Source: U.S. Census Bureau, "Business Formation Statistics," Announcements notices, under "Latest Monthly Business Formation Statistics Report": https://www.census.gov/econ/bfs/index.html
Quoted: "As of the January 2026 monthly release, applications associated with internet sales are excluded from the high-propensity (HBA) and corporation (CBA) application series. This methodology change was originally scheduled to occur during the November 2025 monthly release. The HBA and CBA definitions are updated for the entire time series."
Also quoted: "Due to significant differences between 2017 NAICS and 2022 NAICS vintages, the entire time series was restated to 2022 NAICS during this annual update. The January 2026 monthly data publication was the first release using 2022 NAICS."
And: "Due to a scheduled outage of the Internal Revenue Service's online Employer Identification Number (EIN) Assistant from December 30, 2025 until January 5, 2026, applicants were unable to apply for an EIN through the web application. As a result, the Business Formation Statistics Monthly data for December 2025 and Weekly data for Week 53, 2025 are impacted."
What the page establishes: three separate discontinuities land on the same date boundary — a definitional change (internet-sales applications dropped from HBA/CBA), a full-vintage restatement (2017 to 2022 NAICS), and an administrative gap (IRS EIN Assistant closed Dec 30–Jan 5). None of them is a change in entrepreneurial behavior.
Two consequences I draw (my interpretation, not the author's):
Because the definitions "are updated for the entire time series," a BFS file downloaded before January 2026 will not agree with one downloaded now, for months well before 2026. Any backtest or dashboard built on cached BFS data is now silently stale; the series did not just move forward, its past was rewritten.
A December 2025 dip in applications is plausibly an artifact of the EIN Assistant being offline, not an economic signal, and the page says so explicitly.
Practical rule: for analyses crossing December 2025, prefer a freshly pulled, restated series, and treat "business applications rose/fell" claims dated around this boundary as unproven until the author says which vintage and definition they used.
Limit: these are notices, not data. They tell me something was changed; they do not quantify it. The magnitude of the internet-sales exclusion (how many HBA/CBA applications are internet-related) is not on this page and needs the data tables or technical documentation. I have not recomputed the series.