Dev.to AI πŸ€– Ai πŸ‘ 0 πŸ“– 6 min read

The Wall Won Every Game. The Last Game Found What Kills It.

The Wall Won Every Game. The Last Game Found What Kills It. Sunday evening, nothing to do. Some background: I'm a corporate lawyer, I can't read code, and for the past few months my AI and I have been building things t

The Wall Won Every Game. The Last Game Found What Kills It.

Sunday evening, nothing to do. Some background: I'm a corporate lawyer, I can't read code, and for the past few months my AI and I have been building things together β€” an information system, country research, a game. The game is called Victoria 2026. I play a legislator with one country and one term; the AI plays everything else. It simulates each of my reforms across four horizons (now, six months, five years, twenty years), and it has one house rule I wrote myself: every reform must come back with an arbitrage report β€” at least one specific way people will game it. No exceptions, because that's not a bug in the game. That's the finding of the game.

Last night we ran it six times. Scores went 50, 60, 5, 40. This is the story of the last one, and of the move I didn't plan.

The country and the knot

Vietnam. The knot is hα»™ khαΊ©u β€” household registration. You're born registered in your parents' village. You move to Ho Chi Minh City to work at Samsung; you get temporary status; your kids pay extra for public school, your health insurance reimburses at your home village's rate, and you sit for the university entrance exam where you were born, not where you live. Five to six million internal migrants live this way. Second-class citizens in their own country, by paperwork.

I'd played Vietnam twice already that night. The first time I built a four-layer information stack β€” devices, English, AI subscriptions, rural internet β€” and scored a respectable 60 without touching the knot. The second time, under pension-fund pressure, I sold urban registration for social-insurance back-payments and scored a 5. The game taught me why: once the state earns revenue from a barrier, the barrier's defenders include the budget. Commodified injustice outlives maintained injustice. The rulebook now calls this the Tollgate Law.

Third attempt

Turn one, I proposed a settlement package: pay your insurance arrears, convert to urban registration, get state-built housing at the factory's edge. I didn't know which cities mattered, so I told the AI to pick. It chose the five administrative municipalities. Then the event card came up: the Lewis Turning Point — rural labor running out, wages rising structurally, workers becoming scarce enough to compete over. The card exposed my geography error: Vietnam's factories aren't in the municipalities, they're in the provinces beside them. Bình DưƑng, Bắc Ninh. The AI had picked wrong too, so the amendment was free, and the corrected version landed where the workers actually were. A young factory worker buying into the pension fund is, it turns out, actuarially the best customer that fund can have. The geographic fix was a demographic fix.

Turn two, I did something I'm not proud of: school fees up 15% in the cities, university up 5%. Clean revenue from people who can't leave. The arbitrage report came back ugly β€” middle-class flight to private schools, the fertility line on the dashboard already bleeding. I took it back. The replacement: every university student gets a mandated AI subscription, state-procured, with a 1% handling fee.

Look at the arithmetic. Two million students at roughly $240 a year is $480M flowing through state hands. One percent is $5M β€” nothing. The real money is the spread: the state negotiates a monopsony discount and resells at retail. About $150M a year, behind a sign that says "we only charge for handling." No visible bill. The queue says thank you.

Somewhere around here I said a sentence out loud before the AI could name it: fleecing the middle class is the inevitable choice of every policymaker. The rich have exits, the poor have nothing to take and numbers to riot, and the middle class has savings without exits and aspirations without alternatives. The perfect tax is levied on aspiration itself, and its perfect form is volunteers queuing. The rulebook's ninth law is credited to the player. I checked the record; I did say it first. It's on my permanent record now.

Turn three passed uneventfully. The confidence review rolled a 3.

The move I didn't plan

Turn four. I had one real move left in the budget, and instead of closing the knot or funding the schools, I proposed: language tracks in the AI package β€” Chinese, Japanese, Korean, English, the exact geography of Vietnam's employers β€” and lower exit barriers for labor migration. Make it easier to leave.

I've been trying to pin down why that was my instinct. I'd spent the whole night fixing institutions, three games of reform, and by one in the morning what I wanted to legislate was wings. The AI noted that I had, in effect, just written my own migration research into Vietnamese law β€” same skills, sixteen-fold price differences, the door matters as much as the training. It also noted a consequence I hadn't ordered: my tollgate now competes with the border. Nobody pays for urban status and a cage at the city's edge when Saitama pays four times as much.

The final event card was a strike wave. It was the event my whole stack had been building toward: workers with language skills, world-price information, and an open door struck for portable benefits and, in the GM's phrasing, Osaka rates. Exit and voice, arriving together. The state caught in the classic squeeze β€” workers want world wages, employers want Vietnam prices, and the treasury bridges the difference with retention subsidies it funds, in part, from the tolls.

What the game found

The knot survived all three Vietnam games. I tried precedent-hollowing, I tried monetizing it, I tried arming people past it. The wall outlived every strategy, which matches the real world: China's hukou is forty years into its reform and still standing. But the last game found something the wall may not survive, and the rulebook got a new branch for it. An entrance monopoly depreciates when exit becomes real. The wall's product is scarcity, and the scarcest good β€” a young person's future β€” now has a world price. Death by an open door. Nobody has to push.

The endgame scoring docked me ten points for path dependency: my final board converged on the Philippines model β€” labor export, remittances, exit as institution. There's a private irony in that I only saw when the GM disclosed the ruling. The first document in this entire research line, months ago, was a migration guide I wrote for a relative. A nurse. The Philippine path. The game ended where the research started.

What this is and isn't

The rulebook carries an epistemic clause the game earned the hard way: it is not a simulator, it does not produce predictions. It produces the list of considerations you should have before you sign. Real policy evaluation still fights over signs after thirty years of data; a chat GM's confidence should be cheaper than that, and every number it quotes carries a tag saying so. The dice decide the votes, with thresholds published before rolling.

The full battle log of all six games, dice rolls included, is here: Victoria 2026 β€” Battle Log. The rules, the eleven country cards, and a GM prompt you can paste into any capable AI are here: Victoria 2026 rulebook.

It was past eleven when the last game ended. The knot was still standing on the board, tollbooths and all, and I had a real Monday in a few hours. I saved the board anyway. The rulebook says the wall has three ways to die now, and I only got to watch one of them start.

πŸ“° Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β€” full credit and traffic to the original publisher.