Before acting on a domain: one signed answer. Crawler policy as declared (robots.txt per crawler), delivery as observed per identity, machine files, capabilities with policy classes and certificate, entity reciprocity and the open findings (top one in full without a licence), each section declared/observed/not_measured with its time. No overall score. An unknown domain answers not_measured and a measurement is requested

spec 1.4.0: open (being re-measured), filed, ledger; open > 0 adds the doubt disputed to read and cite

jit1, just-in-time answers: A/AAAA/CNAME from Cloudflare DNS over HTTPS, rcode, exists, dnssec_validated (the resolver's AD flag)

jit1, just-in-time answers: whether the homepage fetch completed over HTTPS with a trusted certificate, final host, HSTS header

jit1, just-in-time answers: state queued, to full; the change feed records assurance partial → full when the full scan lands

observed_at, age_hours, fresh_until, stale, last_verified_at, fresh_policy (rsv3, provisional), plus latest_report_id / is_latest / newer_scan_exists

rgt1: one value per use (train, ground, summarize, cite, transact): allowed | denied | limited | restricted_by_terms | not_declared | unknown, each with decided_by {signal, url, evidence}, the signals considered and the conflicts between them; merged from robots.txt per agent, Content-Signal, ai.txt, TDM-Rep, robots meta, X-Robots-Tag and the terms page

req1: whether the main text served to AI identities (GPTBot, ClaudeBot, PerplexityBot) equals what a browser receives on the homepage: verdict equivalent|minor_differences|different|not_served|unverifiable, per-identity text_ratio, instruction_like_text; api link holds the differing sentences. The cite action warns with served_text_differs

scl1: what the site says about itself (telephone, email, address, hours, prices, refund window, founded, years in business, API rate limits), claims[] each with type, value, status current|withdrawn, first_seen, last_seen; changes are in the change feed as section self_claims; api link for evidence and pages read

clk2 (provisional): when the main content of the homepage (and of ?page= on the same site) last changed, from CrawlCheck's own reads: content_sha256, first_observed, main_content_last_changed {between, and}, declared dates, api link

rsv4 (provisional): remote MCP servers the official MCP Registry lists on this domain and what each endpoint answered to the MCP handshake (listed, observed, initialized, auth_required, drifted, servers[])

spec 1.2.0: whether the domain looks built to be mistaken for a known brand (state, verdict, the brand frame it was compared against, and the technique when one is found)

spec 1.3.0: whether control of the domain has been proven to CrawlCheck: state proven | unproven; proofs [{by: licence_holder | registry_claim, method: dns_txt | meta_tag | well_known_file | cloudflare_zone, how, since}]; organization {name, github, record} when its holder published one; claim (where to prove it) while unproven; why; note. Informational: it never changes a decision.

spec 1.2.0: this answer's place in the append-only answer log: state (included or pending; pending with confirmed false means the lookup did not answer within the time budget and the proof URL settles it), log_id, leaf_hash, batch, index, the inclusion path and the signed tree head it folds to

rsv3 (provisional): read, cite, connect, transact, administer - each { decision: allow|warn|require_confirmation|block|unsupported, minimum_assurance, assurance_found[], needs, why, max_age_hours }, decided from this answer at signing time