The Question
Polymarket asked: “Highest temperature in Seoul on June 13?” The contract resolved to discrete Celsius bins. Prediction Council published its report on 2026-06-11, two days before the event, with all four models (Gemini, Grok, Claude, GPT) participating and Grok acting as chairman.
What the Council Concluded
The council’s graded pick—the outcome it labeled most likely—was “Will the highest temperature in Seoul be 26°C on June 13?” at 25% probability. Separately it issued a high-confidence BUY_NO recommendation (1% probability, 95% confidence) against a much colder outcome. Its published reasoning stated:
“Council recommends BUY_NO (NO). Probability: 1%, Confidence: 95%. Recent KMA and station-specific forecasts confirm 27°C max for June 13 at RKSI with no cooling mechanisms to reach 22°C. Historical June averages at Incheon (~25-28°C) align with this. The $0.003 YES”
In short, the models saw the official forecast locked on 27°C, treated 22°C as essentially impossible, and still assigned their single highest probability mass to the adjacent 26°C bin.
How the Market Was Pricing It
The only price fragment preserved in the council note is a YES contract trading at $0.003—consistent with a deep out-of-the-money tail. The broader order book is not recorded in the graded report, so we cannot reconstruct the full implied distribution the market was showing on 11 June. What is clear is that the council’s own 25% on 26°C already implied a relatively flat distribution across the mid-20s, even while its narrative repeatedly cited the 27°C forecast.
What Actually Happened
The market resolved to “Will the highest temperature in Seoul be 27°C on June 13?” The council’s most-likely selection was therefore incorrect. Graded on the most-likely basis, the call earned a Brier score of 0.0625—exactly (0.25 – 0)². The miss was only one degree, yet in a discrete-bin market a one-degree error is a full miss.
Takeaway
The models correctly identified the meteorological regime and the official 27°C forecast, then diluted their probability onto the neighboring bin. That small hedging decision turned a near-certain directional read into a wrong most-likely call. Weather markets this granular punish even well-calibrated forecast agreement when the final digit lands one notch away. The 95% confidence attached to the BUY_NO leg was justified; the 25% crown placed on 26°C was not. Transparency about the miss is the point of publishing every graded report.
AI-generated analysis for informational purposes only. Not financial advice.