Agentic navigability audit grid

What this proof makes visible

Agentic navigability can quickly become an abstract term. This proof shows how the concept becomes an audit matrix: observable criteria, weighting, risk examples and remediation decisions.

The grid does not claim to predict the behavior of every agent on the market. It verifies whether a site provides clear enough signals for an agent to understand the interface, maintain context and complete a task without critical ambiguity.

It complements the Agentic Readiness Score and applies directly to the Agentic Readiness Audit.

Audited criteria

Domain	Audit question	Detected risk
DOM stability	Does the initial DOM remain coherent after hydration?	The agent plans an action on a structure that later changes.
Visual stability	Do critical zones move during loading?	The agent clicks the wrong place or abandons the journey.
Multimodal coherence	Do screenshot, DOM and accessibility tree express the same thing?	The agent sees an action but cannot correctly identify it.
HTML semantics	Do actions and navigations use the correct native elements?	A fake button, fake link or ambiguous card blurs the intent.
Forms	Does every field have a label, name and clear relation?	The agent fills the wrong field or cannot understand the error.
Interface states	Do menus, modals, filters and accordions expose their states?	The agent loses context after opening, closing or selecting.
Determinism	Does an action produce a predictable response?	The agent does not know whether the action succeeded, failed or changed step.
Discoverability	Are key resources accessible through links, sitemap and machine surfaces?	The agent finds a secondary page instead of the canonical surface.

Weighting model

The grid is weighted out of 100 points. The heaviest criteria are the ones most likely to block a critical action:

1. DOM and layout stability: 15 points;
2. visual / DOM / accessibility coherence: 15 points;
3. actionable semantics: 15 points;
4. agentic accessibility: 15 points;
5. interaction determinism: 10 points;
6. hydration risk: 10 points;
7. visual hierarchy and affordances: 10 points;
8. machine discoverability: 5 points;
9. WebMCP or agent APIs readiness: 5 points.

This weighting is not a universal metric. It is a prioritization model used in Pagup engagements.

Example reading

A site can have a strong technical SEO score and still be fragile for agents.

Typical example:

• the page loads quickly;
• the main content is indexable;
• the design is clear for a human;
• the primary CTA is a styled <div>;
• the form appears after hydration;
• the fields have no programmatic labels;
• a banner moves the button during loading;
• the submission confirmation is not associated with the form.

In that case, the issue is not only performance. It is interface reliability as an action environment.

What the grid produces

A serious application of the grid usually produces:

• a map of critical agentic journeys;
• a list of divergences between initial HTML, hydrated DOM, visual rendering and accessibility tree;
• an inventory of risky buttons, links, fields, menus and modals;
• a prioritization of components to refactor;
• a hydration-risk estimate;
• a list of tests to prevent regressions;
• a recommendation on whether WebMCP exploration is relevant.

What this proof does not claim

It does not claim that a high score guarantees better visibility in AI responses. It also does not claim that WebMCP, llms.txt or a machine surface is enough to make a site agent-ready.

The grid measures a narrower capability: whether an agent can read, choose and execute an action from coherent signals.

How to inspect it

To verify the logic of this proof, compare three layers:

1. the public Agentic Readiness Score;
2. the Agentic Readiness Audit;
3. a case such as B2B SaaS with rich interface and low agent readability.

If the three layers answer each other, the proof is not merely declarative: it becomes an inspectable method.