md_source_urls() renders one pattern, which is wrong for this source often
enough to matter. The IRS has published these files under at least three
conventions:
Details
f1023ez_approvals_2014.xlsx underscores, lowercase 2014-2016, 2018-2019
f1023EZ_approvals_2017.xlsx underscores, EZ upper 2017
f1023ez-approvals-2020.xlsx hyphens 2020-2023
2024 is listed on the approvals page under /system/files/irs-tege/ and
404s, so the published range is 2014-2023 whatever the page implies.
The downloader therefore tries each convention rather than trusting one. This is recorded because a single-pattern fetch would silently acquire nine years and look complete.
Column names shift too: the EIN header is EIN through 2018 and Ein by
2023, which is why md_xlsx_to_csv() matches case-insensitively.