Adaptation

Weather derivatives, three edges deep

One system in two halves: a Bayesian inference engine maintaining a posterior over the true airport temperature, and a low-latency execution layer firing against it. Each edge died as market makers adapted, so I built the next one: observation latency, then a Markov decision layer over dying contracts, then the forecast itself. The first two traded live and were retired on an out-of-sample measurement that said the window had closed; the third was calibrated offline and never deployed.