145 of 783

September 24, 2026

Originally published at Poisson Labs.

I measured whether Jev, TypeSafe’s decision model, gives the same yes or no twice on the same request near its thresholds. Roughly one near-threshold decision in five flipped at 0.50 and 0.60 while the headline rate barely moved. The full note is on Poisson Labs, and the study repo is public at github.com/poisson-labs/jev-replay.