Welcome Back

The first decision was structural. Rather than build five game sites, I built one machine and made every game an application on it. The desktop is the navigation. The browser inside it is the content layer. The worlds are apps.

This is the choice the whole project rests on. It gives five unrelated games a reason to sit together, and every new world inherits an audience already sitting at the machine.

It also set the tone. No onboarding, no tooltip, no press start. You are handed a room and left to work out what is clickable.

My favorite thing in the whole project is also the smallest. The login screen asks for a password, and there is no correct one. Whatever you type opens the machine. Because you had to guess at it, getting in feels like getting in, and people reliably decide they worked it out. Nothing on screen ever admits otherwise, because explaining it is the only way to spend it.

There is no right password. That is the whole trick.The first thirty seconds

Pick a resident, move into a starter house, earn Maple Bucks, decorate rooms, shop, play the arcade and read the town paper. There is no plot and nothing to save.

Nine districts, each a painted scene. Fifteen residents with their own conversations, friendship, gift preferences and letters they mail you. A town hall, a bank, a post office, shops, a cafe, a school, a newspaper and a six cabinet arcade.

The progression is deliberately domestic. You are not leveling up. You are finishing a sticker set, fishing at Lakeville, sitting a class at the academy, eating something out of your own pantry, and noticing the shop stock has changed since yesterday. It resets when you shut the computer down, on purpose. It is a day at the family computer, not a save file.

A cluttered photographic scene, a rhyming clue on a strip below it, and every object named in the rhyme hidden somewhere in the picture. Click it and the words strike through. One round per scene, finished when the rhyme is empty.

The clock is sized to the work rather than fixed: twelve seconds to read the rhyme, plus a per object budget by difficulty, so a scene stays fair when it gains or loses an object. A wrong click costs two seconds. A hint costs five and rings the object. You finish on stars, not a score.

It looks nothing like the town on purpose. Dark stained wood, brass, parchment, and an inscriptional Roman serif.

Four of the six rooms are the house I grew up in. The bedroom, the kitchen, the office and the garage, rebuilt from photographs down to the wallpaper border, the tile, the bookshelves and the cat. Nothing in the game says so. The riddles never mention whose house it is and my name is nowhere in it. It is simply the most personal thing in the project, sitting in plain sight as a puzzle about noticing things.

The office is the one that closes the loop. There is a computer on that desk, and it is the computer this whole project is built inside.

The photographs sit inside a frame rather than filling the width, and that is a decision rather than a shortfall. Filling it would crop almost thirty percent off the height of every picture, and in a hidden object game that means losing objects outright. One scene hides a brass elephant low enough that it would have been cut in half. So the picture stays whole and the bars are dressed as what they are: wood, a vignette and a shadow.

A point and click adventure under the water. Swim between rooms, talk to the people who live there, pick things up, and solve problems by carrying the right object to the right place. Five Moon Pearls are missing and the town has decided somebody took them.

Everything in it is a record rather than a component. A scene, a character, an item and a puzzle are all data, and a single engine file is the only place a requirement is ever checked or an effect ever applied, so a hotspot, a line of dialogue, an exit and a gift all behave identically. A new room needs no new code.

There is no modern quest interface. No objective markers, no map, no floating labels. The room name is captioned small in the corner, object names appear on hover, and a notebook page called Things I'm Wondering About is the entire quest system, with three escalating hints you have to choose to open.

It is also the one world that remembers you. The town and the riddles are an afternoon on a website and wipe when the computer shuts down. Finn is an adventure with a save file, the same exception a real adventure disc on that machine would have had.

Nothing in it beeps. The rest of the machine speaks in plain sine tones, which made the reef sound like a spreadsheet, so this world got two sound generators of its own: a bubble, which is a tone that jumps up in pitch very fast and dies just as fast, and a sweep of water moving past you. Every sound in the game is built from those two. Picking something up is three bubbles climbing. A wrong answer is two going down.

Stand still for ten seconds and fatter bubbles start up the edges of the frame with a quiet puff of sound. Any click resets it. The room notices you are there.

Wander snowy locations, play minigames, furnish an igloo, and turn up to seasonal parties. The most social feeling world, built without a multiplayer server.

The trick is atmosphere: crowds in the paintings, a party calendar that actually changes, and locations that look like somewhere people just left.

Last on purpose. It leans on the minigame framework, the room system and the seasonal calendar that the worlds before it already paid for.

