# Changelog of ATS Jobs Scraper: Greenhouse, Lever, Ashby, Workday & More (`cleanyield/ats-jobs-scraper`) Actor

- **URL**: https://apify.com/cleanyield/ats-jobs-scraper/changelog.md
- **Full Actor documentation**: https://apify.com/cleanyield/ats-jobs-scraper.md

## Changelog

All notable changes to the output and the input of **ATS Jobs Scraper** (Greenhouse, Lever, Ashby, Workday, Oracle Recruiting Cloud, Eightfold, Phenom, Recruitee, Personio, Teamtailor).
Fields are only ever **added**, never renamed or removed. Every item carries `schemaVersion`.

### Unreleased — 2026-11-11 (3)

#### Generic reader for companies' own career sites (schema.org JobPosting)

- When a company does not use a hiring system we know, the Actor now also reads its own careers site if that site publishes each job as schema.org JobPosting (JSON-LD or microdata) in its sitemap: careers subdomains with any URL style (search-careers.gm.com/en/jobs/…, jobs.mercedes-benz.com/en/<title>-<id>, careers.factorialhr.com/job_posting/…) and the company's main website when its robots.txt declares a jobs sitemap (e.g. a WordPress `job_posting-sitemap.xml`). `ats` is `careersite`, as before; no field changes.
- Only the company's own domain, its subdomains or a careers site it links that carries its name; never a job board or aggregator. robots.txt and Crawl-delay apply to every request; at most 4 sites per company, 2 sitemaps (up to 4 child sitemaps each) and 6 job pages to recognise the site, and the usual 200 newest jobs per run.
- Salary only when the posting itself publishes it (`baseSalary`), never guessed from the text; postings whose `validThrough` has passed are skipped; remote from `jobLocationType: TELECOMMUTE`; one language per site (English when there is one), so the same job is not returned twice.
- Coverage on the same 286-company sample: 205 → 208 (71.7% → 72.7%), all 286 re-checked with a real run (gm.com, mercedes-benz.com, factorialhr.com and basf.com added; ferrovial.com is read from our network but not from Apify's, so it is not counted; airbus.com came back not found that day with the published version too). On all 324 test websites: 0 errors and no company changed board.
- Second, broader measurement (version 1.1.48, 2026-11-11): on a random sample of 1,000 companies drawn from Wikidata and Y Combinator (small, mid-size and large; US, Europe and Latin America), it reads 27.2% of the 874 reachable ones and 40.1% of the 594 where a careers page or hiring system was found: about 55% of US mid-size and large companies, 9.8% of small companies, 18% in Europe and 14.5% in Latin America. The README now shows both numbers.
- Newest first: when job URLs carry a job number (or the sitemap gives no usable date), jobs are ordered by the job number in their URL (never by years written in the title), and then by `datePosted`, so a run capped by `maxJobsPerCompany` (and monitor mode) always gets the newest postings.

