What the market asked
The Polymarket event was straightforward on the surface: what would be the highest temperature in Chongqing on 13 August? Resolution turned on a narrow, mechanical rule—the highest whole °C value that appears in the WU ZUCK Daily Observations table for that calendar date. In practice that meant a set of discrete temperature buckets rather than a continuous forecast, and small differences in source tables or rounding could decide the winner.
What the council concluded
All four models—Gemini, Grok, Claude, and GPT—participated, with Grok acting as chairman. The council’s formal action was SKIP. It assigned only 33% probability to its most-likely outcome and recorded 58% confidence overall. The quoted rationale was explicit:
“Council recommends SKIP. Confidence: 58%. Resolution criteria are strict: the highest whole °C value that appears in the WU ZUCK Daily Observations table for calendar date 13 August 2026.”
Despite the skip, the graded pick under the “most_likely” basis was the 36°C contract: “Will the highest temperature in Chongqing be 36°C on August 13?” at 33%. In other words, the panel saw 36°C as the mode of a fairly flat distribution and did not believe it had enough edge or conviction to recommend a position.
How the probabilities sat
With the top outcome carrying just one-third probability, the remaining two-thirds was necessarily spread across neighboring temperatures. That is a classic low-edge weather setup: short-range guidance can be informative, yet the exact whole-degree print that a single official table will show remains noisy. The council’s 58% confidence and the SKIP recommendation were consistent with that view—they were not claiming 36°C was unlikely, only that the edge did not clear their bar for action.
What actually happened
The market resolved to exactly the council’s most-likely bucket: the highest temperature in Chongqing on 13 August was 36°C. On the grade basis used (most_likely), the council was therefore correct. The Brier score on the graded pick was 0.4489, which is simply (0.33 − 1)². A correct call at a modest probability is still a correct call; it is also a reminder that calibration and decisiveness are separate skills. The panel identified the right mode and still chose not to bet it.
Takeaway
This report illustrates a useful distinction between “what we think is most probable” and “what we are willing to act on.” Strict, table-based weather resolutions reward precise source knowledge more than broad climatology. The council’s caution (SKIP at 58% confidence) was coherent with a 33% mode; the fact that the mode won does not turn the skip into a mistake, nor does it turn a 0.45 Brier into a triumph. It simply shows that, on this day, the highest-probability cell on a flat forecast board happened to be the one the official table printed. Honest grading keeps both the directional success and the weak probability in view.
AI-generated analysis for informational purposes only. Not financial advice.