News Monitoring

Relevance-scored news tracking across multiple sources.

Implemented in farsight/modules/news.py, class NewsMonitor. Enabled with --news (or --all/--modules news). Runs last in the pipeline. Its results are also reused by Organization Discovery’s news-based acquisition detection.

Primary source: GNews

If the gnews library is installed (it’s a core dependency, so this is the default path), Farsight uses it to search Google News for mentions of the target, run in a thread executor since the library itself is synchronous.

The fallback is placeholder data — know this before you rely on it

If gnews isn’t available, _monitor_alternative generates synthetic placeholder articles — this is explicit in the source, not a hidden bug: the code comments read “Placeholder for demonstration” and “In a real implementation, this would use a real news API.” It fabricates up to 5 articles from a fixed set of publisher names and title templates (e.g. "{target} Announces New Security Measures", "{target} Partners with Leading Cybersecurity Firm") with random dates within the lookback window. If you ever see News Monitoring results that look suspiciously generic or formulaic, check whether gnews is actually installed — pip install gnews (already a core dependency of farsight-recon, so this should only come up in an unusual environment where it failed to install) restores real results. The web UI surfaces this directly: its gnews_available health-check field drives a visible warning banner when the fallback is active.

Relevance scoring

extract_relevant_articles filters and ranks articles with a simple additive score, independent of whatever ranking the source (GNews) already applied:

score += 5  if the target string appears in the article title
score += 3  if the target string appears in the snippet
score += 2  per substantial word (>3 characters) of the target that
            appears in the title
score += 1  per substantial word that appears in the snippet

An article is kept only if its score reaches at least 3 — a snippet-only mention of the exact target string clears that bar on its own, but a single partial-word match in the snippet does not. Kept articles are sorted by this relevance score, and separately, the final result list is sorted newest-first using a multi-format date parser (RFC 2822 with and without timezone, then ISO date, then ISO datetime) before being capped at news_results_limit (default 10).

Output shape

{
  "articles": [
    { "title": "...", "url": "...", "published": "2026-06-01", "publisher": "...", "snippet": "..." }
  ],
  "total_articles": 9
}