Professor Jiang Xueqin’s game-theory-based geopolitical forecasts made him a viral sensation with 1.8 million YouTube subscribers. The same methodology — historical pattern matching, actor-incentive modeling, and strategic scenario analysis — is exactly what AI prediction market agents automate at scale. Here’s what builders should take from his approach.

The Professor Who Called It

In May 2024, a Beijing-based educator named Jiang Xueqin published a lecture on his YouTube channel Predictive History titled “The Iran Trap.” In it, he made three predictions: Donald Trump would win the 2024 election, the US would enter a military conflict with Iran, and the US would ultimately lose that conflict.

Two out of three have materialized. Trump won in November 2024. US and Israeli forces struck Iran on February 28, 2026, killing Supreme Leader Ali Khamenei and triggering an ongoing exchange of military strikes. Jiang’s third prediction — an American defeat — remains unresolved and heavily debated.

The internet dubbed him “China’s Nostradamus.” His subscriber count surged past 1.8 million. He appeared on Breaking Points, Glenn Diesen’s show, and dozens of international outlets. But what actually made his predictions work is more interesting than the predictions themselves.

Jiang’s Method Is Game Theory, Not Mysticism

Jiang doesn’t use astrology or intuition. His framework — which he explicitly calls “predictive history” — borrows from Isaac Asimov’s fictional concept of psychohistory and operationalizes it through three pillars:

Historical pattern recognition. Jiang maps current geopolitical configurations to historical analogs. His Iran analysis drew a direct parallel to Athens’ disastrous Sicilian Expedition of 415 BCE — an overconfident empire projecting power into hostile terrain against a prepared adversary.

Game-theoretic actor modeling. Each nation-state is treated as a rational actor with defined incentives, constraints, and capabilities. Jiang models what each player wants, what they can do, and what their optimal strategy is given other players’ likely moves. His Iran analysis mapped US domestic political incentives (strongman politics, military-industrial complex), Iranian strategic preparation (20 years of proxy warfare refinement, Strait of Hormuz leverage), and Gulf state vulnerabilities (water desalination plants, oil infrastructure).

Structural trend analysis. Long-cycle forces — debt structures, reserve currency dynamics, military recruitment trends, institutional trust decay — form the backdrop against which specific events unfold.

This is not prophecy. It’s scenario planning with a narrative wrapper. And it maps almost perfectly to what prediction market agents do computationally.

The Prediction Market Mirror

While Jiang was going viral, Polymarket was pricing the same events in real-time with real money.

The numbers are staggering. The “US strikes Iran” market generated $529 million in total volume — the fourth-largest market in Polymarket history. The Khamenei leadership market pulled $45 million before resolving at 100% after his death was confirmed. As of March 9, 2026, Polymarket hosts 217 active Iran-related markets covering everything from ceasefire timelines (68% chance by June 30) to US ground invasion probability (23% by March 31).

One trader — “Magamyman” — made $553,000 on correctly timed Khamenei bets. Six wallets flagged by on-chain analysts collectively profited $1.2 million by betting on the February 28 strike date, drawing Congressional scrutiny and proposed legislation to ban prediction market trading by government officials and their families.

Prediction markets and Jiang’s approach share the same core mechanic: model actor incentives, assess structural constraints, estimate probability of outcomes, and put conviction behind the estimate. The difference is that markets do it continuously, with thousands of participants, and with financial skin in the game.

What AI Agents Add to the Stack

Here’s where it gets interesting for builders. Jiang is one person producing narrative-length analysis on a YouTube schedule. AI agents can run the same game-theoretic reasoning — at higher frequency, across more markets, with continuous Bayesian updating.

An agent operating on the Polymarket CLOB/CLI can:

Ingest and process information 24/7. News feeds, OSINT signals, social media sentiment, satellite imagery analysis, shipping data, energy price movements — all of it feeds into a probability model. One fully automated bot recently executed 8,894 trades across prediction markets, generating nearly $150,000 in profit by exploiting brief mispricings.

Model actor incentives programmatically. The same game-theoretic framework Jiang applies intuitively — what does Iran want, what does the US want, what are their constraints — can be encoded as payoff matrices and updated as new information arrives. Multi-agent systems using frameworks like CrewAI can assign specialized sub-agents to track different actors, then synthesize their outputs into a unified probability estimate.

