How to read this board
The market number
The market figure is what Polymarket's odds imply for that city and that day as they stand when you open the page. Prices are collected every 15 minutes, so it is minutes old, not hours. Every strike trades as a probability (20 degrees, 21 degrees, and so on), so the figure is those prices weighted into one temperature. It is not published anywhere. It is arithmetic on the odds we read.
Only strikes priced between 2 and 98 percent go into that average. A strike at 0.1 percent or at 99.9 percent is the market saying the day is already decided, and letting it into the average would drag the number towards a certainty the rest of the ladder does not share. Between three and ten strikes usually qualify, and if fewer than three do, we say we hold no market number for that city rather than quote a number built on one strike.
The forecasts beside it are each model's newest reading for the same day. Models publish on their own schedules, roughly every three to twelve hours, so those readings are not all the same age, and the heading on the board tells you how old the oldest of them is. A model that has not revised is not wrong, it simply has not published since.
If we hold no odds at all for a day, the column shows the strike the market settled on and labels it settled. That is an outcome, not a forecast. A dash means no price was recorded for that city and day.
Where the forecasts come from
Forecasts come from Open-Meteo, which republishes the national weather services: ECMWF, DWD ICON, NCEP GFS, HRRR and NAM, MeteoFrance, UKMO, JMA, CMA, KNMI and others. Ten to eighteen models cover a given city, depending on where it is.
The models column is their plain average. There is no weighting and no clever ensemble, on purpose: the point is to show what the forecasts said, not to build a better forecast. The spread column is simply the hottest forecast minus the coolest one.
Gap and spread, and where to look first
The gap is market minus models. Positive means the market is pricing a warmer day than the models expect. The spread is how far the models disagree with each other.
Read the spread first. A wide gap against a wide spread means very little, because the forecasts themselves do not agree. A wide gap against a tight spread is the case worth looking at: the market and a confident group of models disagree about the same day.
Colours are graded in Celsius, so a band means the same amount of disagreement in both sections.
Why the board has two sections
Celsius cities quote exact strikes one degree apart. The US cities quote two degree Fahrenheit buckets (80 to 81, 82 to 83), so one bucket is counted at its midpoint and carries about half a degree of slack. Those pages say so.
We do not convert one market into the other unit on the board, because a Fahrenheit bucket converted to Celsius reads as 21.1 to 21.7, which looks like precision the market does not have. A city page opens in the unit its own market trades in, with a switch if you want the other one.
Which day opens by default
The day with the most cities still being priced, which is usually today or tomorrow. A day that has begun loses its ladder, so it becomes less useful to look at. Every day in the window is in the day picker.
How fresh the numbers are
Prices are collected every 15 minutes and forecasts every 30. The site sits behind a cache that serves a copy at most a minute old, so a page can be a minute behind the newest reading. The footer prints how old the prices are on every page.
A city page states the moment of its latest odds reading, and says when the oldest of the model readings it is showing was taken, rather than presenting numbers of different ages as if they were simultaneous.
What is deliberately not shown yet
Scoring each model against what actually happened, city by city, is collected but not displayed: a handful of city days per city is not enough to rank models inside one city. The columns that show how the market itself did against reality are switched off for the same reason. They come back once there is enough history behind them.
What the market priced 12, 24, 48 and 72 hours before each day is archived as it happens, even though nothing shows it yet. That archive is what the day-by-day comparison of market against models will be built on, which is why the site keeps collecting it while the pages quote the current market instead.