Seed: 289,640 roles, built 2026-08-28

Prebuilt roles for the slow half of a scan, so a new user is not sat watching a counter.

289,640 roles from 8,779 employer boards, built 2026-08-28. This is passes two to four of a scan: Ashby, Greenhouse and apply.workable.com. Pass one is not here and never will be, because anyone can have it themselves in about five minutes, fresher than any file.

Use it

job-radar seed load https://github.com/maccydee/job-radar/releases/download/seed-latest

That reads the index, works out which shards your locations.countries needs, and downloads only those. A UK reader takes 38MB and 41,038 roles rather than 242MB and 289,640.

What is in it

Every field the adapters produce, including the advert text, because without a description a role cannot be scored, cannot be checked against a dealbreaker and cannot have a CV written from it.

What is deliberately absent: score, fit, reasons, application status. Those are answers to a question only your own config asks. A seed is a saved fetch, not a saved decision, and seed load screens what it reads against your config exactly as a scan does.

Shards

One per country, plus two that go to every reader: unplaced for roles whose country could not be read, and multiple for roles open in more than one country. Neither is evidence that a job is somewhere else. A role open in London and New York is a UK role.

Reader	Download
UK	38.3 MB
DE	27.7 MB
IN	29.2 MB
US	133.8 MB
everything	242 MB

What it does not do

It does not replace a scan and it does not make one faster. None of these board APIs has a "changed since" parameter, so a scan still reads every board. Roles die in days and this file is a day old at best. Run a scan anyway; its answer wins on every field.

docs/SEED.md has the format and how to build your own.