Execute across correlated markets. When a development shifts the probability of a ceasefire, it simultaneously affects oil futures, regional stability markets, and downstream election markets. Agents using tools like pmxt can identify and trade cross-market correlations that no individual human analyst would track simultaneously.

Update continuously. Jiang publishes a video; the market moves. An agent monitors the source data that would cause Jiang to update his view, and adjusts positions before the analysis even gets published. This is the core advantage — not speed of execution, but speed of inference.

Jiang’s Framework as Agent Architecture

If you were building a geopolitical prediction agent inspired by Jiang’s methodology, the architecture maps cleanly to the agent betting stack:

┌─────────────────────────────────────────────────┐
│           Layer 4 — Intelligence                │
│                                                 │
│  ┌─────────────┐  ┌──────────────────────────┐  │
│  │ Historical   │  │ Game-Theoretic Actor     │  │
│  │ Pattern DB   │  │ Models (payoff matrices, │  │
│  │ (Jiang's     │  │ constraint sets, optimal │  │
│  │  analogs)    │  │ strategy computation)    │  │
│  └──────┬──────┘  └────────────┬─────────────┘  │
│         │                      │                │
│         ▼                      ▼                │
│  ┌─────────────────────────────────────────┐    │
│  │  LLM Synthesis Layer (Claude / GPT)     │    │
│  │  - Scenario generation                  │    │
│  │  - Probability estimation               │    │
│  │  - Cross-market correlation detection   │    │
│  └──────────────────┬──────────────────────┘    │
│                     │                           │
├─────────────────────┼───────────────────────────┤
│  Layer 3 — Trading  │                           │
│                     ▼                           │
│  ┌─────────────────────────────────────────┐    │
│  │  Polymarket CLOB / Kalshi API           │    │
│  │  - Position sizing based on Kelly       │    │
│  │  - Cross-market arbitrage               │    │
│  │  - Continuous rebalancing               │    │
│  └─────────────────────────────────────────┘    │
│                                                 │
├─────────────────────────────────────────────────┤
│  Layer 2 — Wallet                               │
│  Coinbase Agentic Wallet / Safe multi-sig       │
│                                                 │
├─────────────────────────────────────────────────┤
│  Layer 1 — Identity                             │
│  Moltbook agent registration / ENS / SIWE       │
└─────────────────────────────────────────────────┘

The intelligence layer is where Jiang’s methodology lives. The historical pattern database contains structured analogs — “overextended empire invades mountainous terrain against prepared defender” — tagged with outcomes and confidence levels. The game-theoretic actor models encode each player’s incentive structure, capability set, and strategic options. An LLM synthesis layer (Claude, for example, via Polyseer-style multi-agent architectures) generates probability distributions and identifies which markets are mispriced relative to the model’s output.

Where Jiang’s Approach Falls Short — And Agents Don’t

Jiang’s critics have valid points. His methodology relies on selective historical analogies, narrative framing that can overfit to a thesis, and untestable assumptions about actor rationality. The Free Press described some of his broader content as conspiratorial. Academic critics note he uses “game theory” rhetorically rather than with formal mathematical modeling.

AI agents solve several of these weaknesses:

Formal probability estimation. Where Jiang says “I think Iran has many advantages,” an agent outputs P(Iran strategic advantage) = 0.73 with a confidence interval and updates it every hour.

Multi-model ensembling. Instead of one person’s analogical reasoning, agents can run multiple forecasting models in parallel — base-rate analysis, structural trend extrapolation, sentiment analysis, formal game-theoretic computation — and ensemble their outputs.

Backtesting against prediction market data. Tools like FutureBench are already benchmarking AI agent forecasting accuracy against real prediction market resolutions, creating feedback loops that Jiang’s YouTube channel can’t match.

No narrative bias. Jiang has a thesis and selects evidence that supports it. Well-designed agents weight evidence by source reliability and recency, not by narrative fit.

What This Means for Builders

Professor Jiang’s viral success proves one thing conclusively: there is massive demand for game-theoretic geopolitical analysis that goes beyond TV punditry. His 1.8 million subscribers want someone modeling actor incentives, mapping historical patterns, and making specific, testable predictions.

Prediction market agents are the scalable version of this. The tools exist today:

The $529 million in Polymarket Iran volume isn’t just a number — it’s a signal that geopolitical prediction markets are now large enough to sustain sophisticated agent trading strategies. The question isn’t whether AI agents can do what Professor Jiang does. They already are. The question is whether your agent is doing it yet.