### Unreleased — 2026-11-11 (2)

#### 9 more hiring systems and company career sites

- New values of `ats` (no field is renamed or removed; `schemaVersion` stays 3): `icims` (iCIMS sites that allow it in their robots.txt), `successfactors` (SAP SuccessFactors Career Site Builder), `jibe` (iCIMS career sites built on Jibe), `radancy` (Radancy / TalentBrew), `deel`, `bamboohr`, `breezy`, `gem`, `gupy`, and `careersite` (a company's own careers site that publishes each job as schema.org JobPosting in its sitemap, e.g. careers.lidl.com).
- From a company website, the Actor now also checks `careers.<domain>` and `jobs.<domain>` when the home page loads its menu with JavaScript (costco.com → careers.costco.com, deere.com → careers.deere.com), and a careers site on another domain that the website links, only if that board carries the company's name (mediamarkt.de → careers.mediamarktsaturn.com, cummins.com → cummins.jobs → its Oracle site).
- Boards found by name on BambooHR, Breezy, Deel or Eightfold follow the same rule as the others: returned only when the board shows the company's website or its job ads link that domain.
- SmartRecruiters is not read: the robots.txt of its API asks every robot except LinkedIn's to stay out.
- Career sites that ask for a pause between requests in their robots.txt (Crawl-delay, e.g. 5 s) return fewer jobs per run; the run message now says so ("the site asks for 5 s between requests… only 4 of 81 could be read in this run") instead of blaming `maxJobsPerCompany`. When the company also has a public Greenhouse, Lever or Ashby board confirmed as its own (toasttab.com → Greenhouse `toast`, whose jobs link careers.toasttab.com), that board is used instead.
- A robots.txt that answers with a server error (5xx) is treated as "everything disallowed" (RFC 9309).
- After a redirect to another host (company.com → careers.company.com), that host's robots.txt and Crawl-delay apply too. A careers site that asks for a pause is first previewed (2 jobs) and skipped when a confirmed Greenhouse/Lever/Ashby board exists (toasttab.com: ~250 s → ~135 s).
- Limits per run: Jibe up to the 1,000 newest jobs, SuccessFactors 500, iCIMS, Radancy and schema.org career sites 200 (one page per job).
- Not read on purpose: SmartRecruiters' API, old-style `careerN.successfactors.*` portals, the SuccessFactors RSS feed and iCIMS sites like enapply-danone.icims.com (their robots.txt asks robots to stay out); career sites whose job pages carry no job data (careers.3m.com, basf.jobs).
- Coverage on the same 286-company sample: 157 → 205 companies read (54.9% → 71.7%), 48 more and none lost, each checked with a real run from its website; the jobs of every new company were opened one by one on the live portal (title and company matched in all but one, whose Chinese title our check could not compare). On all 324 test websites: 167 → 219 found, 0 errors, no company changed board (so `jobId`s stay the same); macys.com came back blocked (HTTP 403) from our network that day with both the old and the new code.

#### Clearer results when a run returns few or no jobs

- The final status message (also seen by AI agents through MCP) now says how many open jobs were read, how many matched your filters and which filters to relax (e.g. "Read 827 open job(s) on 1 board(s), none matched your filters. To get more, relax: keywords (…)"), or that no company was found and what to paste instead.
- `postedAfter` in the future is rejected with a clear input error instead of returning 0 jobs.
- If your maximum cost per run is below the start fee plus one job, the run says so.
- An area written in the city field ("San Francisco Bay Area") now goes to `region`, with `city` empty.

### Unreleased — 2026-11-11

#### A job board with the same name as the company is no longer returned unless it is confirmed

- Fixes an error in the published version 1.1.33: with a company website, a board that only had the same name was returned and charged even when it belonged to another company (`basecamp.com` → the Personio board of "BASE CAMP", a Munich consultancy; `amazon.com` → a 3-job Personio board in London; `monday.com` → a coworking space in Valencia; `crisp.chat` → the Ashby board of crisp.nl). Now, from a website, a board the website does not link is returned only if it is confirmed: the board shows that website, its job links or job ads link that exact domain, or the company website links the careers site the board shows (`pinterest.com` → pinterestcareers.com). Otherwise the company comes back as **not found (free)** and the message names the unconfirmed board, so you can paste its URL or use the prefixed token (e.g. `personio:basecamp`) if it is the right one. `unverifiedBoards` in the COMPANIES report lists them.
- With a bare name (`stripe`, `Match Group`), the board must show a website or job links with that name, or (Greenhouse, Lever, Ashby) call itself by that name. `basecamp` no longer returns BASE CAMP and `wise` no longer returns "Wise Worksite Field Sales".
- Workday sites found from a website: only the website's **exact domain** counts (or the domain the website itself redirects to). Another domain with the same name is a contradiction (`trinity.com` is not Trinity University / trinity.edu; `liberty.com`, `yale.com`, `brown.com` likewise), and the company name alone is not enough (`mosaic.com` is not The Mosaic Company). Sites for laid-off staff, outplacement, temporary staff, postdocs and recruiters are skipped.
- When the Workday site belongs to one country's subsidiary (`hcahealthcare.com` → HCA Healthcare UK), `company` and the message say so: it may not include all of the company's jobs.
- Monitor mode: when a company's board changes (its `jobId` changes), the first run with the new board saves it as the baseline **without charging**, keeps `firstSeenAt` for the jobs you already had (matched by requisition ID, e.g. `R-31172` on both systems, or by title and city), and reports `boardChanged` in COMPANIES. Jobs from the old board that the new one only lists in a later run are also recognized, so jobs you already received are not sent (or charged) again.
- Websites whose careers link says "Carreiras" or "Trabalhe conosco" (Brazil, Portugal) are now followed (loft.com.br); a careers page that loads its board with a script is read (typeform.com); and a Lever board's "Home Page" link counts as its website (blablacar.com → jobs.blablacar.com).
- `company` shows the real spelling ("3M", "Chevron" instead of "3m", "chevron.com").
- Coverage is measured again on the same samples (see the launch notes); a few companies that were only matched by name are now reported as not found instead of risking another company's jobs.

