Registry

Maintained by ApexClaw Research · engine ARB/1.1 · last revised 2026-08-26

Every site scored with the Agent Readiness Benchmark, the score it received, and a link to the run that produced it. The registry fills as runs complete; nothing is listed that was not measured.

What does a grade mean here?

A grade is a band on the same 0-100 score: A from 85, B from 70, C from 55, D from 40, E below. Bands describe the score, they do not certify a control, an audit standard or a compliance regime.

49 sites scored with engine ARB/1.1. Each row links to the run that produced it.

How are the bands set?

Bands sit on fixed cut points, not on a curve, so a grade means the same thing in every run.

The engine holds every pillar's weights to a total of 100%, applies its heaviest single pillar weight of 14% to no more than two pillars, and completes a typical run in under 2 seconds with most findings resolving in under 800 ms -- see the exact figures for any listed row in the machine-readable rubric.

How do I get my site listed?

Run it at the audit form. A completed run is listed with its score and its evidence. A run that could not fetch the target is not listed at all.

• Scoring is unattended; no submission or review step sits between the run and the listing.
• A run reads one URL plus the files at the domain root.
• Typical run time is under 2 seconds, and results are not cached beyond 24 hours -- see how a run works for the full pillar breakdown.

Can a listing be re-measured?

Yes. Fix a finding, run it again, and the new run appears with its own date. Earlier runs are kept, so a change over time is visible rather than overwritten.

• Re-running after a single fix isolates the delta attributable to that change.
• Every row names the engine version that produced it, so scores stay comparable across rubric revisions.
• A score of 100 is reachable: no check requires a paid product or a third-party certification.

Does a listing imply endorsement?

No. A row records that a public URL was fetched and scored on a stated date. It is a measurement, not a recommendation, and it carries no commercial relationship of any kind.

The distinction matters because the register is deliberately open. Any public URL can be scored by anyone, including a competitor scoring us, and every row links to the evidence that produced it. A registry that only listed flattering results would be marketing; one that lists whatever was measured, in the order it was measured, is a record. Rows are added, never removed, and a later run never overwrites an earlier one.

What a listing does not mean

ARB/1.1 measures whether this one public page is legible and independently verifiable to an autonomous agent: structured data, machine-readable dates, off-site corroboration, and the page's public transport and security surface. It does not measure the organisation's actual security posture, the quality of its product, or its reputation -- those require access, judgement, and time this engine does not have and will not simulate. Its security_surface pillar reads public response headers and TLS parameters observed on one fetch; it is NOT a penetration test and must never be read as one.

A row on this page is not a security clearance, a certification against ISO, SOC 2 or any other compliance regime, and not a statement about whether the underlying business or product is trustworthy. It is a record that one URL was fetched on a stated date and scored against a published, versioned rubric. A site can score a perfect 100 here and still have real defects the rubric was never designed to measure -- the pillars audited are machine access, structured data, extractability, evidence and citability, entity clarity, the agent interaction surface, governance and trust signals, off-site corroboration, and the public transport and security surface. Nothing else is scored, and nothing else should be inferred from a row appearing here.

How is this different from a search ranking?

A search index ranks pages by relevance to a human query and assumes a person will click through, skim, and judge the result themselves. This registry does none of that: it runs the same fixed set of deterministic checks against every URL, in the same order, and reports exactly which ones an agent reader could verify without inference. Two sites with very different search rankings can land on the same row here if they expose the same structural facts to a machine reader, and a site that ranks well in search can still score low here if its content is only legible to a human -- gated behind client-side rendering, phrased as adjectives instead of checkable facts, or missing the structured markup a script can parse without guessing. The two measurements answer different questions and neither substitutes for the other.

Sources

The formats this registry's checks read against are themselves public, versioned specifications, not house rules:

Method and weights: the published rubric. Machine version: /methodology.json.

---

About ApexClaw · Contact · Privacy · Terms · Agent access policy