Can its public pages be reached and understood?
We check redirects, access rules, useful content, structure, and consistency across a small set of pages.
How the scanner decides.
One point for each applicable check. Public evidence determines what counts, what passes, and what still needs to be verified.
How we decide
From the first public page to the final score.
Find what applies.
Start with the public website and declared capabilities. Must-have checks count even when evidence is missing.
- Yes
- Include the applicable check
- No
- Exclude unless must-have
Read the evidence.
Every applicable check has the same weight. More pages provide evidence, not extra points.
- Pass
- One point earned
- Fail
- Zero points earned
Calculate the score.
Divide passed checks by counted checks, then multiply by 100. Apply the strictest unmet tier gate to the overall score.
- 18/20
- 90 / 100 · Tier A
- Skip
- Unassessed stays out unless must-have
A timeout is not a failed observation. Unassessed checks stay out of the score unless marked must-have. A tier gate still requires a passing result.
What counts. What doesn’t.
Every public website is checked for reliable access and useful, understandable content. Additional capability checks normally apply when the site advertises or publishes that capability. Must-have checks and tier gates are explicit exceptions.
Extra ways to use the site
- API
- Developer tools
- Authorization
- Agent Skills
- A2A
- ARD
- AI Catalog
- MCP
- WebMCP
- Commerce
- Bot Identity
An undeclared feature is neutral unless one of its checks is marked must-have or carries a tier gate.
Six tiers. The same rules.
Each tier describes a band of final scores. Must-have checks count missing evidence as zero; unmet tier gates cap the overall score.
- S
- 95–100
- Exceptional
- A
- 85–94
- Excellent
- B
- 70–84
- Well prepared
- C
- 50–69
- Getting there
- D
- 30–49
- Limited
- E
- 0–29
- Needs attention
Scores are displayed as whole numbers, rounded down. Tier boundaries are product conventions, not guarantees of agent success.
What this means in practice
These examples show how scope, must-have checks and tier gates affect the result.
Eighteen checks pass. Two need work.
All 20 counted checks have enough evidence to reach a conclusion, with no unmet tier gates.
90 / 100 · Tier A
18 ÷ 20 × 100. Each failed check has a concrete finding and, when supported, a fix.
One of those checks is still unresolved.
Eighteen pass, one fails, and one non-mandatory check could not be verified. No tier gate applies.
94 / 100 · Tier A
18 ÷ 19 × 100, rounded down. The unassessed check is excluded from the score and listed in scan coverage.
The site has no API or WebMCP tools.
It publishes useful pages without claiming to offer those capabilities. None of their checks is must-have or a tier gate.
No points lost
Undeclared capabilities stay out of the score. Relevant ideas may appear in Go to next tier, separately.
A must-have check is missing.
Eighteen checks pass, one fails and one must-have check is N/A.
90 / 100 · Tier A
18 ÷ 20 × 100. The missing must-have check earns 0/1 instead of leaving the denominator.
The raw score reaches S, but a gate has not passed.
A check is configured with maximum tier B and is failed or unassessed.
84 / 100 · Tier B
An illustrative raw score of 98 is capped at 84, the highest whole-number B score. Go to next tier puts the modules blocking tier A first.
Every point has a check behind it.
See what each check looks for, why it matters, and which evidence supports it. Discover, Access and Use regroup the same checks; they never add extra weight.
What the number cannot say.
The score describes the evidence we could inspect. It is not a probability that an agent will complete a task.
- Whole numbers, not false precision.
- Displayed scores are rounded down to a whole number. The final score, after any tier cap, decides the tier, so the display never lifts a site into a higher tier.
- One score. Three perspectives.
- Discover, Access and Use each show the percentage of their own assessed checks that pass. The overall score comes directly from all checks, then applies any tier cap. Category scores show uncapped percentages.
- Missing evidence stays missing.
- N/A and unassessed checks stay out of the score unless marked must-have, when they earn 0/1. Missing supporting pages are noted in scan coverage. A tier gate requires a pass. A scope with no applicable checks shows N/A; if nothing could be assessed, it has no score.
- Tier gates require a pass.
- A check with maximum tier B caps the overall score at 84 until it passes, even if the raw score is 98. The strictest unmet gate wins. Removing a gate does not guarantee the next tier: its score threshold still applies.
- A broken promise still counts.
- When a site advertises a capability, its applicable checks remain in scope even if the promised file is broken. Different protocols activate independently.
- Go to next tier is a separate question.
- Modules with checks blocking the next tier appear first, followed by remaining improvements in descending potential points. Points count unique checks that could pass, not a promised increase out of 100.
- Prepared does not mean proven in use.
- The scan reads public pages and contracts without running code, signing in, calling tools or making transactions. Closed or unreachable targets receive no score. A capability endpoint is assessed within its own scope.