Definitions, formal results, programs, experiments, admission decisions, and proposed architectures are complementary forms of support. None is a substitute for all the others.
Standard categorical mathematics and a broad family of proposed categorical models for AI coexist in one expansive introduction.
Computational role
Implementations, notebooks, and a Lean companion realize selected constructions at different levels of coverage.
Empirical role
Application chapters report task-specific results for diagrammatic learning, geometric transformers, causal systems, and related architectures.
Boundary. A categorical formulation, executable notebook, or machine-checked theorem does not establish the empirical adequacy of every associated AI claim.
Learning sketches, typed obstructions, tangent response, quotient, localization, repair, and admission form the most explicit axiomatic framework in the sequence.
Computational role
LINCS-X systems instantiate the six-stage workflow and selected formal companions under declared observers and repair languages.
Empirical role
Application chapters test whether optional structural signals help under particular estimators, baselines, and admission rules.
Boundary. A structural theorem does not imply that a scalar observer is informative or that an available repair improves a task; admission remains separate.
DIAL, higher-DIAL, and DIAL-X are proposed contracts, separated from cited theorems on tangent categories, Weil algebras, and classical Lie theory.
Computational role
CLIC, OPTIC, RELIC, AGENTIC, ARTISTIC, and DILATE supply component realizations with explicit target and artifact contracts.
Empirical role
The book reports controlled component demonstrations and staged testbed ladders, with ARTISTIC providing the most visible repair evidence.
Boundary. The studies do not establish general autonomous creativity; ARTISTIC repair is not transformational creativity, and local DILATE evidence is not a newly admitted artistic generator.
ORACLE formulates categorical identification under doctrinal priors, then specializes it to persistent UOCL, PACC learning, UODL, coherent repair, and lifelong transport.
Computational role
The book specifies learner interfaces, observers, comparison maps, and stabilization obligations, while leaving most implementations and machine-checked refinements as future work.
Empirical role
Grounded examples from infant cognition, automata, world models, particle physics, LEGO, Rubik’s Cube, online learning, and agentic safety motivate and test the scope of the abstractions.
Boundary. Formal categorical structure does not by itself establish identifiability, tractability, empirical adequacy, or autonomous AGI; each requires separate assumptions and evidence.
From status to source
See the selected record behind the chapters.
The curated evidence library connects public papers and software, together with nonpublic frozen experiment packets, to the precise chapters they inform. Every item carries its own claim boundary. Open the evidence library →
Shared status vocabulary
What each label licenses.
DefinitionDeclares an object or term inside a stated framework.Does not establish existence, usefulness, realizability, or uniqueness.
Formal resultDerives a consequence from named assumptions.Does not establish empirical adequacy, novelty, or value.
ImplementationShows that a specified construction executes.Does not by itself establish semantic correctness or superiority.
Controlled demonstrationSupports a bounded claim under stated controls.Does not establish generalization beyond the registered environment.
AdmissionAssigns accept, reject, or abstain under a named certificate.Does not imply universal truth or permanent scientific acceptance.
Proposed architectureStates a falsifiable construction and its obligations.Does not denote a finished or empirically superior system.
Environment counts describe manuscript structure. They are not a score for rigor, originality, or correctness.