How a News URL Becomes a Feed Record

The feed does not build itself. An operator or authenticated service submits a URL. The ingestion path extracts a bounded record, validates it, and writes it to the configured feed store.

The local development command

GREENCIO_RUNTIME_PROFILE=development npm run url-news "https://example.com/article"

The explicit development profile permits the labeled local fallback when Google Cloud Storage (GCS) is unavailable. Preview, staging, and production do not use that fallback.

Extraction

1. Validate the URL before the outbound request.
2. Fetch a bounded HTML response.
3. Read standard metadata and a limited article-text excerpt.
4. Use a generated summary when the model is configured; otherwise use source metadata or extracted sentences.
5. Derive relevance, impact, sentiment, tags, and named entities from the resulting text.

The extracted record retains the source URL, source name, publication date when available, and discovery time. A generated summary does not replace the source link.

Validation and identity

The write path validates record shape and bounded values before persistence. It merges records by stable identity so a retry does not need to create a duplicate article.

Storage behavior

Configured environments write the canonical feed tofeed/latest.json in GCS. The write uses the current object generation to detect a concurrent replacement. If a strict environment cannot read or write GCS, the operation fails. It does not silently write a local production feed.

What still needs review

Extraction can misread a page, and a generated summary can omit context. Before publication, review the title, source date, quantities, attribution, and any statement about future plans. A relevance score measures keyword fit. It is not source quality or factual confidence.

Browse the public intelligence feed or inspect the repository.