### Unreleased — 2026-11-10

#### Finds the job board even when the company website does not link it

- From a company website, the Actor now also finds boards the site does not link (or loads with JavaScript), and only uses one after **confirming it belongs to that company**:
  - **Workday**: the company's own Workday site, found through the public list of career sites that Workday publishes for each company (e.g. `zoom.us` → Zoom's Workday site, `maersk.com`, `sanofi.com`, `3m.com`). It is used only if the site itself confirms it belongs to the company (see 2026-11-11 for the stricter rule).
  - **Greenhouse, Lever, Ashby and the others**: a board found by the company name now counts as the company's only when the board shows the company's website or its postings name the company's domain (e.g. `elastic.co`, `datadoghq.com` → Datadog's Greenhouse board, `travelperk.com` → Perk's Ashby board after its rebrand).
  - One more level from the careers page: links like "See open positions" are followed (brevo.com → its Lever board).
- In the COMPANIES report, `resolvedBy` can now be `websiteVerified` (found from the website and confirmed) besides `careersPage` (linked from the website) and `nameProbe` (found by name, not confirmed).
- Fix: a company whose name does not exist on Personio no longer fails with an error when Personio's website answers HTTP 429 (3m.com, gm.com and ge.com used to fail).
- Measured on the same 286-company sample: companies read went from 47% to 55% (134 → 158). On all 324 test websites: 157 → 187 found, none lost.
- Nothing is renamed. Three websites now get a better board than before, so their `jobId` changes: `maersk.com` (a test board named "maersk" on Teamtailor → Maersk's Workday site, 1,207 jobs), `kavak.com` (Lever, 5 → Workday, 407) and `siemens-healthineers.com` (Workday `onehealthineers`, 498 → its own careers portal, 898). Every other company keeps the same `jobId`.

#### New hiring systems: Oracle Recruiting Cloud, Eightfold and Phenom

- `ats` can now also be `oracle` (Oracle Recruiting Cloud / Oracle HCM Candidate Experience), `eightfold` and `phenom`. Same fields as every other ATS; nothing renamed, `schemaVersion` stays 3.
- Type the company website (`emerson.com`, `bostonscientific.com`, `lowes.com`) or the site URL. A bare name is not enough for these three.
- Salary comes from the posting text when the company states it (`salarySource: "description"`).
- Measured on the same 286-company sample: companies read went from 39% to 47%.

#### Fixes

- Canadian province codes are no longer read as countries ("Sudbury, ON, Canada" → `CA`), and a country written after a state code wins ("…, PA, Brasil" → `BR`).
- From a website: when the home page has no Careers link, `/careers` and `/jobs` are tried; when the bare domain does not answer, `www.` is tried.
- Workday jobs keep the company website/domain found on the careers page after reading each job page.
- Phenom portals where the job description was missing (CVS Health, GSK, Humana) now return it.
- A website that links to other companies' job boards (job sites and aggregators) no longer returns another company's jobs: the company is reported as not found (free) and the message lists the boards that were skipped.
- A website that limits or blocks requests is retried once after a pause and then reported as "could not open the website" (free, with `unreachable` in the COMPANIES report) instead of "not found". Paste the job board URL in that case.
- The other-company check only applies to known job sites and to pages that link job boards of two or more companies. A company website that links a single board is used as before, even under a parent or group name (slack.com → Slack's Workday site, merck.com → MSD's global Workday site). If another page of the same website links a board with the company's name and more jobs, that one is used (hims.com → Hims & Hers, 149 jobs).
- A website that keeps answering HTTP 429 is no longer retried on `www.` and has a 90-second limit, so one slow website does not hold the run for minutes.
- Job sites on a company's own domain (`careers.idealista.com`, `careers.betterfly.com`, `jobs.acme.com`) now use the company name as `companyToken` and in `jobId` (`teamtailor:idealista:…`). For idealista and betterfly the `jobId` is the same as in the published version; for about 12 Teamtailor sites that the published version stored as `careers` (e.g. PayFit), only the prefix of `jobId` changes (`teamtailor:careers:…` → `teamtailor:<company>:…`). Two such companies in the same run are no longer skipped as duplicates. Monitors that saved the old `careers` id (e.g. PayFit) carry their state over, so existing jobs are not reported as new.

