Readiness
earned weight ÷ conclusive weight
Pass earns full weight. Warning earns half. Fail earns none. Unable to verify is excluded rather than treated as failure.
Transparent by design
Every score begins with evidence.
AXplore runs deterministic checks against public website signals. No language model decides whether a check passes, and the same evidence always produces the same score.
Public evidence only Fixed scoring rules Inspectable findings
Scoring model
Two numbers. Two different truths.
Readiness describes verified performance. Coverage tells you how much of the applicable assessment was actually verified.
Assessment coverage
conclusive weight ÷ applicable weight
Coverage falls whenever evidence cannot be verified. A high readiness score with low coverage is an early signal, not a verdict.
Result states
Uncertainty is visible.
- Pass
The required evidence was found.
- !Warning
Evidence exists, but is incomplete or malformed.
- Fail
The check conclusively found a barrier.
- —Not applicable
The optional interface does not apply and has no score impact.
- ?Unable to verify
Evidence could not be retrieved; coverage falls, not readiness.
Deterministic assessment
16 checks. Nothing hidden.
Each check has a fixed weight, a specific evidence source and an explicit limit. Optional interfaces never penalise sites that do not use them.
01
Discoverability
Can an agent locate the site, its public map and its preferred identity?
Check & evidenceWeightWhat it verifiesKnown limit
HTTPS transportNormalized URL scheme7%The normalized origin is served over HTTPS.Certificate quality, redirect chains or mixed content on inner pages.
Homepage availabilityHTTP status and content type10%The homepage returned a successful status through the scanning transport.Availability from other regions, or behaviour behind bot protection.
robots.txtRoot response and excerpt8%A plain-text robots.txt was retrieved at the site root.Whether its rules are enforced, or rules served conditionally per user agent.
XML sitemapParsed XML root8%A parseable urlset or sitemap index exists at the location declared in robots.txt, or at /sitemap.xml.Additional undeclared sitemap locations, or whether every listed URL resolves.
Canonical URLHomepage link element4%A canonical link element is present in the homepage HTML.Whether canonical targets are correct across the rest of the site.
02
Content & structure
Can an agent extract useful content and machine-readable facts from the homepage?
Check & evidenceWeightWhat it verifiesKnown limit
Markdown negotiationNegotiated content type7%The content type returned when the homepage is requested with Accept: text/markdown.Negotiation on inner pages, or the quality of returned Markdown.
llms.txtRoot response and excerpt5%Substantive non-HTML content at /llms.txt.Whether the guidance is accurate, current or honoured by any tool. This is an emerging convention.
Structured dataParsed homepage JSON-LD10%JSON-LD blocks on the homepage and whether they parse.Schema correctness, deeper-page coverage, microdata or RDFa.
Readable page contentExtracted text length10%How much readable text the homepage returns without executing a user journey.Whether that text answers the questions an agent actually has.
Commerce offer detailProduct and Offer JSON-LD fields7%Where Product or Offer JSON-LD exists, price, currency and availability are present.Deeper product pages or details expressed only in prose. Without Product schema, it is not applicable.
03
Agent access & interfaces
Does the site declare access rules or optional interfaces that agents can use?
Check & evidenceWeightWhat it verifiesKnown limit
AI crawler accessGrouped robots.txt directives11%Root-level default access and named AI crawler policies, parsed by user-agent group.Edge rules, rate limits or blocks applied outside robots.txt.
OpenAPI documentParsed /openapi.json shape4%A JSON document with an OpenAPI or Swagger field and non-empty paths.Documents at non-standard paths or in YAML, and whether documented endpoints work.
OAuth metadataParsed well-known JSON3%Authorization-server metadata with issuer and authorization_endpoint fields.Metadata at non-standard paths, or whether documented flows work.
MCP discoveryParsed well-known JSON2%A discovery document with transport, servers or MCP markers.MCP servers advertised only in documentation or at non-standard locations.
A2A agent cardParsed well-known JSON2%An agent card with a non-empty capabilities array.Agent cards at non-standard paths or discovered through other conventions.
AI plugin manifestParsed well-known JSON2%A plugin manifest with schema_version and name fields.Manifests at non-standard paths or described only in documentation.
Validation
What we know — and what we do not.
Check accuracy
Saved real responses and hand-authored edge cases must produce their expected results. A disagreement fails the test suite.
Stability & sensitivity
Identical evidence produces identical scores, and no single check can move overall readiness by more than 15 points.
Predictive accuracy
The first 20-site study found a moderate positive relationship with task success (r = 0.43). That is not enough evidence to claim prediction.
See the method in practice