The whole evaluation set, free, no signup and no order. Five files cut from the index on 2026-10-10 — including the board-by-board completeness verdicts, which is the file the argument actually rests on. What the index proves, and what it does not, is the page to read beside them.
Every board in the index, not a sample of them
One row per board: what was fetched, what the vendor published as its own total, and the verdict. Boards that came up short are graded count mismatch and kept, not dropped.
The provable share, derived rather than typed
Boards, roles and verdicts per ATS platform, with the count-publishing share computed on the last row. This is where the headline percentage comes from.
Graded openings and closures
A stratified cut across all three confidence grades, so the sample shows the grading rather than hiding behind the high-confidence rows. Each row carries the sentence explaining its grade.
Postings, one per company
The delivery schema, on real rows: company, ATS, title, location, department, the vendor's updated date, and the employer's own URL.
The history, including its holes
Every recorded day since the first. Short history is a known limitation and the days that failed to record are visible rather than interpolated. Its two board columns count boards that returned at least one role, which is why they read lower than the directory total above; the difference is the reachable-but-empty boards and the unreachable ones.
Deterministic: the same index produces the same pack, so a pack regenerated after a coverage change can be diffed against the one you already hold. Regenerated nightly, so these numbers track the index rather than a date somebody typed.
8,459 of 15,756 boards (53.7%) sit on an ATS platform that publishes its own job total, so those fetches are checked against the vendor’s figure board by board, every day. The rest are labelled unverifiable rather than presented as complete — that distinction is the product.
Of the 15,756 boards listed, 14,881 returned at least one open role on the cut date, 677 were reachable and simply had nothing open, and 198 could not be reached at all and say so on their row. The buyer page headlines 14,881 because that is the number with roles behind it; this pack lists all 15,756 because a board with nothing open is still coverage.
Of those 677 empties, 332 were checked against the ATS
vendor’s own published total and it was also zero — employers
provably hiring nobody on the cut date, not boards that failed to read.
Filter boards-completeness.csv to roles_fetched = 0
and verdict = complete to see them.
888,945 posting rows collapse to 803,959 unique roles, a 9.6% dedup collapse across 12 platforms. History runs 2026-08-29 to 2026-10-10, 39 days recorded, forward-only: it is not backfilled and cannot be.
No personal data. The stored schema is company, ATS token, job id, title, location, department, updated date and URL — no candidates, no applicants, no names, no contact details. Every endpoint is the ATS vendor’s own documented public API: no login, no session cookie, no CAPTCHA, no IP-block circumvention.
Raw postings are self-serve on the Apify Store at $5.00 per 1,000 unique deduplicated results, including a verified-boards reader that returns only the boards whose count can be proven. The full graded event feed is delivered as daily CSV and scope decides the price, so that one starts with an email.
Name the employers or the platforms that matter and the answer comes back plainly, including when it is no.