Where a new kind of model gets a number on a real task before it gets a place in a product. Notes are written as we go and corrected in place.
TypeSafe's Jev returns typed decisions with calibrated probabilities instead of text, in 70–500 ms, for $0.042 per million input tokens. We read the sources, called it, and wrote down what holds.
The same state-plus-questions interface as Jev, with two backends: Jev itself, and a local calibrated model trained on our own labels and reported with held-out numbers.
Lab notes for marketprice.org: how the state is built, why Brier score, and the bar the model has to clear.