Where No One Looks is a first-person game about a man who cannot leave his apartment, and the novel that will not leave him. I designed its loop, its rules and its inspection mechanic — then built the parts that could be proven in engine. This is the case, including the parts that did not work.

Most portfolio cases open with a finished game. This one opens with a problem: how do you build a game whose only verb is attention?

No combat, no traversal, no economy of loot. One apartment. One narrator who is lying to himself. A novel that keeps leaking into the room.

I wrote four constraints before writing a single mechanic, and designed against them for eight weeks:

One location, no loading screens.
If the player leaves the apartment, the game is over. There is no second map to hide behind.

No cutscenes.
Everything the story needs has to be visible in the room, at the player’s own pace.

The plot must survive a player who does not understand it.
Understanding is a second run, not a gate. Confusion is a valid state, not a failure.

Every system has to work without a writer in the room.
If a designer cannot author the moment alone, the moment does not ship.

Voices of the Void is a signal-hunting game, and the closest thing I found to a working answer. Its structure is two states, and the exchange rate between them is the design.

State A is work: scan the sky, decode the signal, manage needs.

State B is exploration: map the valley, find anomalies, spend the time you bought.

Work pays for automation, automation buys back time, time is spent exploring — and exploration gives you reasons to keep working.

Then the loop degrades, honestly and in the player’s favour. Automation shrinks the work half from the game, to a chore, to a button. Most games hide that. Voices of the Void lets you watch your own obligation become optional.

What I kept: the exchange rate between obligation and freedom, and the rule that exploration must be paid for rather than handed out.

What I refused: the grind as content, the sandbox without a thesis, and the fantasy of being a competent employee.

The question I carried into my own project: what happens to a game when work stops being content? In Where No One Looks there is no automation to buy. The work is looking — and it never gets cheaper.

A game about looking has to know what the player is looking for. I reduced the pitch to three feelings and turned each one into a rule that could be implemented or rejected.

«I am reading a place, not a plot.»
Rule: information is spatial — nothing is explained in text that could be found in a room.
In engine: inspectable objects; the word field replaces surfaces.

«The book is editing the room.»
Rule: every intrusion is coded — audio, light, material — and reversible in the player’s memory.
In engine: cel-shade pass, hatch stencil, light shafts.

«I am not alone.»
Rule: the Shadow is never shown. It only writes, and its handwriting is the most reliable narrator in the game.
In engine: word field layers, inscriptions on surfaces.

R01. The player never knows more than the protagonist.
If he is confused, the player is confused. That is uncomfortable on purpose.

R02. Reality and illusion are coded, never ambiguous.
A change in the room has a cause the player can name afterwards: a sound, a light, a texture.

R03. The illusion escalates in legible steps.
Beautiful nature, then castles, then dark decay. At any moment the player should be able to say which version of the world they are standing in.

R04. Contradiction is content.
Two sources disagree about the same event. The game never adjudicates. The player does — and the room reacts to the choice.

R05. The Shadow only writes.
Its inscriptions contradict the protagonist more precisely than any scene could.

R06. The apartment edits itself between visits.
Objects appear, move, disappear. Nothing is announced directly. The player’s memory is the only reliable source.

R07. Motivation is social, not systemic.
The sister is a relationship, not a quest marker. The game pushes through guilt, not through objectives.

The core verb is hold-to-inspect. Aim at an object, hold, and the room resolves into a field of words — the object’s history, the protagonist’s associations, the Shadow’s corrections. Release, and the room re-forms.

The word field is not an overlay on top of the game. It is the game’s only readable language: everything the player learns, they read in the room.

Three decisions shaped the implementation more than the art direction did. Each one is written down as a decision record, because the reasoning matters more than the result.

Words live in UMG, not in the post-process.
Text has to stay crisp, localizable and legible at any resolution. The post-process handles the strokes; the interface handles the language.

Depth of field instead of a screen blur.
A full-screen blur destroyed the words at the edges of the frame. Narrowing the depth of field keeps the text sharp and pushes the room back instead.

The hatch is a post-process pass driven by a cursor trace.
Custom stencil ½, triggered from a ~3 Hz trace. It costs almost nothing and survives the cel-shade pass, which is why the effect is affordable at all.

What exists is a test ground, not a game yet. The demo corner holds four systems: cel-shade with outline, the hatch and inspection post-process, light shafts built as billboards, and atmosphere — dust and a single moth.

No gameplay loop is implemented. The room is a test bench: it exists to prove that the look and the interaction can carry the fiction before any content is written.

Three things broke in ways I could not design around. They are here because a case that lists only successes is an advertisement, and because the fix is usually less interesting than the diagnosis.

Auto-exposure breathing.
The inspection overlay changed the average luminance of the frame, so auto-exposure compensated by pumping the whole image while the player held perfectly still. Freezing exposure fixed the breathing and killed the room’s light adaptation. Three possible fixes are documented; none is good enough yet.

Screen blur versus legibility.
Blur was the obvious way to focus attention and the fastest way to make text unreadable. Replaced with depth of field: the words stay sharp while the room falls away behind them.

Lumen and ray tracing.
On the target hardware the cost was not defensible for a game that lives in one room. Both are off. Cel-shade and baked lighting absorb the loss — but the decision limits what the light can do later, and I would revisit it before the vertical slice.