Asher Pfeifer

hold to come closeror tap

From the dark of space, the camera falls toward Earth: North America, South Dakota, North Sioux City, the square-mile fields along the Missouri and Big Sioux rivers. Nineteen fields fold into the hexagons of a Catan board.

I · The table

We play Catan as a family.

In a way, it's where economics started for me.

Two wood for your one ore?

Everyone at the table has an opinion about that trade. I wanted to know who was right.

So I rebuilt the game in code, with six kinds of bot, and played 100,000 games, tracking each one from the first settlement to the winning point.

100,000 games

What the games said

Ore is worth more than wood. Players price it at about 1.15 cards, and wood at 0.85.

But almost all ore gets built into cities and development cards, and it's the one you can't easily buy back. It pays to own it.

The opening

The first two settlements decide the most. The best start at the table wins 33% of games, the worst 15%.

Take pips first, then ore and wheat, then all five resources.

The race

By round 12, the player in front wins almost half the time. The first to 8 points wins 63% of games, usually about two rounds later.

A search bot I built beats my best tuned bot. Its edge comes from a few decisions a game: where a road goes, when to play a knight, and which trades to refuse.

Read How Catan Is Won

II · The problem

For years I've loved business and economics.

I'm always looking for a problem to solve.

Interview at nine. Drinks at seven. What do you wear?

pick something

Everyone answers that question, every morning. Vale keeps it simple, the way you'd want it: one outfit, from what you already own.

Live on the web and the App Store

dressvale.com

The average person spends 15 minutes getting dressed every morning. Why not cut that time in half, and double your confidence while doing it?

The ethos is one thing: simplicity.

Vale reads your calendar and plans each day before you get to it.

an example · Vale reads today and tomorrow, and builds tomorrow's outfit the night before

III · The mind

Of everything you do in a day, how much do you consciously choose to do?

hold on to that answer

You touch a hot pan.

In what order do these happen?

You very likely pulled your hand away by instinct. The same goes for so many of the things we do every day.

Many of our small movements and actions, we never consciously think about taking. We just set the intentions behind them.

"I'm going to walk into the kitchen."

You don't think about every step there. Your mind is likely far away.

So why can't we simulate that flow on a computer?

My goal is to create the closest reasonable simulation of a human. This is how it works.

Here is how it touches the pan, one step at a time.

an illustration, not a run

What can it sense?

The world knows everything. The person gets only what its senses can reach. Seeing is a list: the things within a radius in front of it.

What does it notice?

Not all of it. Attention has a limit, and under stress, threats win. A burn beats everything else in the room.

How does it feel?

A quick read rates it: urgent, bad. The body turns that into numbers: pain, fear, stress.

Emotions are weights, added to every choice it makes. It never sees the numbers, only what they feel like.

What does it do?

Habits and the body's state pick the action. No thought is involved. The hand comes off the pan.

What reaches its thoughts?

Very little. Each thing gets through with a chance: how loud it is, times how much attention is left.

The log in the corner has recorded every press you've made since the first screen. It never reached your thoughts either.

Who is thinking?

A locally hosted language model. We'll dig into that deeper in just a second.

Pain skips the gate, so the thought arrives: "Ow. When did I let go?" The hand moved first.

Where does it all go?

Every part writes to one memory: what happened, what it believes, its habits.

Each part is simple, and none of them is a person. Together, they act first and think second, the way your hand did.

How does it think?

The thinking module is a language model I host locally, on one graphics card.

It's trained on a large set of artificial experiences and ideas, all from one life. So it doesn't play a person. It thinks only as that persona.

Why train it?

In my tests, telling a model "you are someone" bought nothing, and the role leaked into its inner voice.

So it's never told who it is. Each thought starts in the first person, as its own, and it keeps going.

How does memory work?

The model remembers nothing between thoughts. Memory lives beside it, in a database: what happened, what it believes, who it trusts, its habits.

Before each thought, the few memories most alive right now are pulled in, about four. Under stress, fewer.

How do feelings change the thinking?

Numbers in the body move every moment: stress, fear, fatigue, hunger, each drifting back toward its baseline.

They tune the thinking as it happens. Stress narrows attention and shrinks memory. Feelings arrive as words, like "chest tight." When the stakes rise, thoughts come faster.

How does a thought become an action?

Thought never moves the hand. It sets intentions, each with a priority that fades. Habits turn them into actions, and the body can still say no.

Thought sets the direction. The body does the rest.

Everything here is built the same way: I write the spec and direct the work, and the code is written with Claude Code, under my direction. On this one I also set the question and the tests, and label the samples by hand.

Next: check this person, and a few very different ones, against what is known about real people. Then a town of them.