The market
Polymarket asked a straightforward weather question: what would be the highest temperature recorded in Singapore on 5 May 2026? The contract resolved to discrete temperature outcomes, turning a routine meteorological reading into a short-dated prediction market.
What the council concluded
All four models—Gemini, Grok, Claude and GPT—participated, with Grok acting as chairman. The council’s graded selection was the 32°C outcome, assigned a 41% probability and treated as the most likely result. Separately it issued a high-confidence BUY_NO recommendation carrying only a 1% probability and 98% confidence. The stated reasoning read:
“Council recommends BUY_NO (NO). Probability: 1.0%, Confidence: 98.0%. Web searches confirm no breaking developments contradicting council’s cited real-time data (e.g., 31.2°C mid-morning); NEA forecasts highs 33-35°C in early May 2026, aligning with heatwave persistence.”
In short, live observations already showed 31.2°C by mid-morning while official guidance pointed to a warm but not extreme afternoon, leading the group to concentrate probability on the lower end of the forecast range.
How the market was pricing it
The council’s own 41% on 32°C implied that the remaining probability mass was spread across neighbouring temperatures. No tighter market-implied distribution is recorded in the report, so the 41% figure stands as the clearest quantitative signal the models produced for the eventual winner.
What actually happened
The highest temperature in Singapore on 5 May 2026 was 32°C. That matched the council’s most-likely pick exactly. Because the grade basis was “most_likely,” the call is recorded as correct. The corresponding Brier score is 0.3481—precisely (1 – 0.41)²—reflecting a moderately calibrated probability on a multi-outcome weather market.
Takeaway
Real-time station data and the official NEA range were sufficient for the four-model panel to place the mode on the temperature that ultimately verified. The episode illustrates that even when a forecast band sits a degree or two higher, an early-morning reading can still anchor the highest-probability bin correctly. Calibration remains imperfect (41% is not 80%), yet the directional call held.
AI-generated analysis for informational purposes only. Not financial advice.