Take for example cpsidp=20140100218301 in 2015 and match this person with his observations one year ago, you see that he was was reported. Male for mish==1-4, but Female for mish=5-8.
Similarly cpsidp= 20140100303701 was reported to be a white person for mish=1-6, but black for mish=7-8.
These kinds of changes in sex, race, and age are known to exist in the CPS source data, which is why IPUMS-CPS recommends users validate all links using these three variables. The common practice is to drop erroneous linkages, but it is ultimately up to the researcher to decide how such cases should be handled.