The family whose computer this is has a calendar, an inbox, a buddy list, a grocery list and a standing argument about ground beef. None of it is invented. It is mine.

That was not the original plan. The plan was a period accurate family, written the way a set designer would write one. It turned out that inventing a convincing family is much harder than remembering a real one, and that the real one was better. So the machine quietly became a place to put things I did not want to lose, and almost none of it is announced.

Nothing below is labelled in the product. There is no about page, no author note in the calendar, no wink. You would have to already know.

Everybody on the messenger is offline when the machine boots. That is not a technical shortcut. It is the state the computer is actually in: nobody has signed on in years. A buddy only appears online once you message them and they answer.

My dad is on that list. He died when I was ten. He is the reason there is a red velvet birthday on October 7, a treasure hunt every February, donuts in July with a warning next to them, and a sticky note on the monitor telling whoever sits down not to install anything.

I did not build this to be sad about. Those are good memories and this is somewhere they can sit, in a machine that feels exactly like the one they happened around. If somebody plays it and never notices any of this, that is the correct outcome. It is a game about a family computer. It just happens to be a real family.

A game that looks like 2004 and is wrapped in a modern interface is a costume. So there is no global design system across the worlds. Each one gets the interface its own software would have had, and the only thing they share is a set of rules about how clicking works.

The rules exist because the art is photographic or painted, which means nothing on screen is a real element. Everything clickable is a box measured against the picture, and every one of these rules is the answer to a bug that shipped once.

The temptation with a project holding several games is to unify them. That would be the wrong call here. Half the pleasure of a family computer is that the software on it did not match: one program is bright and flat, the next is dark and serious, and the mismatch is what makes it feel like a machine with things on it rather than a product with modes.

So each world is allowed its own look, and the discipline is in holding each one steady rather than blending them.

Three finished games with nothing in common on the surface, built on one set of mechanics that never changes. This is the part that made a second and third world affordable, and the part that would let somebody else add a fourth.

Next.js exported as a static site, React and TypeScript, Zustand for state, Tailwind for layout, deployed from a repository to a host. That is the whole stack, and the constraint is the point: everything in here has to work as files on a server, the same way the sites it is imitating did.

127 TypeScript files, about 23,600 lines, 83 components and seven state stores, grown across three archived build generations. The shell stopped growing at generation three, which is when the work moved into worlds, which is exactly what the architecture was built to allow.

The design writing is the part I would show first in an interview, because it is the part that made the rest finishable.

It starts with a template of 25 fixed sections, so every world begins from a structure rather than a blank page. World one filled it to about 21,700 words: every system specified before it was built, from the economy and inventory to the school, the careers and the town paper.

Seven design pillars sit at the front, written as tests rather than values. Would a resident actually do this. Does it give people a new way to improve their home. A feature that fails several of them does not get built, and that sentence is in the document so the answer is decided in advance instead of in the moment.

Alongside the design documents sit the build notes, which are the more useful half. They record what is actually built rather than what was planned, every decision and the reason for it, and the traps worth not repeating. One of them exists purely because a launch once went out with no analytics and no visitor counter, silently, because three build time variables were never set.

The most useful habit in this project was treating my own work like a client deliverable. Each build got a numbered review document: notes written against specific screens, screenshots with the problem marked, and a clear line between what was broken and what was a decision I wanted revisited.

One round was structural: the viewport was wrong, the icons were too large, two card games were unplayable, and the whole thing did not work on a tablet. Another was almost entirely copywriting, cutting written gags out of the family documents in favor of things a real family would actually have typed.

Written rounds do two things mental notes never do. They create a record, so nothing quietly stays broken across three builds. And they separate taste from defect, which is the difference between a list of complaints and a usable brief.

A real mother writing a grocery list is funnier than a written gag.Review round four

That note is a good example of the value. It is not a bug. It is a decision about voice that arrived halfway through, got written down once, and has applied to every string written since.

The most recent round is the clearest case for doing it at all. Two hint lines still told the player that part of the mystery was not built yet, months after the ending had been written. The game was finished and was still saying it had stopped. Nobody playing it would have reported that as a bug; they would just have quit. It got found by reading, not by testing, and the note now says to search for that exact phrase after any change to the content.

Personal projects usually end by being abandoned. Each world here ended with an audit instead, because I wanted the word finished to mean something specific.

