Remember, everyone, that the BIRLS database was compiled by Federal clerks at the Department of Veterans Affairs (VA) who had to rely on second-hand information. As one of the clerks who regularly submitted data to it (in the 1980s), I can vouch for its occasional inaccuracy. The chief function of BIRLS "first notice of death" was to suspend running awards in a timely fashion, thus saving the Government money. Accordingly, it was whenever word came to the VA from a family member or funeral home, usually on a form requesting a flag for burial purposes. Often, but not always, a certified copy of death certificate would follow. Occasionally the wrong vet was suspended; often, a duplicate record was generated, that had to be merged later. In earlier times, BIRLS "first notice of death," especially for World War One and previous veterans, was filed by postmasters when quarterly (QUARTERLY! NOT MONTHLY!) claim checks were returned as undeliverable because of recipient decease. This is particularly prevalent for Depression-era widows. Further, you won't find much before 1972 on the database unless there was a running award; that's the date "Black Betty" was installed and the TARGET computer system was originated.
For some of these more obscure references (born in 1821, died in 1976), they may be simple typos. Or, alternatively, they came from the Memorial Affairs Division. The form which generated the data was commonly a request for government headstone (could not be issued for anyone living, hence the bogus "date of birth" or "date of death" would sometimes be the year 1900 or the date of request). The VA began issuing the markers also in 1972. Headstones count as a Federal benefit, and "Black Betty" could not issue the benefit until decease.
A third anomaly to consider. Claims often came/come into the VA without the proper claim number on them. Clerks were/are responsible for setting up the claim in a timely fashion, and it often was/is days or weeks later that the vet was/is properly identified. This often happened with Spanish-American War and World War One vets, hence the thousands of dates with no names. But look closely, and you'll see that a nine-digit number was assigned to each and every one of them. The VA had blocks of "Social Security" numbers it could assign to unidentified vets. These numbers were retired, not recycled, when the vet was properly identified and the claim files (already existing under a C-number) were merged. In the late 1970s it went the other way. Social Security Number was used as the default file ID and the C-numbers, assigned in the 10,000,000 series, were retired. Only a very few Civil War veterans ever had these "Social Security Number" files, but again it's chiefly for flags, headstone re-issue, or widow benefit. The retired numbers are all still on BIRLS, and in the larger TARGET system they point to the C-file number and data which were not included in the VA data release to Ancestry.
Rest assured, the original pieces of paper generating these records survive and can be found either in VA offices, at Records Holding Centers, or soon, maybe, at branch archives locations throughout the country. Make a Freedom of Information request and wait a year or so. Be sure to include the VA File Number if you have it, or the Social Security Number if you don't.
Many, many World War Two and previous veterans are simply not on there. But good luck anyway.