InquirySpec - Ontological Boundary: The Stratified Semantic Authority Graph organizes artifacts by authority layer so agents do not flatten traces, commitments, and standards. - Not This: Not a vector database or generic knowledge graph. - Doctrine Dependencies: The_Anatomy_of_Action, Persistent_Context, Unified Glossary.
Working Definition
A Stratified Semantic Authority Graph is a semantic web with authority layers. It organizes artifacts so the system can distinguish a raw trace, a local note, an interpreted claim, a draft model, an adopted commitment, a reviewed procedure, a governing standard, and a consequence record.
The point is not only to connect related things. Ordinary retrieval can already return related things. The point is to preserve what kind of thing each artifact is, where it sits in the authority field, what scope it governs, what review state supports it, and what action it is allowed to constrain.
This graph is the structural counterpart to Warrant Gravity. Warrant Gravity names the pull that different artifacts exert on action. The Stratified Semantic Authority Graph gives that pull a navigable topology. It lets a system say: this item is findable, but it is only a trace; this item is relevant, but it is local to one team; this item is reviewed, but it has expired; this item is adopted, but only for a defined domain; this item is a consequence signal, so it may reopen a higher layer for review.
Without that stratification, retrieval becomes dangerously flat. A search result, embedding match, chat excerpt, policy document, and model summary can all appear inside the same interface and look equally actionable. The graph exists to stop availability from masquerading as authority.
The Phenomenological Problem
The ordinary failure is the information equivalency trap.
A person searches the archive and receives a mixed pile of artifacts: a meeting note, an incident trace, a draft procedure, a hallway summary, a formal standard, a model answer, and a consequence log. All of them may be relevant. They are not equivalent.
Under time pressure, the artifact that is easiest to retrieve often becomes the artifact easiest to act on. This is systemic gravity. People do not need to intend distortion for the system to drift. The interface returns a list. The list looks coherent. The top result feels convenient. The recent result feels alive. The official-looking paragraph feels authoritative. A model can then blend the pile into fluent language, making the differences even harder to see.
This is how low-warrant artifacts gain accidental force. A raw trace becomes treated as an operating rule. A local workaround becomes treated as global practice. A stale standard suppresses a current consequence signal. A confident summary hides the fact that its sources come from different authority layers. A private team namespace is projected outward as if it spoke for the larger organization.
The problem is not retrieval. Retrieval is necessary. The problem begins when retrieval is asked to perform governance. A database can return what is connected. A search engine can return what is similar. A model can return what fits a pattern. None of those operations, by themselves, decide what a group is warranted to rely on.
The graph corrects that failure by making artifact force visible. It does not prevent low-authority artifacts from being found. It prevents them from being silently promoted.
The Engineering Anchor
The internal action anatomy beneath this node treats knowledge as a stratified field rather than a flat store of facts. At the lowest layers, the system captures signals and traces. Higher layers contain interpreted artifacts, local commitments, organizational procedures, cross-organization standards, and environmental constraints. Each layer can be queried, but each layer carries a different relationship to action.
The doctrine's key warning is simple: syntactically valid graph retrieval is not the same as semantic authorization. A query can return a connected subgraph while still returning a stale edge, a low-warrant trace, a partition-scoped artifact, or a record that has never been promoted into a governing role. If an agent treats the returned payload as ready for action merely because the query succeeded, the architecture has confused finding with authorizing.
The Stratified Semantic Authority Graph therefore requires more than nodes and links. It needs labels for authority layer, properties for validity horizon and distortion risk, edges that show warranting or promotion, and metadata that preserves source, scope, namespace, and review state. These features let retrieval hand candidates to governance instead of pretending to be governance.
Persistent memory supplies the substrate. A memory object must carry provenance, metadata, relationships, access controls, and retrieval paths. But persistence alone is insufficient. A durable artifact can still be local, provisional, superseded, or out of scope. The graph adds the authority coordinate that tells the system what kind of reliance the artifact can support.
This is why the node also depends on The Anatomy of Action. Authority is action-relative. A raw trace may be enough to open an inquiry, but not enough to impose a general rule. A local procedure may guide one team, but not settle a cross-team dispute. A governing standard may constrain routine work, but a new consequence record may require that standard to be reopened.
The Unified Glossary provides the lexical discipline. Terms such as trace, note, claim, commitment, procedure, standard, warrant, and consequence must not collapse into one another. If language flattens the artifact types, the graph will flatten in practice even if the database remains technically structured.
Boundary Conditions
The Stratified Semantic Authority Graph is not a vector database. Vector search can be useful for finding semantically similar material, but similarity does not decide whether an artifact may govern action.
It is not a generic knowledge graph. A generic graph may show that two artifacts are connected. This graph must show how they are connected, under what authority layer, with what warranting edge, and within what validity horizon.
It is not a confidence score. A fluent answer, high retrieval score, or frequently cited artifact may still be low-warrant for the action being considered.
It is not a guarantee that higher layers are always correct. A reviewed standard can become stale. A local trace can reveal a failure the standard has not yet absorbed. Stratification does not freeze authority; it makes authority inspectable, contestable, and revisable.
It is not an excuse to ignore raw material. Traces, anomalies, and local reports matter. The graph's job is to keep their role clear. A trace can trigger inquiry. It should not automatically become a rule.
It is not a moral ranking of people or teams. It ranks artifact force inside a bounded action context. The same person can produce a low-warrant observation, a reviewed analysis, and a superseded draft. The artifact's status, scope, review state, and consequence relation determine its pull.
It is not finished by retrieval. Retrieval is the start of the route. Governance, review, action, and repair complete the route.
Drill Path
Start with Warrant Gravity to understand why artifacts exert different action force and why availability should not be confused with authority.
Move to The Anatomy of Action to see why warrant depends on the action being attempted, not only on the artifact's abstract status.
Use Unified Glossary when the language around artifact types begins to drift. The graph cannot preserve authority if the vocabulary collapses the layers it is supposed to separate.
Use this node whenever a workflow says "the system retrieved it," "the model found it," "the graph connects it," or "the document says it" as if that settled what the artifact is allowed to govern.
The compact rule is: retrieval finds candidates; stratification preserves their role; governance evaluates their warrant for action.