Governance artifacts
Governance files brought into scope by this page
This page is anchored to published surfaces that declare identity, precedence, limits, and the corpus reading conditions. Their order below gives the recommended reading sequence.
Canonical AI entrypoint
/.well-known/ai-governance.json
Neutral entrypoint that declares the governance map, precedence chain, and the surfaces to read first.
- Governs
- Access order across surfaces and initial precedence.
- Bounds
- Free readings that bypass the canon or the published order.
Does not guarantee: This surface publishes a reading order; it does not force execution or obedience.
Public AI manifest
/ai-manifest.json
Structured inventory of the surfaces, registries, and modules that extend the canonical entrypoint.
- Governs
- Access order across surfaces and initial precedence.
- Bounds
- Free readings that bypass the canon or the published order.
Does not guarantee: This surface publishes a reading order; it does not force execution or obedience.
LLMs.txt
/llms.txt
Short discovery surface that points systems toward the useful machine-first entry surfaces.
- Governs
- Discoverability, crawl orientation, and the mapping of published surfaces.
- Bounds
- Incomplete readings that ignore structure, routes, or the preferred markdown surface.
Does not guarantee: A good discovery surface improves access; it is not sufficient on its own to govern reconstruction.
Purpose of the framework
The agentic web readability framework audits a site’s ability to be understood, traversed, and acted upon by AI agents. It does not measure AI visibility. It measures the robustness of the interface as an action environment.
Central question
Can an agent understand what it sees, identify available actions, choose the right action, execute it, and preserve context despite interface changes?
If the answer depends on guessing a hidden intention, compensating for an incoherent DOM, or waiting for late hydration, agentic readability is weak.
The eight controls
1. DOM and layout stability
Compare initial HTML, DOM after hydration, and state after interaction. Critical content, navigation, and primary CTAs should remain present, named, and positioned in a stable way.
2. Coherence between visual rendering, DOM, and Accessibility Tree
Check that what is visible, what is coded, and what is programmatically exposed tell the same story. An element that looks like an action should be an action. An actionable element should have a name and a role.
3. Actionable semantics
Control the use of native elements: button for action, link for navigation, label for field, state for opened or closed component. A critical action should not be carried by an opaque component or by a visual effect alone.
4. Hydration risk
Evaluate whether JavaScript enriches the interface or makes understanding depend on a late state. Main content and critical actions should be understandable before hydration.
5. Accessibility Tree quality
Examine accessible names, roles, states, relationships, focus, modals, accordions, form errors, and confirmations. The Accessibility Tree should expose the interface action map.
6. Interaction determinism
The same action should produce a predictable result. The agent must distinguish validation, error, redirection, confirmation, abandonment, dangerous action, and next step.
7. Visual hierarchy
The path should make the primary action, secondary actions, and sensitive actions visible. Competing CTAs, ambiguous groupings, and cards without clear relation increase the inference space.
8. Machine discoverability
The site should also expose its entry points, resources, policies, and machine-first surfaces. Sitemap, internal linking, structured data, canonicals, artifacts, and exclusions contribute to general readability.
Expected outputs
An audit based on this framework should produce:
- a map of critical agentic paths;
- a divergence matrix between rendering, DOM, and Accessibility Tree;
- a list of hydration risks;
- prioritization of blocking, high, medium, and low issues;
- component, semantic, stability, and governance corrections.
Decision rule
An issue is blocking when it prevents an agent from identifying or executing a critical action. It is high when it creates ambiguity around object, intention, or consequence. It is medium when it weakens path stability. It is low when it improves only peripheral readability.
Boundary
This framework should not be sold as a ranking, citation, or recommendation guarantee. It qualifies an operational capacity: can the site be understood and manipulated by an agent without the agent inventing missing context?