The check runs against the game data rather than by clicking around. Every id in the catalog is matched to a real file on disk, and every image path referenced anywhere in the code is resolved.

• Every catalog id matched to a file on disk. Every image path in the code resolved.
• Type check clean. Static export clean, with no warnings.
• Every scene loaded in a real browser: no console errors, no failed requests.
• Boot, login and all 16 desktop applications opening correctly.
• The adventure's content audit at zero problems, and its last full playthrough completing 54 of 54 steps from new game to the ending card.
• Every room in that world reachable, with the exit graph walked and all 25 scenes confirmed.
• Analytics live, and a visitor counter reading from a real endpoint.
• Open items logged honestly as decisions rather than quietly dropped.

The last line matters more than the clean ones. The audits record what is deliberately left undone as well as what is finished, with the reasoning attached, so the next world starts from a truthful picture instead of an optimistic one. The list shrinks as things get fixed: a batch of late scenes that had drifted toward photoreal has since been regenerated to match the rest, and every item in that world now has a puzzle that awards it, so nothing in its catalog is unobtainable.

Measuring things is the day job, so the temptation was to build a dashboard for a game nobody has played. What exists instead is the smallest honest setup: the page is instrumented, the counter is real, and the event layer is specified but not built, because events you designed before watching anybody are usually the wrong events.

There is also no conversion here. Nothing is sold, nothing is signed up for, there is no funnel. That makes it a cleaner measurement problem than most commercial work: the only question is whether somebody explores, and for how long, and how far in they get.

Written, not built. Each one exists to answer a question the design is currently guessing at.

• Boot completed, and how many people quit at a black screen with a power button on it
• Login completed, and how long somebody sat there deciding what to type
• First application opened, and which one it was
• Browser opened at all, since the fake internet is invisible from the desktop until you go looking
• World entered, and whether anybody enters a second one
• World finished, for the one world that can be finished
• Minigame played, and which of the twelve
• Hidden interaction found, which is the closest thing to a quality signal
• Suggest a game submitted, which is the only inbound the project has
• Session length, against the one to three hour estimate
• Return visits, which is the number the whole retention argument rests on

A product with no instructions is a product making a lot of assumptions. Every one of these is a bet, and the review rounds already lost a few of them.

Each of these was a design decision that somebody's actual behavior overruled, and every one of them was found by using the thing rather than by reading the specification.

• The drawing app had no undo, because the original did not. A child would expect one. It has one.
• Exits were invisible strips. Hunting for them is an obstacle, not a puzzle. They announce themselves on hover now.
• A character standing near an object silently ate every click on it. Objects and exits now sit above people.
• Two hidden object targets were ambiguous rather than hard. Removed, not made easier.
• Objects were named what they are rather than what they look like. Renamed to what a player would call them.
• The found markers were unreadable against a busy photograph. Retuned until they read at a glance.

The honest state of all of this is that it has been tested by one person who already knew the answers. The experiment that would actually settle it is in section 21.

Walking a finished prototype into a publisher does not get you a check. It gets you about ten questions, and none of them are about the art.

These are the questions, with the answers I would actually give. I have marked which ones the build has already settled and which ones are still a hypothesis, because the fastest way to lose a room is to present an assumption as a proven thing.

Describing it as five games undersells the part that took the longest. The accurate description is a fictional family computer holding an expanding collection of playable worlds: I recreated the feeling of finding things on a 2000s family computer, then built an architecture that lets completely different games live inside the same machine.

Said that way, every piece of the project has a job.

Everything above is reasoning. The only thing that turns it into evidence is watching people who had nothing to do with making it try to use it, with no explanation first. That experiment is simple enough to run as soon as the thing is public, and the analytics and the visitor counter are already wired in so the first answers arrive as data rather than as a feeling.

What I would be watching for:

• Do they understand what they are looking at?
• Do they know what to click?
• How long do they keep exploring?
• Do they ever find the browser?
• Do they open more than one application?
• Do they enter more than one world?
• Which world holds them longest?
• Do they find anything hidden?
• Do they come back without being asked?
• Do they tell somebody else about it?
• What do they ask for next?

The order matters. Finish it, show it, find out whether anybody wants more, and only then build the machinery to make more. The other way around is how a good idea turns into an expensive one.

Here is what I built. Here is why I built it this way. Here is what I have proven, and here is what I have not.The only pitch worth making