What the market asked
The Polymarket event asked a straightforward weather question: what would be the highest temperature in Karachi on May 7? The market resolved across discrete temperature buckets. The council’s graded pick focused on the outcome “Will the highest temperature in Karachi be 36°C or higher on May 7?”
What the council concluded
On 2026-05-06 the four-model council (Gemini, Grok, Claude, GPT), chaired by Grok, issued a BUY_YES recommendation. It assigned a 37% probability to the broader YES side and 75% confidence in its process. Its most-likely single outcome was the 36°C-or-higher bucket at 30%.
The stated reasoning was concise: “Council recommends BUY_YES (YES). Probability: 37%, Confidence: 75%. No breaking developments contradict the PMD heatwave forecast of 35-38°C max in Karachi, confirmed in latest reports from May 6.” The models essentially treated the Pakistan Meteorological Department heatwave guidance as still intact and saw no contradictory short-term data that would force a downward revision.
What the market was pricing
The council’s own 30% on the 36°C+ bucket already implied that the market (and the models) saw substantial probability mass on cooler outcomes. A 37% YES probability on the heatwave side left the majority of probability on temperatures below that threshold. In short, neither the council nor the market treated 36°C+ as a high-conviction lock; it was simply the single outcome the models judged most probable among the available buckets.
What actually happened
The winning outcome was “Will the highest temperature in Karachi be 34°C on May 7?” The council’s graded pick (36°C or higher) was therefore incorrect. The Brier score on the graded basis was 0.09—respectable calibration given the low base-rate probability assigned, but still a miss on the directional call that mattered for the report card.
The PMD 35-38°C guidance that anchored the reasoning did not verify. Either the forecast itself had residual error, local marine or cloud effects suppressed the peak more than anticipated, or the “no breaking developments” check on May 6 simply failed to capture the final evolution of the airmass overnight into the 7th.
Takeaway
Weather markets punish over-reliance on a single official forecast range, especially when that range sits only a couple of degrees above the eventual observation. The council correctly avoided extreme conviction (30% most-likely, 37% YES), yet still anchored too heavily on the heatwave bulletin without sufficient discount for typical forecast busts at the daily maximum. When the verifying temperature landed two degrees cooler, the graded pick failed. Transparent post-mortems of exactly these misses are how the council’s calibration record improves.
AI-generated analysis for informational purposes only. Not financial advice.