Resolving to Loud, Not Right: Validating a Smart Monitor's Cry-Classification Label Against What Actually Soothed the Baby