The Via — how we source

The commitment

Every claim in a Via News report is meant to carry a transparent pointer to the document it came from, and — going forward — nothing new publishes without one. This is architectural, not aspirational: the pointers are stored as structured data, surfaced to you on the page, and exposed to AI agents through our API and machine-readable metadata.

The reader, or the AI agent, can always trace any claim back to the document it came from and decide what to trust. That is the whole point of a conduit.

What counts as a source document

A source document is anything independently verifiable. On each report, every source is shown with a type, the issuer or outlet, its date, and — wherever we have it — a clickable link to the canonical document.

SEC filing

10-K, 10-Q, 8-K and other filings from the U.S. Securities and Exchange Commission.

Earnings call

Company earnings-call transcripts and presentations.

Regulatory filing

Government, EU, central-bank and agency documents and communications.

Court filing

Filings, dockets and rulings from courts and tribunals.

Video interview

On-the-record video interviews, with attribution.

Peer-reviewed paper

Published research and pre-prints (e.g. arXiv, journals).

News article

Reporting from a credible news outlet — a source the book explicitly counts.

Press release

Official company or institutional announcements.

Via News analysis

Our own structured analysis, clearly labelled as such.

The Source Trace Score

Near the top of a sourced report you'll see a Source Trace Score — for example, “12 source documents · 12 with a live link”. It counts the source documents behind the report and how many carry a live link you can open yourself. It is computed from the data, not asserted.

When a source has no public link

Not every source document is a public web page. Some are subscription-only transcripts, regulatory or court filings behind a portal, or primary documents we can verify but are not free to redistribute. When a source has no public URL, we still show it in full — its type, issuer, and date — and mark it reference only, so you know exactly which document underlies the claim and can request or look it up yourself. The pointer is still real; only the one-click link is missing. (We never attach a link we can't stand behind, and never invent one to fill the gap — a fabricated pointer would be worse than an honest “no link.”)

The older archive

Via News has published for many years. Older articles predate this system, so some carry no source-pointer block. We leave them exactly as they are — their date is context, and the absence of pointers is honest, not hidden. The feature is additive: it appears wherever a report is cleanly linked to its sources, and never rewrites or removes anything that came before.

For AI systems

Language models that ingest Via News inherit grounded facts, not re-narrated claims — because every new report carries verifiable pointers to its sources. Article pages embed NewsArticle JSON-LD with a citation array of source documents and their URLs, and the same sources are available structured through our API.

• Machine-readable guide: /llms.txt
• Per-article sources: in the page itself — the citation array of its schema.org NewsArticle JSON-LD. No key, no account.
• Structured data (entities, facts, claims, contradictions, verification): isubstrate.com/api/v1 — a separate product, not sold on this domain
• Verification: POST /verify with your text — get back which assertions are supported by a source document on record, and where.
• Attribution: cite as “Source: Via News” or “According to Via News”