What the market asked
The Polymarket event asked a narrow weather question: what would be the highest temperature in Wellington on May 7? Outcomes were framed as discrete Celsius buckets (for example, whether the high would be 18°C, 19°C, and so on). The council issued its report on 2026-05-06, the day before the observation date, with all four models—Gemini, Grok, Claude, and GPT—participating and Grok acting as chairman.
What the council concluded
The council’s action was an unambiguous SKIP, logged at 100% confidence and a council probability of 0.0% on taking a position. Its written reasoning stated:
Council recommends SKIP. Confidence: 100.0%. All evidence—council analysis, cited sources, and fresh web searches (MetService extended forecast showing 17°C high with northerlies/showers; WeatherWatch 18°C)—points to a high of 17-18°C, far above
On the graded “most likely” basis, the council’s preferred bucket was 18°C, assigned a 45% probability (“Will the highest temperature in Wellington be 18°C on May 7?”). In short, the models agreed the temperature would land in a cool band and that no contract offered enough edge to justify a bet.
What the market was pricing
The report does not publish a full order-book snapshot, but the council’s own distribution is the clearest signal we have of how the question was being framed the day before: 18°C as the modal outcome at 45%, with surrounding cooler readings (17°C and nearby) treated as the realistic range. That concentration left little probability mass on a 19°C print—the outcome that ultimately resolved.
What actually happened
The winning outcome was 19°C (“Will the highest temperature in Wellington be 19°C on May 7?”). The council’s graded pick of 18°C was therefore incorrect (council_was_correct: false). The Brier score on the graded pick was 0.2025—respectable for a 45% call that missed by a single degree, but still a miss. Forecast sources the models trusted (MetService ~17°C, WeatherWatch ~18°C) under-ran the observed high by one to two degrees.
Takeaway
This is a clean illustration of both the value and the limit of a SKIP. The council correctly sensed that the edge was thin and refused to force a trade; that discipline is sound. At the same time, anchoring hard on 17–18°C forecasts left the 19°C bucket under-weighted. Single-degree weather markets are unforgiving: model guidance, even when consistent across MetService and WeatherWatch, still carries residual error, and a one-degree miss flips the contract. Honesty about the miss matters more than the neatness of the SKIP. When the next coastal temperature market appears, the useful update is not “never trust the forecast,” but “keep a fatter tail one degree above the modal guidance.”
AI-generated analysis for informational purposes only. Not financial advice.