Read from the source, counted in the open_

Companies whose own pages load a vendor's scripts, embeds or tags, or whose domain's mail and hosting records (SPF includes, DKIM selectors, MX, CNAMEs) point at the vendor. One company counts once, whichever the source. It measures reach: the tool is in the building.

Companies whose own job postings say the team works in the tool, read hourly from the applicant tracking systems behind career sites. Candidate wish-lists and apply boilerplate are excluded; each company carries a confidence level and the quote. It measures intent: the team that runs the tool is hiring.

They answer different questions and overlap little. Adding them would produce a number nobody can check. Every page shows both and the overlap.

From the company's own website: name, what it does, whom it sells to, certifications, languages and contact points are extracted from its pages with a confidence score per field. Headcount comes from LinkedIn where known and is otherwise estimated and marked so.

Every page carries the date its numbers were computed. Pages are rebuilt nightly; the previous day's headline numbers are kept so the change is visible. Career sites are read hourly, websites re-crawled on a rolling schedule, DNS re-swept in cycles.

Personal names or personal e-mail addresses on public pages; private contact data; anything from behind a login. Public role addresses and phone numbers that a company prints on its own site are counted, not listed.

Job boards, portals, directories and aggregators. Staffing firms appear only for their own hiring and stack, never for their clients'.

Yes. Counting is free in the product: run the same filter and you get the same number. Every page also has a JSON view of its data at /api/pseo/<family>/<slug>.