Most weather traders check one app. My bot reads 164 forecasts.

I am a data engineer 30 years in and I trade weather contracts on Kalshi as a hobby. For a while I did it the way everyone does. Check a forecast app pick a side and hope. Then it hit me that the market price is just that same single forecast everyone else looked at.

When Kalshi prices a temperature contract it is pricing whatever the average trader believes after glancing at one app. A weather ensemble does NOT work that way. The GFS runs 31 simulations from slightly different starting conditions. ECMWF runs 51. The newer AI ensembles NOAA AIGEFS and ECMWF AIFS add more. My bot pulls all four at once up to 164 independent forecasts for the same city and day.

Then it counts. If 118 of 164 members clear the threshold that is a 72% model read. If the market is at 40 cents that gap is the trade. It only fires when at least 3 of the 4 forecast families agree. Most scans produce zero trades. That is the design working NOT a bug.

It all runs on my RackNerd VPS on Ubuntu, logs every decision to SQLite, and the data sources are free.

What's everyone else here building? I'd like to hear what bots you run and where they fight you the most. For me the boring parts, fills and accounting, have been way more trouble than the model ever was.