How we know

One rule decides who is here

A company is indexed if its applicant tracking system exposes a real publish timestamp. That single requirement is what makes “open for 47 days” a fact rather than a guess, and it is why some large employers are missing: their boards publish no date we can stand behind, so we would have to invent one.

Where a system offers more than one date, we take the one that means first published. Greenhouse exposes both first_published and updated_at; the second moves every time a recruiter edits a word, which is how a role open four months comes to look like a role opened last week on other sites. Eightfold tenants get creationTs rather than postedTs, because some of them refresh the posted date in bulk and a requisition open sixteen months would otherwise report as opened yesterday.

What we refuse to publish

• Dates the employer rewrites. Apple restamps some requisitions at request time. We detected them by fetching the board twice nine minutes apart and comparing, and we drop those listings rather than publish an age we know is manufactured.
• A board we cannot read confidently. Google publishes its date inside an unlabelled positional structure. The adapter resolves it against two independent checks and fails the entire board if they disagree, rather than guessing at a field and being wrong about three thousand roles.
• A board that is not who it says. A token can return real jobs and belong to a different company. One board we tried under “Wise” served a US insurance field-sales outfit. Every board is identity checked before anything from it is published under a company’s name.
• Dates that cannot be true. Every timestamp is checked against the date its applicant tracking system began operating, and against the clock. A listing dated before its own board existed is dropped.

What is ours, and labelled as ours

Job function, level and the “standing” marker are our readings of the job title, not fields the employer filled in. They are kept in their own place, separate from the employer’s own department and ladder labels, so an inference is never mistaken for a citation. Where a title says nothing useful we leave it unclassified rather than guess.

We do not normalise one company’s seniority ladder onto another’s, and we do not convert or annualise published pay. Both would present our arithmetic as the employer’s number.

What a search is searching

A search here matches job titles, company names and team names. That is a choice, not a shortfall: a title says what a role is, while the posting underneath it says everything a recruiter could think of. Measured on live boards, the word “partner” appears in 85% of one employer’s descriptions and 100% of another’s, and “remote” in 98% of a board with no remote roles on it. Searching all of that by default would make most ordinary words match most of the index.

But the opposite is just as true, and it is the one you would hit: nothing puts “Kubernetes” or “CUDA” in a job title. So we do read the postings, and when a search finds too little we say how many more mention your words in the text and let you ask for those — labelled, so a passing mention is never presented as the job itself.

We store that text to search it, not to republish it. Every listing here links to the employer’s own posting, which is the copy worth reading and the only one that is current. Not every board hands it over: some publish the posting only one page at a time, and we fetch those slowly rather than hammer an employer’s hiring system, so at any moment some listings have been read and some are still queued.

A second question: does the role belong

Passing the timestamp test is not enough on its own. Some boards publish perfectly good dates for work this index is not about — one requisition per retail store, per warehouse shift, per bank branch. Those are filtered or excluded, because leaving them in means every figure on the site is quietly describing them instead.