GENERATIONAL
UN-FUMBLE

a life in San Francisco · the AI wave · real statistics

You arrive in San Francisco at the start of the AI wave as a random person. Twelve years. Real statistics decide the odds; every six months a situation lands that only you can decide. The companies are fictional; the numbers are not.

$0
of venture money deployed worldwide so far today
$0
of San Francisco rent paid so far today
0
tech workers laid off so far today, at this year's pace

Born the same month

Four other people arrive the month you do, living by the same statistics on the same market. You see them at every milestone, and at the end.

Decisions, written for you

Every six months a situation is written for your exact life: your employer, your partner, your bank balance, the last thing that happened. Each new year gets a review. The game's statistics still decide the odds and bound every effect.

Why you cannot just buy the chip stock

Each run reshuffles which sectors win, and winners can flip. Rent is due monthly whether or not your paper equity is up. At the end, 450 reruns of your life put a number on how much was luck.

Who you arrive as
About one in four SF residents works in tech. The rest of the city lives through the same wave.
Romance
With romance, the apps open in month three and you set the rhythm after your first swipe. Without it, nobody asks.
Pace
At 1× a month takes four seconds and a year about a minute, so a full life is ten to fifteen minutes with your decisions. The dial at the top goes to 8×. Time pauses whenever a decision needs you.
Language
The screen, the feed and the decisions written for you all follow this choice. Switching reloads the title screen.
Story decisions
checking… · Every six months a situation is written for your exact life; the game's statistics still decide the odds.
My lives
Lives you finish are kept in this browser only.
Data and methodology

Every life is drawn from these distributions, then your choices and a seeded random number generator do the rest. Measured means a published statistic; estimate means the game interpolates between published figures. Gathered October 30, 2026.

Calibration check

500 lives played by a neutral autopilot that also takes the half-year decisions, twelve years each. Median net worth ~$580k; 37% end above $1M, 4% above $10M, 5% below zero. Tech workers: median $3.4M. 5% leave the country when a visa lapses; 2% leave the city broke; the autopilot marries 79% of the time when it wants a family. Across a strategy grid (cautious, moderate, aggressive, founder, and a cheat that knows the sector winners), no strategy wins more than ~65% of head-to-head lives on the same draw, and the cheat does not beat a moderate player. Founders have the highest mean and the lowest median, as in the Carta and Horsley Bridge data.

How story decisions work

A model writes the situation and the options from your current state, but it only proposes effects from a fixed schema (cash, burnout, skill, network, relationship, job risk, offer odds, a few actions), each clamped to bounds the engine enforces, and risky options carry an explicit probability the engine rolls. No story decision can hand you a company, a windfall above two months of income, or a job. The calibration harness, which has no model, plays a handwritten deck under the same rules.

Self-hosting and other models

Hosted with the Node server in server/, this site holds the model key and writes the decisions for everyone. The browser only ever talks to this server; no key is typed into or stored in the page. Inside claude.ai the page cannot reach the network, so decisions come from Claude through the viewer's account. To use another OpenAI-compatible model, point the server's OPENAI_BASE_URL at it.