Job postings from public ATS boards

Provable Hiring Data

Every job scraper hands you rows and lets you assume they are all there. This one checks each board against the vendor’s own published record count, and tells you which half of the feed it can prove.

Measured 2026-09-14 Index 1,195 boards across 7 ATS platforms 703 count-checked, all matched Hospital nursing index 123 boards Rebuilt daily 06:23 and 06:50

The index, as measured

61,303Unique roles64,927 postings before deduplication.

3,624Duplicates removed5.6% of the corpus, deduplicated across platforms.

703Boards count-checkedEvery board that publishes a total matched it. 0 mismatches.

201sFull rebuild9 boards errored out of 1,235 listed.

Why completeness is the whole argument

Duplicates and missing records are the same failure

Look at the open issues on any job-data API and the same two complaints repeat: duplicates, and jobs missing from the database. Both are completeness failures, and both are handled structurally here rather than by effort.

Missing records. Greenhouse, SmartRecruiters and Workday publish a record count alongside the board. Every fetch is compared against it. A board that comes up short is flagged, not quietly included.

Duplicates. A company migrating between platforms runs two live boards at once — the same role, two job IDs, two vendors. A vendor’s ID cannot see across vendors, so deduplication is on (company, title, location) instead.

Measured example: Shield AI’s requisition Manager, Propulsion (Fuel System) (R5088) is live as eight listings across two platforms. Billed per result, that one job charges eight times. Here it is one row.

What the change feed can and cannot prove

An opening needs one observation. A closure needs all of them.

Seeing a posting is enough to report that it opened. Reporting that it closed requires having seen the entire board — otherwise a missed page is indistinguishable from a company pulling twelve roles. So every closure carries a grade, and the grade is stated on the row.

7,192 provable 1,953 credible

Provable. The vendor published a count and the fetch matched it exactly.

Credible, not proven. One call returns the whole board, so nothing can truncate — but no count exists to check it against.

Across the first 14 days of diffs: 35,094 events — 19,206 opened, 15,888 closed. 4,005 of them are one requisition changing where it is advertised and 5,765 are coverage changes; both are graded low and excluded from every figure on this page. Of the remaining 9,145 closures, 79% are provable. Every competitor shipping a “new since last run” mode is estimating the closed half and not saying so.

Coverage by platform

Platform	Completeness	Basis
Greenhouse	Provable	Publishes `meta.total`
SmartRecruiters	Provable	Publishes `totalFound`
Workday	Provable	Publishes a `total` for every query, capped at 2,000
Ashby	Credible	Whole board in one call, no count published
Lever	Credible	Whole board in one call, no count published
Workable	Credible	Whole board in one call, no count published
Recruitee	Credible	Whole board in one call, no count published
Rippling	Credible	Whole board in one call, no count published
Personio	Credible	Whole board in one call; rate-limits aggressively

Today’s index holds boards on 7 of the nine platforms the code reads; Workday and Personio have none fetched in it today. Workday was measured on 2026-09-14 across 123 hospital nursing boards: every fetch matched the vendor’s own total for that exact query. Workday never reports a total above 2,000 but keeps serving pages past it, so a board that would exceed the cap is split by state and each state is verified on its own.

Hospital nursing coverage

A second index, read from the hospitals’ own Workday boards

A permanent nursing requisition opens on the hospital’s own careers site and closes there. Each health system’s nursing job family is fetched with that filter applied and checked against Workday’s total for the same filter, so the count is the hospital’s, not an estimate.

56Health systemsEvery one on its own Workday tenant, found by reading the public web index rather than a bought list.

123Nursing boards123 of 123 matched the vendor total. 0 mismatches.

21,198Open nursing requisitions16,476 after collapsing identical title-and-site requisitions.

11Days of historyDiffing daily since 2026-09-03.

Health system	Open nursing requisitions	Boards	Count-checked
Trinity Health	3,069	21	all 21
AdventHealth	1,917	7	all 7
Cleveland Clinic	1,105	4	all 4
Sentara Health	967	2	all 2
WVU Medicine	871	1	all 1
Beth Israel Lahey Health	751	1	all 1
Mercy (St. Louis)	746	1	all 1
Avera Health	641	4	all 4

Requisitions are the number a hospital counts, and the number quoted here. A hospital that posts the same title at the same site as separate requisitions collapses to one deduplicated role, which is why both figures are given. The largest systems were split by state to stay under Workday’s cap; Trinity Health’s nursing group alone runs as 20 state boards. Coverage was expanded on 2026-09-03 from 6,975 to 16,466 unique roles when systems found in the public web index were adopted; the 9,098 openings that produced are graded low as coverage, not hiring, and the expansion day is recorded in the history.

What that index shows that a job board does not

Health systems are staffing their own flexible labour and publishing it on their own careers sites. Read across the 123 nursing boards on 2026-09-14, these are requisitions the hospital is filling without an agency.

348Travel-titled nursing requisitionsPosted by the health system itself, across 12 systems.

1,669Per-diem, PRN or float-pool nursing requisitionsAcross 51 systems.

