Skip to content
Raise Lens
HOW IT WORKS

How we source data

Every record in Raise Lens traces back to a live page on a source we monitor: a government funding portal, a public procurement system, a multilateral development bank, a foundation's own site, or a corporate disclosure. Nothing is generated, nothing is crowdsourced, and nothing enters the feed without a source URL attached.

This page describes the pipeline between those pages and what you see.

Where records come from

We monitor 120+ sources across five categories: government grant portals, public procurement systems (national and multilateral), foundation and philanthropic programme pages, accelerator and innovation programme listings, and corporate and institutional disclosure pages that carry procurement-relevant signals. The full breakdown by category and geography is on the source list; which markets they cover is on the coverage map.

A source is only added after it is verified live and its structure is understood. We do not bulk-import link directories, and we would rather cover a source properly than claim it nominally.

What happens to a raw record

Six steps, in order:

  1. Normalisation. Markup cleanup, dates to ISO, currency values converted and standardised, and a content hash stored so we can tell when a record actually changes.
  2. Validation. News posts, job listings, events and press releases are dropped. Mixed feeds are the main source of aggregator noise; this step is why our feed is not full of it.
  3. Sustainability gate. Procurement records are prefiltered by procurement classification codes and multilingual keywords before deeper processing. This is what removes cleaning contracts and office furniture at volume.
  4. Translation. Non-English titles are translated at ingest across 30+ source languages, so the feed reads in English regardless of origin.
  5. Classification. Each record is tagged with sectors from a controlled vocabulary, geography, funding type, stage fit, organisation type, key requirements, a plain-English summary and a 0 to 100 relevance score. Full detail, including error handling and the human review queue, at Accuracy and classification methodology.
  6. Deduplication. The same notice republished across portals is collapsed into one record with an "also published on" footer.

Change detection, not just collection

Sources are re-checked on a schedule: daily for opportunity sources, weekly for the signal crawl. The content hash from step 1 means unchanged pages cost nothing and changed pages are reprocessed. Actual timestamps per pipeline are public at Last refreshed.

What we do not do

We do not scrape sources we have not verified. We do not pass through mixed news feeds unfiltered. We do not backfill gaps with generated content. Where a source category is thin, the honest state is visible on the coverage map rather than papered over.

Related