Production audit protocols

Each protocol has a bounded observation contract. None is a vanity subscore.

Audit

Record what each named protocol observed.

Diagnose

Trace each production finding to evidence.

Verify

Re-run the protocol required by the remediation.

Why recorded protocols beat a universal score

A network failure, an intentional training opt-out, a browser-only content deficit, and invalid JSON-LD are not interchangeable. The result keeps these observations separate, preserves unknowns, and derives a verdict only from required production protocols.

Protocol 1

Direct HTTP acquisition

Records: HTTP status, final URL, bounded redirects, declared content type, byte length, selected response headers, and artifact hash.

Does not prove: That a provider used the same network path or accepted the page downstream.

Protocol 2

Robots policy acquisition

Records: The public /robots.txt response as a separate, hashed policy artifact.

Does not prove: That an absent policy file caused a provider to crawl the site.

Protocol 3

Actor-specific robots decision

Records: The matching group, longest applicable allow/disallow rule, actor purpose, and policy decision for the audited path.

Does not prove: Actual access, indexing, training, or compliance by the provider.

Protocol 4

Raw and rendered representation

Records: Distinct raw HTTP, rendered browser, and no-JavaScript browser artifacts with mechanical hashes and byte counts.

Does not prove: Semantic equivalence or a universal crawler representation.

Protocol 5

Main-content extraction

Records: Deterministic extraction outcomes, word counts, hashes, and material raw-to-rendered deficits.

Does not prove: Content quality, truth, usefulness, or universal language readability.

Protocol 6

Structured-data consistency

Records: JSON-LD parse diagnostics, schema types, raw availability, and bounded consistency for selected material string properties.

Does not prove: Eligibility, correctness of every property, rich-result display, or downstream use.

Protocol 7

Machine-readable dates

Records: Strict ISO dates, supported meta fields, JSON-LD dates, time elements, and HTTP Last-Modified evidence.

Does not prove: That undated content is stale or that a declared date is factually true.

Protocol 8

llms.txt acquisition

Records: The public /llms.txt response as a separate hashed artifact: presence, bounded byte length, media type, and the required title heading.

Does not prove: That any provider fetched, parsed, or used the file.

Protocol 9

AI-usage header policy

Records: Site-declared x-robots-tag AI tokens and TDM reservation headers read from the captured raw HTTP response for each audited actor.

Does not prove: That a provider honored the declared signals or that the headers constrain downstream use.

Execution state is part of the result

SUCCEEDED

The protocol completed and produced its required observation.

FAILED

The protocol failed with a typed cause and retryability state.

ABSTAINED

Inputs were present but insufficient for a defensible conclusion.

NOT APPLICABLE

The protocol does not apply to this representation or actor.