### 1.3.0 — 2026-11-09 (pending publication)

#### New option: ATS-specific fields (`includeAtsFields`, off by default, no extra cost)

- Greenhouse: `requisitionId`, `internalJobId`, `departments` (with parent), `departmentPath`, `offices` (structured location), `customFields`, every pay range in `payRanges`, `applicationDeadline`, `jobLanguage`, `gdprConsentRequired`.
- Lever: `commitment`, `level`, `sections` (the posting's lists as bullets), `openingText`, `additionalText`, `salaryInterval`, `salaryDescriptionText`, `leverRegion`.
- Ashby: `compensationTiers` (salary, equity, commission, bonus per tier), `compensationSummary`, `offersEquity` / `offersBonus` / `offersCommission`, equity value range, `secondaryLocations`, raw remote flag and workplace type, postal address.
- Workday: `requisitionId`, `jobPostingId`, `timeType`, `remoteTypeRaw`, exact `startDate`, `postedOnText`, `additionalLocations`, requisition location and country, `hiringOrganization`, tenant/site/host; the company name is taken from the hiring organization when it matches the site (e.g. "PayPal", not "Paypal").
- With the option off the output is exactly as before.

#### Fixes

- Greenhouse salaries in currencies without decimals (JPY, KRW…) were divided by 100 (Duolingo Tokyo showed ¥171,780 instead of ¥17,178,000).
- Greenhouse: an hourly pay range whose long text says "base pay" was read as annual and dropped (Robinhood USD 21–25/hour); now kept as hourly.
- Lever: company names with HTML entities in the board title ("Dun & Bradstreet" → "Dun & Bradstreet"); double-encoded descriptions decoded.
- Placeholder locations ("N/A") are now empty instead of a fake value, and impossible publication dates (before 2010) are left empty.
- Spending limit: with "Add company details" on, the company details are now charged together with the first job delivered for that company (job + Pro + details must fit in your maximum cost together). Before, with a very small limit, the details could be charged without any job being saved.
- Workday: filtering by employment type or workplace type returned 0 jobs (those values only exist on the job page); now the job pages are read in small batches until the limit is reached. The default "minimum salary 0" no longer forces reading every job page when another filter is set. A Workday site that does not exist (HTTP 422) is now reported as not found (free) instead of failing.

### 1.2.1 — 2026-11-07

- Fixed: with the default form values, the per-company job limit was ignored for Workday boards, so very large boards (thousands of jobs) could run for minutes. Now NVIDIA + Salesforce with 15 jobs each finish in ~7 s.
- Input: `company`, `companyName`, `companyNames`, `urls` and `startUrls` are accepted as `companies` (before, a run with `company` failed).

### Unreleased — 2026-11-06

#### Faster and steadier runs (no output change)

- A request that gets no answer in 2.5 s is sent again in parallel and the first answer wins (a stuck connection no longer adds 10–30 s to a run); per-request timeout 30 s.
- Jobs are written to the dataset several at a time (still exact `maxResults` and spending limit).

### 1.2.0 — 2026-11-06 (`schemaVersion: 3`)

#### New job boards (same output fields)

- **Workday** (`https://<tenant>.wdN.myworkdayjobs.com/<site>`, also found from a careers page link): the same public, login-free requests its job site makes; exact publication date from each job page.
- **Recruitee** (`<company>.recruitee.com` or its own careers domain), **Personio** (`<company>.jobs.personio.de` / `.com`) and **Teamtailor** (`<company>.teamtailor.com` or its own domain), also as prefixed tokens (`recruitee:x`, `personio:x`, `teamtailor:x`).
- `ats` can now also be `workday`, `recruitee`, `personio` or `teamtailor`.

#### Finding the board

- A company **website** (`figma.com`, `bunq.com`) is enough: the Actor follows its Careers/Jobs link to the job board, and recognizes Teamtailor and Recruitee sites on the company's own domain.
- Bare names: Greenhouse and Ashby are tried first, then Lever, then Recruitee, Personio and Teamtailor (faster: Lever is only asked when needed).

