# Changelog of SEEK, JobStreet & JobsDB Job Scraper — 8 Markets (`bluedata/seek-jobstreet-jobsdb-scraper`) Actor - **URL**: https://apify.com/bluedata/seek-jobstreet-jobsdb-scraper/changelog.md - **Full Actor documentation**: https://apify.com/bluedata/seek-jobstreet-jobsdb-scraper.md ## Changelog This changelog describes customer-visible behavior only. Development-only materials are not included. ### Unreleased — Launch readiness - New `coverageMode` input. `best_effort` (default) keeps the existing bounded scan. `fill_target` widens the page and detail-candidate windows and keeps scanning and enriching until `maxResults` matched jobs are emitted, the source is exhausted, or a hard run limit is reached. Runs now report `coverageMode`, `requestedResults`, `targetReached`, `shortfall`, `candidatesScanned`, `candidatesRejected`, a run-level `stopReason`, and a short `resultNotice`; `fill_target` runs that cannot meet the target add a `TARGET_NOT_REACHED` warning. Pricing events are unchanged; more scanning does not create extra charged events. - Actor title and description are updated for the 8-market launch, including advertised export formats, optional company data, and RUN_SUMMARY coverage. - Input and output documentation now states that parsed queries are forwarded exactly without translation or synonym substitution. Location aliases retain original and effective values; no geographic expansion is claimed. - Removed the two automatic MY/ID query substitutions. Original parsed query text is used for both generated searches and URL-derived targets. - Pricing documentation distinguishes dataset-item events, start events, and optional enrichment work. No new billing event is introduced. ### Unreleased — Verified multilingual relevance and multi-query accounting - Bounded runs fetch scan-sized source pages (minimum 100 listings) instead of shrinking the page to `maxResults`. Filters run after fetching, so the old behavior made filtered markets stop at the candidate window with fewer results than requested. HK searches now get a wider candidate window within the same page-count safety cap. - `resultLimitReached` (final output reached the requested cap) and `coverageLimitReached` (source fetch window capped) are reported independently on the run summary and on each search. The legacy `limitReached` field remains available as a broader cap signal; use the two explicit fields to distinguish output limits from incomplete source coverage. - Incremental runs report `baselineComplete` and `expiredRecordsReliable`. A coverage-limited baseline suppresses `emitExpired` markers entirely and adds the `BASELINE_INCOMPLETE` warning; missing jobs beyond the fetch window are no longer mistaken for expired postings. Searches always expose `coverageComplete` as an explicit boolean, never null. - Detail enrichment candidates scale with the output cap (5x, minimum 25) so detail-dependent filters can refill the requested results instead of stopping after a few rejected candidates. - `searchPlan` and `searches` expose `originalLocation` (caller input) and `effectiveLocation` (source label) separately, e.g. `香港` vs `Hong Kong`. Original query text remains unchanged; there is no translation. - Named incremental baselines use a cloud KVS created on first use. State failures are surfaced and no longer silently classify every row as new. - State-key labels are hashed into Apify-safe record keys, including when the caller's label contains punctuation that Apify KVS does not allow. - Search status is separated from detail enrichment status. A completed search stays `sourceStatus: ok` when detail requests are partial; `detailsEnriched` counts successful detail responses per candidate listing. - Coverage-limited searches expose `coverageLimit`, `coverageLimitedSearches`, and `sourceStatus: coverage_limited`; they are not presented as Actor failures. - Thai queries in TH and Chinese queries in HK use exact query text without translation or added synonyms. Complete no-match coverage returns `no_results`; capped searches return `coverage_limited`. - Zero-evidence source suggestions are filtered before output; emitted records always have target-specific literal evidence and `relevanceScore: 1`. - Non-Latin queries require literal evidence. Complete evidence-free targets expose `queryLanguage`, `languageSupport: unsupported`, and `rejectionReason: No verified query-term match`. For SG, Arabic/Korean targets with no first-page literal match terminate as unsupported instead of scanning generic fallback pages; literal matches remain eligible. - `processedSearches`, `completedSearches`, and `fullyCoveredSearches` distinguish executed targets, terminal outcomes, and exhaustive source coverage. Coverage caps count as completed terminal outcomes while remaining not fully covered. - Source request/time and challenge failures remain partial; detail enrichment has its own `detailStatus`. - Added per-target execution plans and attempted, processed, fully covered, skipped, discarded and resumed accounting. Unattempted work no longer counts as completed; speculative requests remain visible even when discarded. - Multi-query and multi-URL runs divide the source candidate allowance across targets so a run-wide output cap does not silently prevent later targets from executing. The final emitted dataset remains capped. - `searchPlan` now exposes each target URL and input source, making hidden default-query and duplicate URL cases observable. - Arabic-script queries require literal evidence; unrelated source suggestions are filtered and an evidence-free target is reported as unsupported. - The SG/HK fast path is limited to Arabic, Hangul and Chinese first pages with no literal evidence; literal matches remain eligible and no request budget is overridden. Thai queries in TH remain untranslated. ### Unreleased — Input and result-quality hardening - Fixed TH Bangkok alias matching across case, whitespace and compound source labels; added HK Hong Kong/香港 label aliases without widening district filters. - Corrected mixed-type input fields to use Apify's JSON editor. Query accepts text, native arrays or `{terms: [...]