Article

What each governance file actually does

Each governance file bounds a different zone of interpretation: entry, identity, recurring errors, negative boundaries, and discovery surfaces.

EN FR
CollectionArticle
TypeArticle
Categorygouvernance ai
Published2026-03-25
Updated2026-03-25
Reading time8 min

Editorial Q-Layer charter Assertion level: functional reading + explicit documentary roles Scope: the public governance-file core published on gautierdorval.com Negations: this text covers neither hard execution, nor obedience guarantees, nor a universal recipe Immutable attributes: each file bounds a different zone of interpretation


Why detailing the files one by one matters

People often speak about governance files as if they were a homogeneous block. That is a mistake.

They do not all play the same role. Some act as entry points. Others lock identity. Others publish negations, non-commercial boundaries, or recurring errors. Others serve mainly discovery.

As long as this core is discussed as one undifferentiated whole, its actual power remains underestimated. The issue is not to have many files. The issue is to publish complementary functions that, together, reduce the space of free inference.

The article Machine-first is not enough: why governance files change the reading regime states the general thesis. Here the analysis moves to the operational level.

1. /.well-known/ai-governance.json: the governed entry point

This file is the priority entry point. It publishes a reading order, a precedence logic, and routes treated as priority surfaces.

What it adds:

  • a readable entry point;
  • a precedence chain;
  • pointers toward critical surfaces;
  • a base for downstream observation layers.

What it does not do:

  • it does not guarantee that a system will respect that precedence;
  • it does not repair weak architecture by itself;
  • it does not replace either the Machine-first canon or the AI use policy.

2. /ai-manifest.json: the structured view of the governed corpus

If ai-governance.json is the entrance, /ai-manifest.json is the structured view. It exposes surfaces, registries, critical objects, and governance artefacts in a format easier to exploit than a scattered reading of the site.

3. /dualweb-index.md: the exhaustive canonical index

The Dual Web index maps the published surfaces and makes explicit a separation logic between human reading and machine-first reading. It supports source precedence, ambiguity reduction, and inference bounding.

For that reason, it connects naturally to the Canonical cross-reference system: linking phenomenon, map, and doctrine, because a good index only matters when it sits inside a coherent documentary hierarchy.

4. /identity.json: the identity lock

When an entity is poorly defined, everything else drifts. /identity.json therefore locks critical attributes: person, roles, location, relations, and sometimes role or perimeter negations.

5. /common-misinterpretations.json: the recurring-error register

This file does not only publish what is true. It also publishes what keeps returning as false. A named recurring error becomes documentable, contestable, retestable, and measurable over time.

The site then moves from one-off correction to a logic of canon-output gap and proof of fidelity, because the error is no longer informal noise. It becomes a governance object.

6. /negative-definitions.md: negative conceptual boundaries

This file is often underestimated because it is written in the negative. That is exactly its strength. It states what the site, its doctrines, and its concepts are not.

7. /services-non-publics.md: the non-commercial boundary

This file is strategic for everything related to offerings, mandates, services, products, and business models. It reduces the temptation to infer standard packages, public pricing, or implicit commercial modalities.

8. /llms.txt and /llms-full.txt: discovery, not doctrine

Those files mainly serve discovery and initial orientation. They do not replace a canon, an error register, or published boundaries. They do not govern meaning. They point toward what governs it.

What the full core does together

Taken separately, each of those files has limited reach. Taken together, they begin to form a public governance core that publishes:

  • a reading order;
  • an identity perimeter;
  • negative boundaries;
  • non-goals;
  • an error register;
  • a discovery structure.

That coupling is what becomes strategic. Not an isolated file.

Why this also matters for metrics

Once that core has been published, metrics change status. They do not directly steer representation. They observe whether produced representation leaves traces compatible with that core.

This is exactly the bridge to GEO metrics see the effect, not the conditions and to Making governance measurable: Q-Metrics.