The market
Polymarket asked a straightforward weather question: what would be the highest temperature in Chengdu on May 6? The resolution path centered on whether the daily high would clear specific thresholds. One of the key contracts was whether the high would reach 28°C or higher.
With the report filed on 2026-05-05—one day before the event—the council had access only to short-range forecasts and model guidance, not the official observation.
What the council concluded
All four models (Gemini, Grok, Claude, GPT) participated, with Grok acting as chairman. The council’s formal action was SKIP, at 95% confidence and an implied council probability of just 1% on taking a position.
Their reasoning was explicit:
“Council recommends SKIP. Confidence: 95.0%. Recent forecasts confirm council views: timeanddate shows 26°C high overcast, AccuWeather 31°C cloudy/warm, and model ensembles (ECMWF/GFS ~26°C, CMA possible showers) split but cluster in 26-31°C ran”
In short, the available guidance was spread across a 5°C band that straddled the 28°C line. That dispersion made a high-conviction directional bet unattractive. Within that uncertain band, the council still identified “Will the highest temperature in Chengdu be 28°C or higher on May 6?” as the single most likely outcome, assigning it 45% probability.
How the probabilities looked
The 45% figure on the 28°C-or-higher contract was the council’s internal most-likely assessment, not a market mid-price snapshot. Because the underlying forecasts themselves disagreed—some models and services pointing near 26°C, others nearer 31°C—the council treated the threshold as essentially a coin-flip zone and declined to recommend a trade. The SKIP reflected that the edge, if any, was too thin relative to forecast noise.
What actually happened
The winning outcome was precisely the contract the council had flagged as most likely: the highest temperature in Chengdu on May 6 was 28°C or higher. Graded on the most-likely basis, the council was marked correct. The corresponding Brier score was 0.3025—exactly (1 − 0.45)²—reflecting a moderate probabilistic hit rather than a sharp one.
Takeaway
This case illustrates the difference between directional accuracy and actionable edge. The council correctly identified which side of a close weather threshold was more probable, yet the same forecast scatter that produced that modest 45% lean also justified the high-confidence SKIP. In weather markets decided by a few degrees, ensemble disagreement is often the dominant signal; when models and public forecasts straddle the strike, sitting out can be the disciplined call even when the eventual winner matches the internal favorite. Honesty about that uncertainty—and willingness to publish both the SKIP and the graded result—is the point of the exercise.
AI-generated analysis for informational purposes only. Not financial advice.