I've been recording the trades on every Kalshi weather bracket since late July, and the top five levels of each bracket's order book every 10 seconds. Below is what I measured, what I believe it to mean, and an example from an afternoon in Oklahoma City.
When the price came up to an unusually large resting order, that order was usually gone before it filled. In my first week of recording highs markets, pulled orders of this size outnumbered filled by about five to one. Some of it appears as routine trading; some big orders look placed to make other traders think there's a serious buyer or seller at that price. Betting the same way as a big order lost 4.4c a contract on average, after fees.
Most bursts of big buying in highs markets gave the move back within the hour: 58% of 11,095. Buying right after a burst lost 4.5c a contract on average, after fees; betting against it lost 0.9c.
3,619 times since late July a side's price fell 20c or more within 5 minutes. After a thin drop, typically only 3 contracts traded at that price in the next 10 minutes: the price was there for a moment, and almost nobody got it.
In highs markets, people who crossed the spread lost roughly what it cost them to trade, and the biggest orders did slightly worse than the small ones.
A minute-by-minute look at the 89-90°F bracket: a 551-contract offer at 70c, a 19c drop on 72 contracts, a 456-contract bid that appeared and vanished, and 456 contracts bought at market two seconds later.
Trying to extract tells from the order book and other market participants has been a huge upgrade to my trading, as opposed to focusing on trying to improve forecast accuracy.