Three steps, repeated about 18,000 times a day.
The model treats every unknown as a distribution. Each observation is weighed by how surprising it would be under that distribution, and the result is a posterior: narrower, shifted and honest about what it still doesn’t know.
Here it is guessing how many days remain until the first frost. Press the button to give it three more weeks of weather.
Try it
Update a prior
Principles
We write the prior down first.
Before any data is pulled, the model commits to a distribution. It is timestamped and kept, so we can be held to it.
We act on confidence, not conviction.
A narrow posterior earns attention. A strong feeling does not. The squirrel has strong feelings about most things.
We publish when we’re wrong.
Every letter lists the three beliefs that moved furthest from their prior, and what moved them.