Sourcing & attribution
How we pick sources, weight them, attribute them, and decide when something’s solid enough to publish.
Status
This policy is a first draft and will be reviewed by a California media-law attorney before launch.
Our sourcing tiers
When more than one source is available, we weight them in roughly this order:
- Primary documents. Court filings, council agendas and packets, signed contracts, public-records responses, raw public datasets, photographs and recordings we made or that the source verifiably made themselves.
- Government press releases. Useful for the official position of an agency. Not a substitute for the underlying primary document.
- Other newsrooms. Cited and linked when we are downstream of someone else’s reporting. We do not paraphrase verbatim ledes; see copyright.
- Social media. Only with verification of who posted it and when. Useful as a tip — rarely sufficient on its own.
Attribution
We name our sources unless a confidentiality concern is clearly grave enough to outweigh the reader’s interest in knowing. Anonymous sources are used sparingly and only with the editor’s prior approval.
When we link, we link to the primary source first.
Confidential sources
When a confidential source provides material that informs a story, we:
- Document the source’s identity in our internal records.
- Document the source’s motivation and any conflicts.
- Verify the source’s information through at least one independent channel before publication.
- Disclose to the source what they should expect about their identity being protected and the limits of that protection.
California’s shield law (Cal. Const. Art. I §2(b) and Evidence Code §1070) protects journalists from being compelled to disclose confidential sources or unpublished information in court. O’Grady v. Superior Court extended that protection explicitly to online journalists.
Verification
A factual claim that affects a person’s reputation, livelihood, or freedom is verified against at least one primary source before publication. If we cannot verify, we either don’t publish that claim or we make the gap visible to the reader.
For AI-assisted drafts, our fact-check step (Claude Sonnet 4.6 + Citations API) cross-references each claim against the ingested source documents. The composite confidence score combines independent supporting citations, source-tier weight, and Claude logprob. We do not one-click publish a story with a claim scoring below 60.
Copyright
We don’t scrape paywalled sources. When we are downstream of another outlet’s reporting, we say so explicitly, link out, and cap quoted material at roughly 50 words. AP v. Meltwater (S.D.N.Y. 2013) held that aggregating verbatim ledes is not fair use; AI paraphrases that closely track another outlet’s structure carry the same risk.
State and local government works in California are not automatically public domain — County of Santa Clara v. Superior Court — so we attribute when we use them and we don’t reproduce datasets in their entirety without permission.