How claims move from source to page

The method is designed to preserve uncertainty, source ownership, and lifecycle state—not to maximize the number of pages.

Updated Core source baseline How we verify claims

The publication workflow

  1. Observe a source. Capture its owner, date, source class, and exact proposition.
  2. Create a candidate claim. State no more than the source supports.
  3. Review the claim. Check date type, lifecycle state, counterevidence, and terminology.
  4. Generate static outputs. Pages, timeline data, feeds, and the sitemap come from the same reviewed records.
  5. Verify mechanically. Test routes, citations, metadata, disclosures, exports, indexability, and preview isolation.

Evidence classes are labels, not scores

Primary

Original research

A paper or dataset establishing its own reported result.

Official

Documentation

The maintained owner record for a current interface or support matrix.

Claim owner

Company statement

Authoritative for what the company announced, not independent proof of every assertion.

Public record

Regulatory filing

A formal record with its own scope and legal context.

Dates describe different events

Announcement, effective, completion, reporting-period, publication, and verification dates are stored separately when the distinction matters. A year-only source stays year-only; the interface does not invent a day for visual neatness.

The first public editorial date of these pages was not established by the supplied evidence, so page schema omits datePublished instead of backfilling one. Recorded substantive revision dates remain available as dateModified.

Search growth policy

The growth strategy is cumulative utility: answer stable questions, cite at claim level, link updates back to evergreen context, and improve pages when new evidence changes the answer. The site does not generate near-duplicate pages for every keyword variant.

[39][40]

Current publication state

Public reference: canonical pages are open to search indexing and listed in the sitemap. The isolated preview environment remains blocked from indexing through its HTTP response header.

[38]

Source monitoring and review due

The editorial queue uses the source registry and a separate driver-specific monitoring registry. First-release sources are active only where a page or claim is in scope; dormant sources stay dormant until reviewed. Review-due dates are triage signals, not evidence that a change occurred.

Automated checks may surface candidates, but publication still requires claim-level source review, access/reuse clearance where needed, and an explicit revision note.

Download the source monitoring registry (CSV) Download the driver source registry (CSV) Open the driver dataset page

Security bulletins: source changed October 1

NVIDIA PSIRT states that from October 1, 2026, it publishes security bulletins only in the NVIDIA product-security repository. Review the affected products, branches and fixed versions in each bulletin; the old web listing is not evidence of completeness after the cutover.

[70]

Evidence trail

Sources used on this page

Source class describes ownership and form. It is not a reliability score.

  1. Creating helpful, reliable, people-first content

    Google Search Central · Current guidance

    Official documentation
  2. Spam policies for Google web search

    Google Search Central · Current guidance

    Official documentation
  3. Block search indexing with noindex

    Google Search Central · Current documentation

    Official documentation
  4. NVIDIA product-security repository

    NVIDIA on GitHub · Current repository

    Official source repository

Core source registry dated ; later page and record revisions carry their own dates. An inaccessible link triggers source maintenance.