A tender must contain at least one of these words in its title or body. One word/phrase per line. Case-insensitive substring match.
6-8 digit EU procurement codes. Lookup ↗
ISO 2-letter country codes (e.g. ZA, KE, AE). Used by World Bank and the territory filter. This is the SCRAPED list — the map scores a much larger universe.
Phrases sent as search queries to the World Bank API. More terms = wider net but more API calls.
UNGM Title field substring searches. Each adds ~3-8s to the run.
Regex matching navigation links / nav text to discard. Tested against title.
Regex matching category-index pages to discard (e.g. "Energy Tenders").
A tender hitting any of these is dropped before classification (airport infrastructure, ATC, ground handling, ARFF, building fire suppression, anti-drone, etc.). One per line.
If any of these appear in the same record as a negative keyword, the record is kept (FF72/CL-415/aerial firefighting etc.). One per line.
TED notices carrying any of these CPV codes are excluded at query time. e.g. 45235300 = runway construction, 34962000 = ATC equipment, 63733000 = aircraft refuelling services.
Newsroom + trade-press RSS feeds, polled on every scrape. Format per line: tier|name|url — e.g. 1|De Havilland Canada|https://dehavilland.com/feed/. Tier 1 = OEM/authoritative, Tier 2 = trade press.
Must contain the [TEXT] placeholder. Drives title/territory/urgency/business-line assignment.
Keywords help Claude understand what each business line covers. Edit each line's keywords below (comma-separated).