Skip to content
Age of Carthage
Preorder
Age of Carthage
LanguageENFR
Preorder

Played 11,819,300 times before the first table

Four generations of testing

Before the first physical copy existed, the rules were written as a program and played by computers, again and again. Four generations of testing, each one asking harder questions than the last. The verified total is 11,819,300 games. Counting every trial run beside them, the number is about 14.6 million. Every result file is kept, and 173 of 173 can be run again and return the same answer.

Three programs that had to agree

One program can be wrong in a way nobody notices. So the game was written three times, on three separate engines, and made to play the same 6,000 deals five times each. That is 30,000 runs. If two engines ever disagreed about who won, the test failed. They never did. Zero mismatches.

The first version could not lose

The most useful thing the testing found was embarrassing. In the first version, Carthage could not fall. Not once, in any setting. A game about a city that might die, where the city never died. The costs were rewritten, the way Pillars fall was rewritten, and the tests were run again until the city could really be lost. That fact is the foundation of everything else on this page.

The coin flip we proved

The last Act has to be a real gamble. The rule we hold ourselves to: with good play, the city stands between 40 and 60 times in 100, at every table size. The model behind the vote gives 43 in 100 at three players, 42 at four, 49 at five, 49 at six, and 55 at seven. Inside the band, every time. When real tables drift, one knob at a time moves it back, and the knobs are written down in order.

The team game ends level, on paper

Cato's Directive does not need luck to be fair. If Rome never fights and Carthage never wastes, Carthage's final total is exactly its whole budget, and so is Rome's. An exact tie, whatever the budgets, whatever the costs. So both sides are forced to fight over margins they cannot see. It is three lines of arithmetic, and it holds at any table size.

Source
  • 11,819,300 verified games across four generations of testing (v2 battery 2,909,900 plus v3 battery 8,909,400), each one reproducible from its own result file. 173 result files, 0 not reproducible.
  • Cross-checked across three separate programs (the game engine, an OpenSpiel wrapper, and an RLCard adapter, run in five configurations): 6,000 deals, five arms each, 30,000 runs, zero mismatches.
  • The vote model's win rates (43/100 at three players through 55/100 at seven) are the design's own model, recomputed by the test suite's verifier at every player count, not a measurement from played tables.

The talk is designed too

Cards make the money story. People make the real one. The Table Talk lines that open each Act are built on three plain facts about people: we copy the crowd when unsure, we fight hardest for what we already built, and a promise said out loud sticks to us even when the rules say it binds nobody. The design also draws on published research: why secret, all at once decisions stop one loud player from running the table, why a loss you can trace to your own choice keeps you playing while a loss that feels imposed makes you quit, and why a game should satisfy people rather than trap them.

No hooks, no timers, clean endings

This game is built not to be addictive. No reset timers, no streaks, no manufactured unfinished business. A game ends, you say Carthago Aeterna Est or you lay the Agora down in silence, and you decide whether to play again. That decision is yours, every time.

What we have not tested yet

Numbers prove arithmetic. They do not prove fun. The first physical tables start now, and the first question we ask is the simplest one: was anyone annoyed within two Acts? Results go on this page as they come, in both endings, at every player count. If you play, the feedback form takes one minute, and your game becomes part of the record.

Quick answers

How many games were simulated?

11,819,300 verified games across four generations of testing. Counting every trial run alongside them, about 14.6 million.

What does "verified" mean here?

Every one of the 11,819,300 games sits in a result file that can be run again and returns the same answer. 173 files, all of them.

Which research does the design use?

Published work on stopping one player from running a table, on why losses you own keep you engaged, and on games that satisfy needs rather than exploit them. The sources are listed on this site.

Has it been played by real people?

The physical tables are starting now, and the results will be posted here. The simulations prove the arithmetic; only tables can prove the fun.

Keep reading