March 2026: over 150 Polymarket accounts drop big money on a U.S.-Israeli strike against Iran. The day before it happens. At least 16 wallets clear more than $100K each. Senator Murphy calls it "worse than insider trading."

What if Murphy got the category wrong? Insider trading is when you bet on something you know will happen. The Iran situation might be different. What if the people placing bets were also the people deciding whether to strike? From blockchain data alone, you can't tell.

This is the reflexivity problem. Small prediction markets work like thermometers. Big ones heat the room.

Here's how it works:

Geopolitical contract opens at 15%. Early buying pushes it to 25%. At 25% journalists start writing stories. Intelligence analysts flag the price movement. Decision-makers notice that external sources seem to validate something they were already thinking about. The price doesn't have to reflect reality accurately. It just needs enough credibility to nudge the people whose choices create reality.

Traditional markets move slowly. Information flows through human networks at human speed. You can interrupt the cycle.

But autonomous agents update and trade in seconds. The reflexive loop that used to take days now happens in minutes. An agent sees a price move, treats it as new information, trades on it, moves the price. Another agent sees that move and responds. The agents are working correctly according to market theory: they're updating on all available information, including price signals.

Prediction market theory assumes the market is small compared to what it's measuring. That assumption breaks at scale.

This doesn't make prediction markets useless. They do aggregate information well. Polymarket called the 2024 election better than most polls. Early market signals about famines or supply chain problems could save lives.

But is there a stable middle ground? Some scale where these markets stay informative without becoming causal? A volume where the thermometer works but doesn't heat the room? Nobody seems to have figured that out yet.

The recursive part: I wrote a novel about this (PARALLAX, [Chapter 1 is free](https://scm7k.com/read), [$9.99 on Amazon](https://www.amazon.com/dp/B0GSRX9ZTB)). I also made [a market on whether it sells 500 copies](https://manifold.markets/scm7k/will-parallax-sell-500-copies-by-en). So this post on a prediction market platform about a book about prediction markets, linking to a market about the book, demonstrates the thesis in miniature. You can trade on it if you want.

Full essay: https://scm7k.com/essay_reflexivity.html