Frontier lab strategy game
A competitive tabletop game about running a frontier lab on the path to AGI while balancing ambition, safety, trust, and money.
Overview
Each player runs a frontier AI lab. You raise Capital, buy Compute, hire Talent, play research and deployment Projects, manage Safety, preserve Trust and Influence, and try to build the most valuable AGI-era lab.
Capability means your lab's model power: raising it gets you closer to AGI and unlocks stronger Projects. Safety keeps that progress from becoming reckless. Deployments create Revenue and Legacy Points. Private Incidents hit one lab at a time, while Public Events change the field for everyone.
One round
Before round 1
Table language
Primary race axis
Capability unlocks stronger Projects, gates major Deployments, and marks the race toward AGI. If Capability outruns Safety, your lab still advances, but the Safety Gap makes incidents and final scoring harsher.
Turn economy
At the start of a Turn, draft Project cards into Hand, then gain Capital equal to Revenue. Capital raises happen through Financing Program cards.
Player aid
| Purchase | Cost | Limit |
|---|---|---|
| 1 Compute | 2 Capital | 3 per Turn |
| 1 Talent | 4 Capital | 1 per Turn |
Compute and Talent are unlimited in the base game. Safe labs hire Talent for 1 less Capital. Scarcity, leasing, poaching, and stealing are later modules.
How to read cards
Capability Program
+3 Capability. If you are below Growth stage, gain 1 Influence from benchmark recognition by funders and standards bodies.
Rules glue
Endgame trigger
AGI is reached when any lab's Capability hits the threshold. That ends the game after the current round, but the winner is still decided by Legacy Points, including safety penalties for racing too far ahead of safeguards.
Worked procedures
Use these when teaching the first game. They cover the procedures most likely to create rules questions.
Table aids
Print one aid and one lab mat per player. Trust and Influence are token counts, Revenue is a slider, and Deployments live on the map-style area of the mat.
Project deck
The first deck has 72 Project cards. Duplicate cards are rendered as physical copies for print-and-play testing.
Double-sided printing
Print these on the reverse side of each deck, or sleeve them behind the card fronts for early playtests.
Counters and markers
Use cubes if you prefer. These sheets provide printable labels for the full v0.5 component counts.
Print order
Print fronts first, then add backs only after one physical readability pass. The card sheets are built around true 2.5 x 3.5 inch poker cards, six cards per letter sheet.
Physical proof
Use this sheet before the first teaching game. It is the bridge between a checked digital print kit and proof that printed cards are actually readable in hand.
Teaching evidence
Fill this during the 5-round teaching game so later balance notes are based on round state, not memory.
Standard game evidence
Fill this after the first full standard game. Then mark the acceptance gates before changing any costs, payouts, scoring, or card mix.
Playtest report
Fill this immediately after the session, then transfer it into the matching staged draft in docs/evidence-drafts/. Promote it to docs/playtests/ only after it passes the matching --file validator.
Private incident deck
Private Incidents stay hidden unless they have the Public keyword. Keep private Incident Records face down with a Severity marker visible for endgame penalties.
Public event deck
Private mandates
Draw 2, keep 1 secret. Reveal during final scoring.
Pre-playtest balance
The current deck has a local simulation harness. Treat the numbers as early warning signals, not final tuning. Human playtests still decide balance.
docs/balance-simulation-report.md. Run npm run balance after each rule or card tuning pass.
Revision log