Site coherence map: governance layer for interpretive paths

Site coherence map

This page formalizes an additional governance layer: the governance of interpretive paths.

This layer is not meant to restate identity, prohibitions, licensing, doctrinal precedence, or response-legitimacy rules. Those functions already exist in the AI use policy, the machine-first canon, the entity graph, and the SSA-E + A2 + Dual Web surfaces.

Its function is different: to make explicit how an agent should connect pages, read their direct neighborhoods, and reduce the need to guess plausible URLs.

In practice, this layer responds to an observed field condition: agents may access a clean, dense, and governed site, then request unpublished URLs simply because they infer a path that seems coherent to them. Those 404s do not necessarily mean that content is missing. They can signal a gap between the site’s real graph and the logical graph reconstructed by the agent.

What this layer governs

This layer governs:

• direct neighborhoods between canonical URLs;
• minimum documentary dependencies between pages;
• recommended reading sequences that reduce hazardous inference;
• concept equivalences between FR and EN versions;
• relations of clarification, dependency, application, and scoping;
• recovery paths when an agent searches for a plausible page that does not exist.

It does not govern:

• the truth of a concept in place of its canonical definition;
• the admissibility of an exogenous source in place of the EAC layer;
• the legitimacy of answering in place of the Q-Layer;
• the doctrinal hierarchy in place of the endogenous canon;
• exhaustive site crawling in the way an XML sitemap does.

In other words, this layer does not replace definition, doctrine, or policy. It governs the passage between surfaces.

Why this layer becomes necessary

In an interpreted web, an agent does not always follow structure like a classical document engine.

It may:

• read a page, then return to governance files;
• connect multiple surfaces without visiting every intermediate URL;
• infer that a page should exist from a slug pattern, a conceptual cluster, or a perceived hierarchy;
• build a local logical path without retaining the real path published by the site.

This is precisely where “ghost 404s” emerge: not as internal site errors, but as external projections of an expected graph.

When that situation becomes recurrent, publishing more content is not always the right response. One often needs to improve the topological readability of the corpus instead.

Normative thesis

A governed site should not only publish what it is, what it authorizes, and what it denies. It should also publish the minimum interpretive paths that allow an agent to move correctly through the corpus without producing fictional neighborhoods.

This thesis introduces an important difference between two forms of internal linking:

1. exploration linking, which helps discover pages;
2. coherence linking, which helps understand which pages belong together, in what order, and for what reason.

A site can already be very good at exploration linking while remaining partially ambiguous from the standpoint of local coherence.

What a site coherence map should contain

The candidate form of this layer is a companion file of the site-coherence-map.json type.

As of the publication date of this page, this name designates a doctrinally proposed surface. It should not be read as the claim that a universal standard already exists, nor as the declaration that such a file is already deployed everywhere in the ecosystem.

At minimum, a site coherence map should expose for each important URL:

• the canonical URL;
• its role (definition, doctrine, framework, clarification, policy, entity-reference, etc.);
• the main concept or main scope;
• first-hop neighbors;
• minimum documentary dependencies;
• the pages that clarify its reading;
• the pages to which it applies;
• the linguistic or conceptual equivalent;
• the recommended reading path when coherence stakes are high.

Difference from other artifacts

Difference from ai-manifest.json

The manifest inventories surfaces, scope, roles, and entry points. It does not describe with enough granularity local page neighborhoods.

Difference from entity-graph.jsonld

The entity graph formalizes entities, concepts, semantic relations, and doctrinal links. It is not always sufficient to govern URL-by-URL traversal.

Difference from llms.txt and dualweb-index.md

These surfaces guide reading and signal priorities. They do not finely describe local documentary dependencies.

Difference from sitemap.xml

An XML sitemap essentially says: “here are the published URLs.” A coherence map should say: “here is how they hold together.” It therefore does not replace the sitemap. It complements it.

Concrete use cases

This layer becomes useful when an agent:

• slightly alters a slug although the page already exists under another form;
• looks for a derived definition without reaching the right canonical node;
• jumps directly from an observation article to a doctrinal layer without passing through the required clarification;
• confuses two close pages because their relation is not explicit enough;
• loops across the same surfaces because no stabilized local path has been published.

In each of these cases, the question is not first “what text should be added?” but rather “which neighborhood should be made enforceable?”

Strategic consequence

This layer opens a different kind of audit.

Instead of starting only from existing pages and their performance, one can start from:

• URLs inferred by agents;
• ghost 404s observed in logs;
• back-and-forth movements between pages and governance artifacts;
• revisits that signal local hesitation or the absence of a stable path.

The goal is no longer only to fix SEO errors, but to reduce the gap between the published graph and the reconstructed graph.

Doctrinal status

This layer should be understood as:

• a topological complement to SSA-E + A2 + Dual Web;
• a coherence-governance surface;
• an artifact designed to reduce fictional neighborhoods, URL inference, and erratic traversal.

It is not, in itself, a promise of visibility, citation, or compliance from third-party agents. It is a reduction of unguided interpretive space.

Implementation principle

If a companion file of this type is published, it should respect five principles.

1. No duplication of truth

It should not redefine identity, doctrine, or prohibitions already published elsewhere.

2. Explicit locality

It should describe local neighborhoods, not only broad global hierarchies.

3. Agentic readability

It should be readable as a traversal instrument, not as narrative commentary.

4. Multilingual compatibility

It should make FR/EN equivalences explicit rather than assuming an agent will infer them correctly.

5. Version discipline

Any substantial modification of neighborhoods, dependencies, or recommended paths should be traceable.

Conclusion

Governance in the interpreted web can no longer be limited to publishing rules, definitions, and identity files.

As agents produce plausible paths, revisit governance artifacts, and reconstruct the local coherence of a corpus, an additional layer becomes relevant: the governance of interpretive paths.

The site coherence map is not meant to replace existing artifacts. It is meant to make them operate better together by publishing not only the nodes of the system, but also the minimum paths of their articulation.

Associated reading

• AI use policy
• Machine-first visibility doctrine
• Interpretive governance
• Internal linking in the age of second-hop retrieval
• Multi-hop retrieval
• Why an XML sitemap is no longer enough: toward a site coherence map for AI agents