llms-full.txt: inspect expanded text, not inferred coverage

GET /llms-full.txt distinguishes observed responses and retrieval unknown. The auditor looks for a bounded leading HTML document, H1/H2 Markdown headings, repeated section-name patterns and configured encoding-like strings.

No source URL/content mapping is constructed. A heading may represent several pages or none. Literal encoding examples may match the pattern without corruption. Absence does not show that AI systems cannot read a site.

First identify an actual consumer and the intended source set. Manually map each exported section to its source URL and compare text and dates. Check raw responses and encoding before treating a text pattern as corruption.

# Reference

Source: https://example.com/reference

Text selected for the intended consumer.

One heading is one heading, not one verified page. The source line helps a manual mapping but is not evaluated as coverage by this check.

Public measurement separates Confirmed issue, Advisory, Unknown, Not applicable and Observed pass. An observed pass applies only to an explicit bounded rule; absence of an emitted failure is not a pass. Missing prerequisites do not earn repair credit.

Start 100. Retrieval unknown returns 0; absent/empty returns 100; leading HTML pattern returns 10. Heading counts, repeated names and encoding-like patterns do not deduct.

Legacy values are clamped to 0..100 and rounded. They are not readiness certification, risk indicators or predicted citation outcomes. Method changes do not rewrite historical records.

The observations above describe the local implementation, not guarantees from these sources. Protocol and provider documents help interpret purpose and perform checks this audit does not implement.

• llms.txt proposal and generated text context