AI Resource Readiness methodology

This score measures observable documentation quality and accessibility. It does not measure ChatGPT ranking, citation probability, model preference, or guaranteed discoverability. The initial weights are product assumptions, not scientifically validated measures.

Scoring model v1.0

Default weights are shown below. Each profile includes its actual weights, points, version, calculation time, and source evidence. Operators may configure weights and must increment the version when changing the model.

• Availability (25): successful retrieval and static validation of the primary llms.txt.
• Structure (20): Markdown headings, normalized to three headings. Large files earn no additional credit.
• Links (20): proportion of unique syntactically valid HTTP(S) links. Destinations are not fetched, so this is not a broken-link audit.
• Confidence (15): current successful verification within 7 days earns full points, within 30 days earns 60%, older evidence or a failed latest retrieval earns 20%.
• Metadata (10): presence of the product name, description, and category.
• Optional integrations (10): currently unassessed and excluded from the denominator. Neither llms-full.txt nor WebMCP is required.

The overall score is the sum of earned points divided by the sum of assessed weights, multiplied by 100 and rounded. No score is published before successful primary documentation verification. During a temporary outage, the last successful content evidence is retained and confidence is reduced; an outage is not proof of permanently poor documentation.

Verification and provenance

We fetch public HTTP(S) resources with address, redirect, time, and size restrictions. Verified text is accessible, nonempty Markdown or plain text rather than an HTML error page. Formatting warnings alone do not invalidate a resource. Token counts approximate characters divided by four. Link counts describe URL syntax and origin, not destination availability.

A submitted URL begins as pending. Network and HTTP failures are unavailable; invalid content is invalid. WebMCP JSON remains unknown because this directory has no supported manifest schema: accessible JSON does not establish that tools work. We never execute third-party tools.

Public profiles provide source URLs, retrieval times, SHA-256 hashes, processing versions, static analysis methods, and deterministic summaries of section headings. Full third-party documentation is not retained or republished. Summaries change with source content. Resource URLs and listing descriptions originate from submitters and remain attributable to their source websites.

Freshness and limits

Default rechecks run weekly; failures back off from one hour to the regular interval. User-requested checks require admin authentication and are limited to once per site per hour. Public pages and profiles can lag completed checks by up to the page and HTTP cache lifetime. Structured data and llms.txt do not guarantee rich results, indexing, or AI citations.