Cortex Taxonomy

Glossary (glossary.md) defines terms one at a time. This page shows how they cluster.

Layers

Cortex separates four canonical substrate roots. Arrows here show operational relationships, not import or dependency direction:

flowchart TD
    A[Algebra<br/>graph/relation laws]
    W[Wire<br/>source, contracts, compiled circuit]
    P[Pulse<br/>durable runtime]
    C[Capability<br/>executor registration]

    A -- "topology laws for" --> W
    W -- "compiled circuit runs on" --> P
    C -- "grants registered authority to" --> P

Algebra is the pure substrate Wire authors over. Mechanized theory names the safety contracts that runtime implementation must enforce. Wire owns the compiled circuit form. Pulse executes compiled circuits durably. Capability exposes executor authority. Value provenance and artifact references are runtime/contract concepts, not a current public Haskell root.

Roles a value can play

A single Wire value often wears multiple hats. The roles are:

The leading @ marks the executor-authority boundary. It stages a registered executor with pure config data; it does not run the executor. Wire-authored CorePure output equations are written directly without @; they lower to the native pure evaluator inside the admitted Wire semantics. See Reference/Wire/executors-and-alphabet.md.

Artifacts a contract can carry

Every contract declares a payload kind that selects validation and rendering:

Namespaces

Cortex has a small closed set of globally-ambient namespaces:

Beyond these four, every name is local: introduced by let, node, or import. The executor alphabet is authority-bearing; ordinary config constructors are not. The syntax distinction is the leading @.

Downstream Logos library

Logos is a downstream reasoning library built on Cortex. It adds reusable LLM-shaped catalogs above the substrate. Logos.Archetypes classifies modes of cognition, Logos.Thought names one bounded model-mediated node evaluation, Logos.Memory owns cognitive context construction, and Logos.Patterns names reusable reasoning programs. None of these grants executor authority or creates new runtime registries.

Each archetype may have an operational activation bundle at Logos.Archetypes.<Archetype>.Activation. The archetype is the semantic definition; the activation bundle is the concrete set of prompt discipline, retrieval corpus, embedding spaces, tool surface, memory policy, evaluation criteria, and runtime contract. Thought frames compose one or more archetype activations.

Logos.Patterns.DeepReport is the planned extraction target for reusable deep-report reasoning contracts, ports, templates, prompts, memory presets, and evaluation policy.

Doc-kind taxonomy

Docs are classified on two axes:

Cross-cutting canon: Templates/, index.md, map.md, glossary.md, taxonomy.md Consumer-specific: Consumers/{consumer}.md, or Consumers/{consumer}/ for a larger public binding Papers: Publications/paper-N-*/

Related

• map.md — where to put a new doc.
• glossary.md — term definitions.
• Architecture/01-overview.md — architectural starting point.
• Reference/terminology.md — normative terminology.