Skills are read from full text only

The same job, counted three times

One vacancy is posted to a company's ATS, scraped into an aggregator, and mirrored on LinkedIn: three records, one real job. Counting all three inflates every figure — unevenly, overweighting employers who syndicate hardest.

Duplicates are matched on normalised employer + title + city. were flagged and excluded; none are deleted, so they stay auditable. A single recruiter can post one role dozens of times.

“Remote” is sometimes a fact, sometimes a guess

Applicant-tracking systems (Ashby, Lever) and several full-text sources carry a real remote flag. The rest do not — there it is inferred from the text, which under-reports. Read the Remote column per role with that in mind; it is flatter than the folklore, and there is no across-the-board tech premium.

Salary is partial, and sometimes imputed

Most employers publish no salary. Medians are computed only over postings that disclose one, and that n — usually far smaller than the role's volume — is always shown. Adzuna additionally imputes some salaries, so treat absolute levels with caution; comparisons between roles stay valid because the same bias applies to all of them.

Indeed and LinkedIn block direct access

Both return 403 to any automated request, including their RSS feeds. Their postings reach this dataset the legitimate way — through aggregators that already index them — which means a sample of those two boards, not their full pool. Stated, not hidden.