The problem with consumer health sensors is they have both high random error and inconsistent systematic error. When your SPO2 sensor gives you 92% one minute and 98% the next while you're sitting still and it is almost always 2% under, you're not getting "noisy but usable" data - you're getting garbage.