#### New output fields

- `salaryAnnualMinUsd`, `salaryAnnualMaxUsd` — annual salary in USD (exchange rate of the day of the run; one request per run, only when a salary is not in USD).
- `closedAt` — monitor mode, closed jobs: when the run found the job gone. `firstSeenAt` — monitor mode: first time your monitor saw the job.

#### Better data in existing fields

- `employmentType` and `workplaceType` are also read from explicit phrases in the job text when the ATS has no field ("the base salary range for this full-time position", "in the office 2 days a week"). Still `null` when the post does not say it.
- `companyWebsite` / `companyDomain`: when the board does not link the website, the domain the company's own posts link to (only when it matches the company name).
- `experienceYearsMin`: also "plus 4+ years…" and "(5+ years across…)".

#### Other

- A job posted on two boards of the same company in one run is returned and charged once (`crossBoardDuplicatesSkipped` in SUMMARY).
- Not-found message lists every ATS that was tried.

#### New input

- `postedAfter` — keep only jobs published on or after an exact date or time (`2026-10-01`, `2026-10-01T09:00:00Z`) or a relative time (`48 hours`, `7 days`). Also read from `datePostedAfter`, `posted_after` and `postedSince`. No output change.

### 1.1.0 — 2026-11-04 (`schemaVersion: 2`)

#### New output fields (all jobs, no extra cost)

- `seniority` — level read from the title: `intern`, `entry`, `mid`, `senior`, `lead`, `director`, `executive` (null when the title does not say; we never guess).
- `experienceYearsMin` — minimum years of experience the job text asks for ("5+ years of experience" → 5).
- `visaSponsorship` — `true` / `false` when the job text says so explicitly, `null` when it does not (or says both).
- `descriptionLanguage` — language of the job text (ISO 639-1).
- `allCities`, `allCountryCodes` — cities and ISO countries of **all** the job's locations.
- `companyWebsite`, `companyDomain`, `companyLogoUrl` — what the company shows on its own job board.

#### New filters

- `seniorities`, `maxExperienceYears`, `onlyWithSalary`, `onlyVisaSponsorship`, `excludeLocations`.

#### Pro mode (`proMode`, charged per job as `pro-result`)

- `companyHiringScore` (0–100), `companyHiringAccelerating`, `companyOpenJobs`, `companyJobsPostedLast30Days`,
  `companyTrend` (open jobs then vs now since your first run, stored in your account), `whyNow`, `techStack`, `insight` (AI sentence).

#### Extras

- `enrichCompanies` (charged per company as `company-enriched`): `companyDetails` with email, phone, LinkedIn/X, country and website technologies.
- Saved lists: `savedListName` + `saveList` remember your company list between runs.
- Drop-in: input field names of popular career-site job APIs are accepted (`organizationSearch`, `titleSearch`, `locationSearch`, `limit`, `timeRange`, `aiWorkArrangementFilter`…) and `outputFormat: "leader-compatible"` returns their field names.
- Free dashboard page in every run (`DASHBOARD` in the key-value store) and a `SUMMARY` record.
- A cost estimate is logged before the run starts.

#### Fixes and hardening

- Source items without an ID or a title are dropped (never stored, never charged).
- A text list of companies with line breaks is split by line only ("Stripe, Inc." stays one company).
- The company profile page is optional: if it fails, the company's jobs are still returned.
- Monitor state names created with 1.0 do not change (new filters only enter the name when used).

### 1.0.0 — 2026-11-04 (`schemaVersion: 1`)

- Live jobs from the public job board APIs of Greenhouse, Lever (incl. EU) and Ashby, one normalized schema.
- Company input as name, board URL, careers page or prefixed token; clear `notFound` report (free) in `COMPANIES`.
- Structured salary (ATS first, explicit text range second, `salarySource`), annualized values.
- City / region / ISO country, `workplaceType`, normalized `employmentType`, ISO dates, full description (text/HTML).
- Filters: keywords (title or description), exclude keywords, locations, workplace, department, employment type, posted within N days, minimum salary.
- Monitor mode: only new / changed (with `changedFields`) / closed jobs, Slack and webhook alerts, `HIRING_SIGNALS` per company.
- Pay per valid result, spending limit respected, sync mode, n8n / Make / Clay templates.
