What This Observatory Measures, and What It Won't Claim

alldone.jobs reads Olive’s job-postings corpus at a grain no existing labor-market index publishes at once: geography down to the ZIP, parsed from the free-text locale field carried on the large majority of closed U.S. postings; the full text of each listing, not only its title and a coded occupation; and occupations defined from the bottom up, by what a posting actually says, rather than by the ceiling of a federal taxonomy drafted before generative tools existed.

The corpus is younger than it looks. Its usable history begins in February 2026, several months of depth at the time of this note, so the research program avoids long trend lines and does not compute its own seasonal adjustment. Where a series needs a seasonal norm, this publication borrows the Bureau of Labor Statistics’ published factors and says so on the chart, rather than treat a few months of data as enough to derive one.

Every finding here is committed to publishing what the data shows, including a finding that an expected pattern did not appear. A neighborhood map of where hiring changed is worth publishing whether or not a national trend sits underneath it — the map is the object; the trend is only sometimes the story.

The standards this note refers to — flow over stock, share of total rather than raw counts, suppression below thirty postings, a named vintage on every number — are listed in full on the methodology page, and every article this publication runs links back to them.