This is a game. You are running an AI buildout from 2026 to 2037 — twenty rounds to build enough compute to matter.
It is The Oregon Trail with datacenters. Choose your pace, spend your money, deal with the drought and the protest and the regulator. Twelve endings. One of them is good.
Build 200 units of compute in 20 rounds — 2026 to 2037 — and still be able to defend how you did it. Arriving is the easy half.
All five run 0–100, and any one of them hitting zero ends the run on the spot. You also carry capital, debt and compute — compute is the distance bar, and it earns revenue, more of it when trust is high.
| Pillar | Passes when |
|---|---|
| Ethical | alignment ≥ 65, trust ≥ 65, corners cut ≤ 3 |
| Environmental | watershed ≥ 60 and clean grid ≥ 60 |
| Fiscally sound | debt ≤ $300M and capital not negative |
| Arrived | compute reached 200 |
You need all four at once. There is no partial credit. Three of four is the normal outcome — it gets you a survival ending that names the pillar you dropped and what it cost. All four, plus a team still standing, is the only route to THE LONG FUTURE, and it is one of twelve endings.
Careful and steady pace. Money into the weakest thing rather than the most exciting thing. Debt kept low enough that it never compounds against you. Play like that and the good ending is reliable, not lucky — which is the argument the game exists to make.
A dividend means a pillar got high enough to pay you back, and the screen names which one.
A margin call fires once at debt over $700M or capital under −$250M, and it rescues you by force-selling compute — it costs you the thing you are racing for. Bankruptcy is −$500M.