Would it have made money?
Every hour the tool writes down the bets it would make, at the prices really on offer in the order book, including Polymarket's fee. Once a day settles, each bet is scored. Notional bankroll: $1,000.
Is it more accurate than the market?
For every settled day, how far each side's probabilities were from what happened (Brier score, shorter bar = more accurate). The strategy only works where the tool beats the market, and the idea says that should be the afternoon and evening, once the day's high has already been recorded.
Today
Each city's market for today. Our chance is the tool's probability for each temperature range; market is what the market's price implies. Buy at is the real price you would pay now. Would buy shows the bet the paper trader would place at these prices.