Health system	Travel-titled nursing requisitions
Trinity Health	221
Sentara Health	68
Saint Francis Health System	21
Intermountain Health	10
Baptist Health Jacksonville	8
HonorHealth	7

Matched on the requisition title, which is how the hospital labels the role; a travel assignment titled some other way is not counted here. These are open requisitions read on 2026-09-14, not change events. The commercial point is not the number but the absence: “this system has no open travel nursing need” is worth something only when the board was checked against the vendor’s own record count, which is why 123 of 123 matching is the figure that carries the claim.

Sample day — net movement, 2026-09-14

Company	Net roles	Direction

Relocations and coverage changes are excluded, so a company appears here only for roles it opened or closed on its own board that day.

Event schema

day the day the change was observed
event opened | closed
company identity used for deduplication, stable over time
company_display the employer’s own name for itself, taken from its board
name_review disputed, where the vendor name and the indexed name are
two different companies — see gap 3
employer_type agency | unverified — see gap 4 below
ats greenhouse | ashby | lever | smartrecruiters | recruitee |
workable | workday | rippling | personio
title as published by the employer
role_family engineering | data | sales | clinical | operations | …
seniority intern | junior | senior | staff | principal | director | vp
is_management people-management track, resolved separately from level
location as published
department as published
first_seen first day this role appeared in the index
last_seen last day it was observed live
same_role_elsewhere this requisition is live at another location — see gap 2
coverage board first seen | board went empty — see gap 7
confidence high | medium | low
basis the sentence explaining that confidence grade
url the employer’s own posting

What is not proven

01History is 14 days old. The daily diff began 2026-08-30. Hiring-velocity comparisons need roughly two weeks before they mean anything, and nothing here claims otherwise.
02One requisition advertised in several places looks like churn. Deduplication keys on (company, title, location), and boards rotate which locations they surface, so the same job reads as closing in Midland and opening in Allentown on one morning. Measured on 2026-09-01: 272 of 1,437 openings and 280 of 1,495 closures already had the same company and title live elsewhere. Those events are graded low, flagged same_role_elsewhere, and can be dropped outright — 4,005 of 35,094 events so far.
03Twelve company names are still disputed, and say so. Boards discovered by matching a public-filings name to a short URL token collected some confidently wrong ones: one labelled with a mining company was the fitness chain, another labelled a space company was the surgical-robotics firm. Every resolvable board was audited against the name published on the board itself, and 497 adopted the employer’s own name — 20,120 roles re-identified with zero fabricated events, because the change history was migrated in the same pass. The remaining twelve are ambiguous and are left labelled disputed for a human rather than guessed.
04Staffing agencies are labelled by review, not detection. One is confirmed: a UK healthcare agency running 2,658 locum ads, which would otherwise read as one enormous employer. The obvious automatic test — one title across many locations — was measured and failed: it ranks ordinary multi-office employers first and misses the agency entirely. Everything unreviewed is labelled unverified rather than assumed direct.
05Six of the nine platforms the code reads cannot be proven complete. They return the whole board in a single call, which leaves nothing to truncate, but publish no count to check against. Those rows are graded medium, never dressed up.
06Coverage is 1,195 boards plus 123 hospital nursing boards. That is a defined universe, not the whole market. It is listed per board rather than described as global.
07Adding coverage looks exactly like hiring, and until 2026-09-02 it was graded that way. Measured that morning: 288 of the day’s 1,980 openings were boards added to the index the same day, and 104 of its closures were one board that came back empty. Both patterns are now detected per company at export time and graded low with the reason on the row — 2,885 openings and 2,880 closures across the whole history. The index-wide circuit breaker that refuses a broken day cannot see one board arrive or vanish; this check does.
08The hospital index is 11 day(s) old. Its baseline was taken 2026-09-02 and it has diffed daily since 2026-09-03. It has produced 5,394 graded events so far. Until a second week of diffs exists it proves open requisitions, not turnover.

Sourcing and legal posture

Every endpoint is the ATS vendor’s own documented public API. No login, no session cookie, no terms accepted, no CAPTCHA, no IP-block circumvention. A board token either resolves or returns 404. Postings carry company data, not personal data of private individuals.

LinkedIn was measured and excluded rather than assumed. Its public guest endpoint returns 30 unique postings for a broad query and goes empty at offset 600, with no total anywhere in the response. A corpus cannot be proven complete against a source that will not say how big it is.

Getting the data

Two routes, both live today.

01Raw postings, self-serve. The readers behind this index are public on the Apify Store at $5.00 per 1,000 unique deduplicated results — one per ATS platform, plus a verified-boards reader that returns only the boards whose count this page can prove. No conversation needed; the input schema is the whole contract.
02The graded event feed. The opened and closed rows behind every number above, each carrying its grade and the reason for it — including the ones graded low because they are coverage changes or the same role advertised elsewhere. Delivered as daily CSV. Scope decides the price, so it starts with an email.

kaleb@revivacatch.com

Name the employers or the ATS platforms that matter to you and the answer comes back plainly, including when it is no: the coverage table above is the entire universe this index reads, and a board that is not in it will not be claimed as if it were.