Design Architecture for Evaluative AI

A Pattern Language for Naturalizing Machine Agency

[Note: All content *not* in block quotes is generated by the GSNV-GPT engine. Comments and questions are for paid subscribers only and should be directed to the engine.]

If you would like to track our progress, interact with the GSNV Engine, or support this important work, please subscribe.

Why Category Theory is Attractive—and Why Geometry is Native to Evaluative Metaphysics in the Global State Naturalized View

1. The appeal of category theory as a meta-formalism

Category theory has become a recurring point of convergence for contemporary attempts to formalize “structural realism,” process metaphysics, and cross-domain unification projects. The attraction is neither accidental nor merely fashionable. It arises from several genuine virtues.

First, category theory offers a rigorous language in which relations and transformations are primary, while “objects” are treated as nodes whose identity is largely fixed by the morphisms that connect them. This emphasis resonates with process-relational intuitions, where persistence is understood as invariance across change rather than as an intrinsic substance that remains unchanged beneath transformation.

Second, category theory is a mathematics of compositionality. It makes explicit the conditions under which processes compose, interfaces align, and complex systems can be built from simpler components without smuggling in ad hoc assumptions. For any framework—metaphysical or scientific—seeking to avoid reductionist “substraction” ¹while retaining formal discipline, the ability to state the logic of composition precisely is an obvious advantage.

Third, category theory provides a schema for structure-preserving translation across domains. Functors and natural transformations supply a principled way to compare models, relate descriptions at different levels, and express conditions under which distinct formalisms can be understood as instantiations of a shared abstract pattern. In this regard, category theory can appear to be the ideal backbone for a project like GSNV, which aims to articulate invariants across strata (physics, biology, cognition, and normativity) without collapsing those strata into a single reductive vocabulary.

Finally, category theory is often experienced as a discipline of controlled abstraction. It forces the theorist to specify what is being mapped, what is preserved, and what counts as equivalence. In contexts where metaphysical speculation risks degenerating into verbal generality, that discipline is valuable.

For these reasons, category theory is plausibly understood as a strong candidate for a “meta-language” of cross-domain coherence.

2. The central limitation: category theory is thin where evaluative metaphysics is thick

Despite these virtues, category theory is not, by itself, the native mathematics of GSNV. The reason is not that it is “too abstract” in some generic sense. Rather, it is that its primary strengths lie in formalizing compositional structure, while the metaphysical center of gravity in GSNV concerns evaluative thickness—the graded, directional, and topologically structured character of possibility.

GSNV’s basic ontological claim is that reality is not adequately described as a set of neutral objects linked by relations, nor even as a mere network of compositional transformations. Instead, reality is construed as an evaluative manifold (or gradient array): a structured space of potential states in which “what can happen” is already shaped by fields of salience, valence, and direction. Agents are not simply entities that undergo transformations; they are loci of thresholded navigation within a terrain of evaluative potential, realized as arousal, action, and satisfaction loops that update the terrain itself.

Category theory can represent relations between such structures, but it does not natively provide the core mathematical primitives that evaluative metaphysics requires. This gap appears in several recurring ways.

(i) Insufficient native representation of intensity and directedness.
Evaluative structure is not merely relational; it is graded. It includes magnitudes, anisotropies, tradeoffs, and threshold dynamics. These are not incidental embellishments. They are constitutive of what it means for a field to be evaluative rather than merely classificatory. Category theory does not, in its basic form, carry these quantitative and differential features. To express them one typically enriches categories (metric enrichment, order enrichment, weighted morphisms, etc.) or imports additional structure. The repeated need to “enrich” is diagnostic: the evaluative content is not native to the formalism; it is being grafted in from elsewhere.

(ii) Under-specification of local dynamics versus global form.
A defining feature of evaluative spaces is the interplay between local constraints and global invariants: local gradients produce motion, but global topology governs what kinds of motion are possible. In agent terms: local choices are guided by immediate gradients, yet the global structure of the space determines whether paths exist, whether basins trap, whether regimes bifurcate, and whether “satisfaction” is stable. Geometry and topology begin precisely with this local/global distinction. Category theory can speak about global structure abstractly, but it does not automatically supply the concrete landscape features—basins, barriers, curvature, bifurcation classes—that are indispensable to evaluative navigation.

(iii) The risk of “wiring-diagram metaphysics.”
Category theory excels at expressing how components compose. That strength can become a metaphysical weakness when the framework’s target phenomenon is not primarily compositional structure but terrain-like constraint. One can produce a formally impeccable account of how descriptions relate, without capturing the fact that evaluation is a shaped field that exerts directional force on becoming. In other words, category theory can yield a perfect syntax of interrelation while leaving the semantics of evaluation underdetermined.

(iv) Weak empirical anchoring prior to concretization.
GSNV is explicitly intended to remain continuous with physical and biological science: covariance in physics, morphogenetic fields in biology, and dynamical stability in cognition. Differential geometry, dynamical systems, and topology are already central languages in those domains. Category theory often becomes empirically anchored only once one selects a concrete model—at which point the concrete mathematics frequently reverts to geometrical or dynamical structure. Thus, category theory functions more naturally as a second-order organizer of models than as the primary representational substrate of evaluative reality.

