[fingerlocks]

If you’re treating a forecast as a single Bernoulli trial, wouldn’t that make them 60% accurate for the opposite of their prediction?

Which is a silly assumption; a forecast isn’t a single yes-no event. it’s not obvious to me that 50% is the worst case success rate.

Would be more interesting to compare their forecast to something like a long term NOAA forecast, but I don’t believe such a thing exists because calculating the future is very expensive.

[hbn]

GP did say it was just in reference to whether the winter will be warmer or colder than the last one, which is 50/50

In which case if they're 40% accurate, you can get 60% accuracy from them by assuming it'll be the opposite of what they say

If they could get their accuracy down to 0% you'd have perfect predictions!