Credits — Balance of Threat

BalanceofThreat

A game about the security dilemma, balancing and bandwagoning, and guns against butter, made for a classroom and played first by the people named below.

“The Assyrian came down like the wolf on the fold” Lord Byron · The Destruction of Sennacherib · 1815

Arta Khakpour

Music

“Wolf on the Fold (Remix)”

by Arta Khakpour, made with Suno

Our beta testers

The very first

Steven Chow

who played it before there was much of anything to play

The first class to play, October 2026

Caden

Devin

Madalyn

Mady

Oli

Shirin

Sydney

Tara

Every bug you met, every alliance you broke, and every strike you launched made this a better game. Thank you.

How it began

An early screen: rival strengths as bars of uncertainty — **September 4**The first idea that worked: you never see a rival's true strength, only a band of guesses.

The region drawn as boxes joined by lines — **September 9**The world was six boxes joined by lines.

The first map: six rounded shapes — **September 10**The first map: six blobs, bravely labelled.

War arrows over the first painted map — **September 12**The first painted map, and the first war arrows across it.

The situation-room screen with photo cards — **September 15**The situation room takes shape.

Three ministers' advice cards — **September 25**The Cabinet arrives, three ministers who never agree.

The two-continent world with attack arrows — **September 30**The world splits in two: Valmera, Gorgon, and the ocean between.

The Garrow Bulletin front page — **October 1**Every country gets its own newspaper.

A first, flat-coloured map of the Gulf — **October 3**The real world arrives: a first draft of the Gulf.

The painted map of the Gulf — **October 4**The Gulf, painted in ink on parchment.

Under the hood

Before anyone played a turn, the computer played this game against itself, hundreds of games at a time, every time a rule changed. Here are some pages from those notebooks.

The security dilemma, measured

The same 200 worlds, played twice: once where every country can see the others' true strength, once in the fog of estimates you played in. In the fog, everyone spent about 15% more on arms in the first 10 turns, from uncertainty alone, and the world ended up poorer. Nobody behaved foolishly. That is the security dilemma.

200 games · verified

$ python run_playtest.py 200
…
6. SECURITY DILEMMA TEST  (the one that matters)

                             noise on   noise off      change
   mil spend, turns 1-10       1253.2      1085.9      15.4%
   world economy               1137.9      1193.7      -4.7%
   wars per game                 9.89        7.25      +2.64

   -> WORKING. Uncertainty alone drives states to arm more heavily,
      leaving the world poorer without making anyone safer.
      That is the security dilemma, and students will feel it.

Held through every rebuild

Every time the rules were rebuilt, the same test ran again. If the fog ever stopped making countries arm, the game would no longer be teaching what it was made to teach.

From the design notesThe balance bug worth remembering

In the very first version, the peaceful trader won 61% of games and the hawk won 1.8%. Being defenceless was perfectly safe: defenders had the advantage, allies came to help, and every attack was a losing bet.

That would have taught “never arm”, the exact opposite of the lesson, and there is no security dilemma if arming never pays. Tuning four rules (what a conquest wins, how much allies help, how much security counts, what aggression costs your reputation) closed a gap of 59 points to about 15.

Balance failures are invisible in a design document and obvious within twenty minutes of automated play.

From the design notesWhy the non-aggression pact was cut

For most of the game's life, a non-aggression pact really stopped an attack between the two countries that signed it.

But the security dilemma rests on one idea: you cannot trust a promise you cannot enforce. A pact that mechanically stopped a war let a signature buy real safety, which is exactly what a signature cannot buy. The historical case is Molotov–Ribbentrop in 1939, then Barbarossa in 1941: the paper was worthless the moment the balance of power shifted.

So it went. Your safety rests on what really holds it up: strength, allies, and deterrence.

From the balance notesGermany, 1938

The computer players were spending 25 to 35% of their economy on the military, in peace as in war. The yardsticks: the world today, about 2%. Great powers in 1913, 3–5%. The Cold War United States, 5–10%. The Soviet Union, 12–17%. Germany in 1938, 17–25%. Total war, 37–60%.

“I want diplomacy and strategy to matter, not just pouring crazy amounts of money into the military,” the designer wrote. So security became enough, not most: you score for being able to defend yourself, not for outspending the world. The computer's arms spending fell to about 11%.

What the tests found

0 of 1,368wars fought between trading partners when trade agreements were first tested: the liberal peace, appearing on its own

26 → 14the terror threat in long games after it was rebuilt on the real record of the war on terror, where it had spiralled out of control

97%how often a “90–99%” joint-attack forecast now comes true after a forecast of 99% lost, and the planner was rebuilt and checked against 2,400 battles

4.35 → 0.30nuclear first strikes in a long computer-played campaign once the computer learned to count the other side's warheads before firing