The 2018 data looks great, with a couple exceptions (HMDAVision is updated once a month with any updates and corrections made available by the CFPB):
Example 1: 2018 Applicant Income has the lowest percentage of empty fields of the past 5 years, which augurs well for good data quality, but the incomes include negative values for the first time. This is consistent with definitions of gross income, but may be a surprise if you're not expecting it. The data also includes some quite high and quite low outliers. It's hard to know if outliers are accurate; HMDAVision helps by allowing you to filter out the applications which didn't reach a credit decision, and may therefore only be self-reported.
Example 2: There is a conflict in some cases between the FIPS code and the state abbreviation in the data – our decision in these cases has been to go with the more detailed information contained in the FIPS. HMDAVision is based on microdata – this allows granular filtering to see applications by lender, by MSA, County, or Census Tract. Or by any of the other dimensions in the data, in any combination. If you ever detect something you think is an anomaly in the data, let us know, and we'll investigate further.