them, the level fails regardless of art quality. Use light pools, waist-high cover, and sight-line debug views during blockout.

Ghost scoring, lethality, and player expression

Stealth audiences replay for self-imposed constraints. Score systems should track dimensions independently:

Detection — spotted, identified, recorded on camera.
Lethality — kills, knockouts, accidents attributed to the player.
Evidence — bodies found, weapons left, doors forced.
Style — target eliminated via environmental accident vs direct action.

Separate axes let a player kill everyone but never be seen, or ghost with only tranquilizers — different fantasies, both valid. Avoid collapsing everything into a single binary “S rank” unless your campaign is short and arcade-like.

Non-lethal must be fun

If tranquilizers are strictly worse than bullets in every dimension (slower, shorter range, no environmental synergy), ghost paths become punishment. Give non-lethal tools unique advantages: silent takedowns from cover, reusable sleep darts with crafting, or hostage negotiation outcomes unavailable to lethal players.

Alert escalation and recovery windows

Alert states form the tension curve. A typical ladder:

Unaware — routine patrol; player has full information advantage.
Suspicious — guard investigates a stimulus; localized search cone.
Alert — sector lockdown, backup called, cameras active.
Hunt — full map pressure; stealth may still be possible but costly.

Recovery windows are non-negotiable. After a suspicious search times out, guards must return to predictable routes — otherwise one footstep error ends the mission permanently. Horror games invert this: recovery is rare, so detection spikes dread instead of tactical repositioning. Match recovery rules to subgenre.

Tie escalation to AI perception with shared stimuli: noise events, last-known-position markers, and communication between guards. When guards radio each other, players learn to isolate witnesses first — a stealth puzzle emerges from systems, not scripted triggers.

Social stealth and disguise layers

Physical hiding (cover, shadows) and social hiding (looking like you belong) solve different problems. Disguise systems need explicit rules players can master:

Territory tags — chef outfit valid in kitchen, invalid on rooftop.
Proximity checks — named NPCs who know faces, generic crowds who do not.
Behavior scrutiny — running, carrying weapons, or entering forbidden rooms raises suspicion even in uniform.
Duration limits — disguises that expire push players forward instead of camping.

Social stealth shines in crowded spaces where physical stealth is impossible. Pair with faction reputation in immersive sims: uniforms borrowed from one faction anger another when discovered.

Gadget economy and distraction tools

Gadgets are puzzle pieces, not infinite win buttons. A healthy economy tracks:

Inventory caps — coins, bottles, and EMP charges are scarce; players choose where to spend.
Noise vs utility tradeoffs — a distraction lure pulls guards but creates a new stimulus to investigate.
Environmental affordances — chandeliers, fuse boxes, and leaking pipes are reusable level gadgets that do not consume inventory.
Carry restrictions — large weapons break disguise fantasy; store gear in lockers near infiltration points.

Playtest gadgets in isolation first, then in combination. The classic failure mode is a coin plus sleep dart plus vent route that trivializes entire wings without teaching new skills.

Worked example: Harbor Dockyard (infiltration sandbox)

Imagine Harbor Dockyard, a single large assassination-sandbox map with three difficulty modifiers:

Objective — eliminate a smuggler boss during a midnight container transfer without triggering coast-guard backup.
Route A — Social — steal a forklift operator uniform during shift change; drive through the loading zone to plant a tracker on the target’s briefcase; environmental crush kill when cranes swap containers (accident bonus, zero personal evidence).
Route B — Vertical — climb exterior scaffolding, cross roof vents, drop into the crane cabin for line-of-sight knockout, hide body in a sealed container marked for export.
Route C — Loud fallback — sabotage fuel drums to lure guards, fight through with suppressed pistol; coast guard arrives in 90 seconds, forcing fast exfiltration by stolen speedboat — mission completes but ghost ratings fail.

Security tiers: public pier (no disguise needed), restricted yard (uniform or forged pass), high-security crane platform (face recognition unless operator disguise plus borrowed RFID). Three chokepoints each have documented bypasses tested in greybox before art. Post-mission debrief shows four independent scores: unseen, non-lethal, no evidence, accident kill — encouraging replays without forcing perfection on first run.

Subgenre decision table

Goal	Favor	Avoid
High replayability	Infiltration sandbox, multi-axis scoring, accident kills	Single-solution corridors, binary pass/fail
Narrative blockbuster	Cinematic infiltration, codec intel, boss stealth arenas	Instant fail on any sighting without recovery
Pure tension	Horror hide, one apex predator, scarce resources	Gadget spam, loud combat as equal path
Hardcore tactics audience	Squad stealth, planning phase, simultaneous execution	Real-time-only with no pause planning
Systemic emergent stories	Immersive sim routes, faction reactions, chaos scoring	Scripted guard telepathy, locked doors without keys

Common pitfalls

Omniscient guards — spotting through walls or instantly knowing your location after a noise event breaks trust; use last-known-position searches instead.
Binary fail states — mission over on first alert teaches reload scumming, not mastery; offer recovery or loud escape paths.
Disguise without rules — uniforms that work everywhere remove the social puzzle; uniforms that fail randomly feel rigged.
Garbage disposal fantasy — hiding bodies in closets without consequences removes evidence tension; bodies should be found on timers or by patrols.
Stealth tacked onto action levels — arenas built for cover shooting rarely offer roof vents or distraction chains; blockout for stealth first.
Minimap wall hacks — showing all guards always collapses observation phase; intel tools should be earned or localized.
Identical AI loops forever — guards that never react to missing colleagues feel like furniture; implement buddy checks and escalating searches.

Production checklist

Document subgenre contract (sandbox, cinematic, horror hide, squad, immersive sim route).
Blockout level with three documented routes before vertical art slice.
Define security zone tiers and disguise validity matrix per zone.
Implement alert ladder with timed recovery to unaware patrols.
Wire shared stimuli between stealth mechanics and AI perception.
Build independent score axes: detection, lethality, evidence, style.
Prototype one distraction gadget and one environmental kill per wing.
Playtest ghost run without weapons; lethal run without detection; both must complete.
Add exfiltration objective equal in weight to infiltration objective.
Record patrol routes in a designer spreadsheet; verify loop lengths match audio pacing.
QA with vision-cone debug overlay; fix any blind spotting or false negatives.
Ship post-mission breakdown UI so players understand why ranks changed.

Key takeaways

Stealth games sell information advantage and quiet mastery, not DPS.
Every chokepoint needs a bypass; every mission needs a recoverable failure window unless you are making horror.
Score axes should separate detection, kills, and evidence so players express different fantasies.
Social and physical stealth solve different problems; disguise rules must be learnable.
Exfiltration is part of the puzzle — leaving clean is as designed as entering unseen.