Lever Jobs Monitor PL/DE

Monitor a supplied list of public employer boards hosted on Lever. Get current jobs whose published country field is PL or DE, source salary ranges when present, and evidence of changes or repeated absence. This Actor does not discover employers or cover the whole Polish/German job market.

One delivered current job observation is one billable event, including jobs labeled unchanged. See the Pricing tab for the active price; at $0.003 per observation, eight delivered observations cost $0.024. There is no startup fee, charge for errors, or charge for zero matching observations. Changes, coverage summaries and snapshots are free key-value outputs. The platform's active price governs a run; this version refuses prices above $0.003 or positive automatic fees.

Start with your employer boards

{
  "boards": [
    {"site": "aircall", "region": "global"},
    {"site": "palantir", "region": "global"},
    {"site": "spotify", "region": "global"}
  ],
  "countries": ["PL", "DE"],
  "maxPagesPerBoard": 10,
  "maxResults": 100,
  "minimumMissingHours": 6
}

site is the identifier in the employer's Lever board URL. Use global for jobs.lever.co, or eu for jobs.eu.lever.co. API endpoints are fixed; arbitrary URLs, login and candidate/application access are unavailable. A board returning 404 is an error, not an empty result.

Changes and snapshots do not duplicate the full current job payload. Snapshots carry hashes for comparison; their checksum detects corruption but does not authenticate a customer-supplied baseline.

For later comparisons, copy the complete NEXT_SNAPSHOT object from the previous run into the optional previousSnapshot input. Keep the same boards, countries and minimum missing interval. Omit it to start a fresh baseline. Scheduling remains your choice in Apify; this Actor creates no schedules.

Run through the API

Use your own token in the APIFY_TOKEN environment variable. This example starts a run; use the returned run ID to inspect status and output when it finishes.

curl --request POST \
  'https://api.apify.com/v2/acts/danielmichalek~lever-jobs-monitor-pl-de/runs?maxTotalChargeUsd=0.03&timeout=240&memory=512' \
  --header "Authorization: Bearer $APIFY_TOKEN" \
  --header 'Content-Type: application/json' \
  --data '{"boards":[{"site":"aircall","region":"global"},{"site":"palantir","region":"global"},{"site":"spotify","region":"global"}],"countries":["PL","DE"],"maxResults":100}'

What you receive

Output	Contents	Job event charge
Default dataset	Current matching job observations; changeType: current and changeSinceSnapshot	One per delivered row, including unchanged
CHANGES	Compact identity/time/changed-field evidence: new, changed, entered_scope, left_scope, reappeared, disappeared	None
OUTPUT	Per-board coverage, error/cap status, selected counts, structured salary coverage and billing summary	None
NEXT_SNAPSHOT	Compact identity, field hashes and state baseline for a later run, when safely available	None

Each current row has a stable recordKey based on region, board and posting ID, the title, raw location/commitment/workplace fields, source link when valid, observation times and salary status. Identity deduplication does not merge similar titles across employers. firstObservedAt is the first observation in the supplied baseline history. sourceCreatedAt preserves Lever's numeric createdAt when supplied. Neither is claimed as publication date; publicationDate stays null.

Salary comes only from the optional structured salaryRange. It is not extracted from descriptions, inferred, converted or assigned a tax basis. salaryStatus distinguishes published, not_published and invalid_published_value; unknown salary is null. Commitment and location remain source text. Country selection uses the explicit source country, not a city name or inferred remote eligibility. Additional location strings can name other countries without changing the primary published country filter.

What absence means

A missing job first becomes a pending absence. Only a second complete read of that board, at least minimumMissingHours later (minimum and default six), emits disappeared. A complete read requires a terminating page and valid unique posting IDs. Failed, capped, malformed or duplicate/racy pagination cannot advance that board's baseline or produce an absence event. Successful boards can advance while a failed board's previous state is preserved.

This is evidence of repeated absence from the published API, not proof that a vacancy was filled or definitively closed. Offset pagination is not an atomic source snapshot: undetected source changes between pages remain possible. Two spaced reads reduce false disappearance claims; they do not eliminate them. A published country change is left_scope, not closure. Reappearing after confirmed disappearance is labeled reappeared.

If current observations, changes, a spending limit or snapshot size truncate the intended result, NEXT_SNAPSHOT is withheld. Inspect OUTPUT.report before continuing a baseline. Retained historical disappeared entries count toward the snapshot limit; start a fresh baseline when needed and retain old snapshots yourself if useful.

Bounds and failures

• 1–5 explicit boards; only PL/DE filters; at most 100 postings/page and 10 pages/board.
• 500 delivered observations or change entries; 5,000 entries and 5 MiB per snapshot; 8 MiB per source page and 32 MiB total source bodies per run. Unused HTML/descriptions are discarded after each page.
• Sequential public GETs, redirects refused, 20-second request timeout, a 180-second source budget; at most three attempts for network errors, HTTP 429 or 5xx. Retry-After is honored within the source deadline; waits over 30 seconds stop the board instead of retrying early. Set a platform run timeout of at least 240 seconds to allow output finalization.
• A full 100-row final page needs another page to establish completeness. Lowering the page cap can intentionally yield incomplete coverage.
• INPUT rejects invalid bounds, unknown fields, changed snapshot scope or checksum corruption. HTTP_*, PAGE_LIMIT, TIME_LIMIT, SOURCE_* and DUPLICATE_ID appear in board coverage. No complete board means a failed run with no job events.
• BILLING_CONFIGURATION rejects unsupported pricing. The SDK's unpriced shape is accepted only when the SDK explicitly identifies a local runtime; missing cloud pricing fails closed. A zero effective spending limit makes zero source requests and emits zero job events. Partial budgets deliver only the affordable prefix and withhold the next snapshot.
• RECONCILIATION_REQUIRED means an interrupted or inconsistent delivery cannot safely be replayed. Writes and charges are not transactional. A durable pending checkpoint blocks automatic replay, including after a lost charge acknowledgement. Inspect the existing dataset and platform charge records before deciding whether to start a separate run; a separate run can bill a fresh observation again.

Access and evidence

Lever's official Postings API README explicitly states that published jobs are publicly viewable and may be scraped by third parties. Only those public GET endpoints are used. This is not a blanket license to redistribute employer content. Respect employer content rights and applicable downstream obligations. No full job descriptions, candidate records, contact enrichment, paid API, browser or proxy are used.

On 2026-10-21, a live read through this source adapter covered Aircall (77 postings), Palantir (306) and Spotify (79). Of 462 source postings, eight matched explicit PL/DE countries (one PL, seven DE), with zero structured salary ranges in those eight. Spotify legitimately matched zero. These are three source checks, not evidence of comprehensive market coverage or paying demand.

Local verification

Node 22 or newer. npm test uses built-in Node tests and saved source-field fixtures. npm run verify:live -- /tmp/lever-check.json performs bounded public GETs through the source adapter and writes a local report; it does not initialize or run an Apify Actor. Normal runtime is npm start after npm ci --omit=dev --ignore-scripts.

The SDK is pinned to apify@3.7.2. Runtime guards check actual dataset contents, the cumulative named-event count delta and the response to each write/charge call. Interrupted delivery stops rather than automatically replaying ambiguous writes or charges.