VextJS
  • English

      • Documentation Data and AI

      VextJS exposes documentation as both human pages and deterministic build artifacts. This makes search, AI-assisted analysis, and documentation quality checks easier without asking readers to trust an opaque crawler or a tracker.

      Public machine-readable assets

      AssetPurpose
      docs-manifest.jsonGenerated page metadata: canonical URL, locale, summary, audience, applicability, stability, related pages, and source hash.
      capabilities.jsonSupported frontend/runtime capabilities and explicit non-goals. Cite it together with the linked detail page.
      ai-gold-questions.jsonCitation-required questions that prevent answers from inventing unsupported capabilities.
      llms.txtConcise English entry points for language models and documentation tools. It is an index, not a crawler-control file.
      llms-full.txtComplete English URL-and-summary index: every public English documentation page appears exactly once.
      zh/llms.txtConcise Simplified Chinese entry points, isolated from the default English index.
      zh/llms-full.txtComplete Simplified Chinese URL-and-summary index: every public Chinese documentation page appears exactly once.
      docs-events.schema.jsonOptional privacy-preserving event contract. No collector is enabled by VextJS.
      docs-dashboard-definition.jsonMetric definitions and collection boundary for a site owner who later chooses a compliant collector.

      The machine artifacts are generated after the documentation build. They contain no build timestamp, so identical source creates identical metadata and hashes.

      Language and completeness contract

      All four llms*.txt files are deterministic UTF-8 Markdown served as plain text. The root files contain English only; files under /zh/ contain Simplified Chinese. llms.txt is intentionally curated so a model can find the main reading paths without loading the whole site. llms-full.txt is the exhaustive index for its locale, with one canonical URL and source-derived summary for every page. docs-manifest.json remains the authoritative bilingual inventory and records each entry's locale and source hash; the build verifies exact per-locale coverage.

      The “full” files are complete indexes, not copies of every page body. A web-capable assistant follows the canonical URLs to the rendered documentation; an offline tool can use the manifest and indexes to choose the exact pages it needs. This keeps the machine entry small enough to parse while preserving 1:1, build-verified coverage.

      How an AI answer should use the docs

      1. Locate the relevant entry in docs-manifest.json and cite its canonical URL.
      2. Check capabilities.json before saying that a frontend feature is supported or excluded.
      3. For RSC, Server Functions, Server Actions, PPR, and bundler assumptions, read Frontend Boundaries and Roadmap rather than inferring support from React, SSR, Suspense, or Streaming SSR.
      4. Use ai-gold-questions.json as a regression set for documentation-based answers. A plausible answer without a source is not an accepted answer.

      Measurement is optional and privacy-first

      VextJS does not ship a tracker, analytics SDK, collector endpoint, cookie, or identity graph for this documentation site. The event schema intentionally allows only page, locale, event kind, referrer class, optional search length, and CTA class. It explicitly excludes raw search text, URL query values, credentials, page content, and user identity.

      A documentation site owner may wire a collector later only after choosing the provider, legal basis, retention, consent behavior, and security review. The JSON files define what an implementation may measure; they are not permission to collect data and do not by themselves measure revenue or conversion.

      To report a documentation gap, open a GitHub Discussion.