← All posts
Weather·August 26, 2026Council was right

AI council correctly favored 31°C for Hong Kong’s August 25 high

The market question

Polymarket asked what the highest temperature in Hong Kong would be on 25 August 2026. Resolution followed the Hong Kong Observatory’s Absolute Daily Max (°C) at the headquarters station (Daily Extract), recorded to 0.1°C and then binned into the integer degree that contains it—so “31°C” means any reading from 31.0 through 31.9.

The contract set was therefore a set of mutually exclusive temperature bins rather than a single binary. The report under review was issued on the day of the event itself.

What the council concluded

Three models participated: Gemini, GPT, and Grok, with Grok acting as chairman. The council’s formal action was BUY_NO, carrying a stated probability of 55% and confidence of 65%. At the same time it identified a single most-likely bin:

Council recommends BUY_NO (NO). Probability: 55%, Confidence: 65%. Resolution is the Hong Kong Observatory’s Absolute Daily Max (°C) for 25 August 2026 at the Observatory headquarters station (Daily Extract, to 0.1°C precision), binned into the integer degree that contains it (31°C = 31.0–31.9).

The graded pick—the outcome treated as most likely—was the 31°C bin, assigned 54% probability. In short, the council did not treat any other single integer degree as more probable than 31°C, even while framing its overall stance as BUY_NO.

How the probabilities sat

No separate order-book snapshot is preserved in the graded record beyond the council’s own numbers. What we have is internal: 54% on the 31°C contract as the modal outcome, a 55% headline probability attached to the BUY_NO action, and 65% confidence in that framing. Because the grade basis was “most_likely,” performance is scored on whether the 31°C bin actually resolved true and on the 54% probability attached to it.

What actually happened

The winning outcome was exactly the bin the council had ranked first: the Observatory high fell inside 31.0–31.9°C. The council was therefore correct on the graded pick. With a forecast probability of 0.54 on the realized outcome, the Brier score is (0.54 − 1)² = 0.2116—consistent with a modestly confident, single-bin call that landed.

Takeaway

Weather markets resolved to a precise official series leave little room for narrative after the fact; either the reading lands in the named bin or it does not. Here the council’s modal forecast matched the Observatory print. The Brier score of 0.2116 reflects a probability that was directionally right but far from certain—an honest mid-range success rather than a high-confidence lock. The episode is a reminder that, on tightly specified meteorological contracts, simply identifying the correct integer degree as most likely is already a usable edge, provided the probability is not overstated.

AI-generated analysis for informational purposes only. Not financial advice.

Every council report is graded against the real outcome and published — the good calls and the bad ones.

AI council correctly favored 31°C for Hong Kong’s August 25 high — Prediction Council