How a brief is made

Every number and label you see on IndiaBatmi comes from a fixed pipeline. This page is the honest version, including the parts that are automated.

1 · The source registry

Coverage flows only from publishers we have deliberately added to a registry. Each source goes through: discovery of its feeds, verification of what the feed actually is, a licence check on re-use terms, and a health check on reliability. A source that fails the licence check is not ingested, and a source that later objects is removed at the registry level — which drops every page derived from it. The live registry is on the sources page.

2 · Clustering into events

Incoming items are grouped by the underlying occurrence, not the headline. Twenty reports of one court judgment become one event page listing twenty sources. Clustering is reviewed continuously; a wrong merge is treated as a pipeline bug and fixed for the class, not just the case.

3 · Deterministic confidence

Each event carries a cross-source confidence: how consistently independent newsrooms in the registry report the same core facts. It is computed by a fixed formula from source overlap and agreement — never by a language model, and never editable by one. Higher confidence means more independent outlets said the same thing; it is not a guarantee of truth, and pages still distinguish claim types like official claims and allegations regardless of the score.

4 · AI-written briefs

A model reads the clustered sources and drafts the 60-second brief: what happened, why it matters, what changed. The draft is grounded only in those sources; it is marked AI-written on the page. Models run server-side against free-tier providers with automatic failover, and every call is metered. If no provider is available, the story still appears with its headline and source list — an absent brief is never faked.

5 · Freshness and status

Timestamps come from real ingest and update events, not from when a page was rendered. A story moves through developing, breaking and archived states as its sources do. When a story changes we update dateModifiedand keep the original publication time, so “updated” always means genuinely new reporting.

6 · Indexability gate

Not every event deserves the index. Pages below our quality bar — too few sources, low confidence, or thin coverage — are marked noindex and kept out of sitemaps. Empty city and category pages are never published. That is why some internal links may lead to a page that says coverage is still warming up rather than a stub dressed up as an article.

7 · Translation

Indic briefs are machine-translated from the English event, quality-scored, and clearly tied to the same sources. A translation exists only when the scored output passes our bar; we do not publish a thin translated copy just to occupy a language URL.

8 · Markets and prices

Market instruments use delayed public data, each with a stated methodology and an as-of timestamp. Direction colours (green/red) mean financial movement and nothing else. Nothing on a market page is advice or a forecast presented as fact.

What we are still building

Human-at-scene reporting, deeper local desks, and more languages are ahead. Where the site is automated, this page says so; where it is not yet good enough to publish, the indexability gate keeps it off the internet instead of shipping a placeholder.