US employers have to file a WARN notice roughly 60 days before a mass layoff. The notices are public, but they are scattered across 48 state agencies in a mix of spreadsheets, PDFs and web tables, and nobody keeps them together.
CanaryWhistle collects them into one corpus and turns them into a forward calendar: which jobs are scheduled to end, at which company, in which city, on which date. On the day this screenshot was taken it listed 27,442 jobs ending within 90 days across 234 notices.
What was hard
The data, not the design. Two states file by county rather than centrally, one state’s export put a data row where the header should be, and an early city parser double-counted more than a thousand California rows. The published total went down after that was fixed, because it had been inflated. There is now a corpus audit in the build chain, because a data product that is confidently wrong is worse than no product.