}`; start URLs accept strings or objects. - Added opt-in `queryMatchMode` (`all_terms`/`phrase`) and literal match evidence. Field-only keyword flags validate the main query when no explicit include or exclude terms are supplied. Legacy keyword scopes remain available. - Added two explicit MY/ID query phrase aliases, preserving original/normalized queries and checking literal evidence instead of accepting generic suggestions. This is not general translation or language detection. - Source-ranked results without full literal evidence now carry explicit relevance warnings; source success is not semantic relevance certification. - Added suspicious salary heuristics with null numeric bounds, readable source text and warning flags; foreign currency alone does not invalidate salary. - Empty completed/filtered searches no longer fail record contract checking. Partial runs expose warning flags, filter counters and safe stage/code logs. - Query-validation mode changes reset the incremental baseline. This does not establish concurrent-monitor safety or any live source-coverage guarantee. ### Unreleased — Optional data groups - Detail concurrency is managed automatically at a default of 8 parallel detail requests; it is not a user-facing tuning input. The request budget still governs total detail work. - Detail enrichment now runs as a continuous worker pool instead of batches, which shortens enriched runs at the same request budget. - While one search's details are enriched, the next query's search page is fetched concurrently; if the run stops early, the scouted search's records are rolled back so results match a sequential run. - Avoid unnecessary extra detail enrichment when no keyword filter is configured. - Accept both single-object and list-shaped location hierarchies. - Clarified that result limits do not cap network requests or spending. - Added first-use `preset` choices: `basic`, `company`, and `full`; each keeps contacts and login-only applicant insights off by default. - Clarified flattened CSV/Excel columns, persistent monitoring configuration, filter semantics, source URL inputs, and the 10,000-result output ceiling. - Added individually selectable public listing and company-data groups; all are disabled by default and report availability when requested. - Public contact fields require explicit opt-in and a separate notification allowlist. Data requiring a logged-in session remains unavailable. - Added summary-level search suggestions and clearer partial-coverage status. - Optional groups report `available`, `not_provided`, `restricted`, `failed`, or `skipped_budget` status instead of implying that every selected field exists. - Nested optional data remains lossless in JSON; CSV/Excel exports should be treated as serialized cell values for nested arrays and objects. - Added opt-in stateless `exportExcel` delivery for `JOBS.xlsx` (1–1000 `maxResults`) with readable field/part rows and an exact Canonical JSON sheet; native Dataset exports remain unchanged. ### 0.1.14 — Result delivery and validation improvements - Improved output handling for larger result sets. - Strengthened input validation and partial-run status reporting. - Improved handling of incomplete or malformed source records. - Improved salary, posting-date, source labeling, and text normalization. - Preserved optional detail enrichment, incremental flags, notifications, and compact output. ### 0.1.11 — Run safety and filter correctness - Added clearer stop reasons and coverage status for incomplete searches. - Improved result-limit behavior after final filtering. - Improved checkpoint and retry behavior for interrupted runs. - Added safer monitoring, notification, and incremental-state handling. ### 0.1.10 — Record handling - A malformed listing no longer causes an entire result page to be discarded. ### 0.1.9 — Larger result limits - Increased the supported maximum result limit to 10,000. - Added run notes when numeric inputs are clamped. ### 0.1.8 — Filtering and notifications - Improved notification failure reporting. - Improved salary text handling for local currencies. - Improved description keyword filtering when detail fields are enabled. - Improved numeric filter validation. ### 0.1.7 — Persistent incremental state - Added named state-store support for cross-run monitoring. ### 0.1.6 — Filters and notifications - Added salary, date, classification, keyword, custom-field, and sort filters. - Added incremental monitoring flags and optional notifications. - Added compact and empty-field output controls. ### 0.1.5 — Enrichment and monitoring - Added optional detail fields, classification filters, and incremental change flags. - Improved retry and source-availability handling. ### 0.1.4 — Multi-market output - Added normalized output across the supported APAC markets. - Added market-aware currency, locale, and source metadata. - Improved placeholder salary detection and text cleanup. ### 0.1.0 — Initial release - Added structured public job-search results with salary, posting-date, category, location, and source-link fields.