Even if it's polluted, you can mark it as is, with reasoning and keep it up, so you can learn what dirty/bad data is, and build quality detectors with the information contained in/with the "dirty" or "low quality" data set.