Websites and PDFs into data.With a source and a history.

A sitemap lists page addresses. A PDF spreads a price list across pages. Each recruitment platform uses its own field names. Passing these through unchanged leaves the next step to handle every source differently.

We built focused tools with defined output fields, read status and enough state to continue the process. On the next run, the monitoring tools accept a saved snapshot and return the changes against it.

Suppose an employer’s website stops responding. A basic monitor treats the empty result as a closed job board. When the website recovers, it announces the same vacancies as new. The recipient gets noise instead of useful events.

Our tool retains the previous board state when a read fails. The first successful check without a role marks it as missing. A second successful observation triggers the closure event. That still means absence from the source: it does not claim someone has been hired.

A separate guard stops mass removals when most roles suddenly disappear from a sufficiently large board. The source needs checking before that change is accepted.

A nested sitemap may be temporarily unavailable, or a crawl may reach its limit before the last file. Neither proves that pages have been removed.

Sitemap Diff permits removal events only when both snapshots are complete and cover the same root sitemaps and filters. Errors, conflicting duplicates and crawl limits appear in the summary. The next process can check coverage before changing its catalogue.

When a table enters an accounting system or catalogue, the reader needs to know where each value came from. Records retain the document, page, table and row position. A content-based document identifier gives identical files the same identifier.

Values that look numeric remain strings. An item code keeps its leading zero, and local punctuation does not silently change an amount. CSV exports escape values that a spreadsheet might interpret as formulas.

A successful response is only part of an integration. The caller also needs to distinguish an unavailable source, an unsupported document and a reached limit. Tools return these states separately so the workflow can stop or retry the relevant step.

Network access has boundaries: public addresses, bounded responses and timeouts. The calling workflow owns its snapshots and decides when to check again and which version to use as its baseline.