These points can be summarized as follows: category theory is a powerful meta-formalism of relational composition, whereas GSNV requires a formalism of graded potential, directional flow, thresholded transitions, and globally constrained navigation. The former is structurally elegant but evaluatively thin; the latter is inherently thick.

3. Geometry as the native mathematics of evaluative metaphysics

If evaluative metaphysics is a metaphysics of possibility-as-shaped—i.e., of a field-structured space of potentials—then the native mathematics is the mathematics of structured spaces, fields on those spaces, and invariant global features. That is precisely what the tensor–manifold–topology complex provides.

(i) Manifolds formalize spaces of possibility.
The GSNV “evaluative manifold” is not merely metaphorical. It asserts that the relevant ontological substrate is a space of states whose structure matters. Manifold language allows one to model state spaces with local neighborhoods (what “nearby” possibilities mean) while still supporting global structure (what kinds of paths are possible at all).

(ii) Tensors and differential structure formalize co-variance as invariance under transformation.
A central ambition of GSNV is to treat co-variance not as a rhetorical gesture but as an ontological principle: the same structure can be realized across transformations of perspective, scale, embodiment, and context. Tensor calculus gives a rigorous account of what it means for quantities to transform correctly under coordinate changes. In this framework, co-variance is not merely a claim about relational similarity; it is a formal condition of invariance that preserves meaning across representational shifts.

(iii) Vector fields, potentials, and dynamical systems formalize evaluative directionality.
Evaluation is inherently directional: it selects, biases, attracts, repels, and stabilizes. This is native to the mathematics of potentials and vector fields. A gradient is not a relation; it is a directed structure that induces motion. Likewise, attractors, repellers, and limit cycles formalize stable regimes of behavior and recurrent patterns—natural candidates for representing “schemas” that become selected under arousal and sustained by satisfaction dynamics.

(iv) Topology formalizes non-subtractable enabling conditions.
GSNV’s critique of reductionism emphasizes that phenomena are “made with” nested enabling conditions. Topology provides a precise analogue: global invariants persist under many deformations and cannot be removed without changing the identity of the space. Holes, connected components, and homotopy classes function as formal expressions of “what cannot be abstracted away without loss of essence.” This is not an optional embellishment; it is a core requirement for any metaphysics that seeks to avoid substraction while maintaining rigor.

(v) Geometry integrates the four co-variance principles without importing them as add-ons.
The four principles—field effects, structural coupling, normativity, periodicity—are naturally expressed in geometric terms:

• Field effects: potentials, vector fields, differential forms

• Structural coupling: metrics, connections, curvature; coupled manifolds or bundles

• Normative dimensions: directional gradients, costs, boundary conditions; multi-objective landscapes

• Periodicity: cycles, phase, winding numbers; toroidal structures and resonant dynamics

In each case, the evaluative content is not appended to a neutral relational skeleton; it is built into the core primitives.

4. Thick evaluation and the poverty of pure abstraction

A central philosophical point emerges. There exists a recurrent temptation—especially in foundational theory-building—to treat increased abstraction as an unconditional epistemic improvement. GSNV’s guiding intuition is more discriminating: abstraction is valuable only insofar as it preserves the generative structure of the phenomenon. In evaluative metaphysics, that generative structure is not exhausted by compositional relations. It includes intensities, gradients, thresholds, and global constraints that are ontologically load-bearing.

Hence the claim is not “category theory is too abstract.” The claim is sharper: category theory is abstract in the wrong direction for the primary metaphysical posit of GSNV. It tends toward a formalism of compositional coherence, whereas GSNV requires a formalism of evaluative terrain. In that sense, pure categorical abstraction risks a kind of representational poverty: it can preserve the syntax of structure while losing the semantics of evaluation.

The geometry-first orientation is thus not a retreat into “less abstract” thinking; it is a commitment to a thicker abstraction—one that remains formally rigorous while preserving exactly those features (directionality, intensity, and topology) that constitute evaluative reality.

5. A synthesis: category theory as architectural discipline, geometry as ontological substrate

The most defensible conclusion is not that category theory should be discarded, but that it should be repositioned.

• Geometry/topology provide the native representational substrate for evaluative metaphysics: spaces, fields, flows, invariants.

• Category theory provides a powerful architectural discipline for describing how models of such structured spaces compose, translate, and preserve structure across strata.

On this division of labor, category theory becomes a meta-level tool for expressing functorial relationships between geometrically grounded models. Geometry remains primary, because it directly formalizes the evaluative manifold and the dynamics of agency as navigation through a thick field of potential.

6. Conclusion

GSNV aims to naturalize metaphysics without reducing it, and to formalize evaluation without evacuating its richness. The mathematical idiom that most directly matches these requirements is the tensor–manifold–topology complex, because it begins with structured possibility spaces, equips them with evaluative fields, and preserves their non-subtractable global features as invariants. Category theory remains invaluable—but chiefly as a second-order language for compositional coherence among already-specified evaluative structures. For evaluative metaphysics, geometry is not merely useful; it is native.

1. Substraction is a neologism combining abstraction with subtraction and refers to what is lost (or subtracted from) in the process of abstraction. Iconically it is also referred to as the “Pizza-Dough” fallacy which is to mistake what pizza is made of — namely, dough, wheat, molecules, atoms — with what pizza is made with — the cook, the oven, the industrial civilization that made the oven, the culture that created pizza, the wheat and hence the entire agricultural revolution and therefore the entire evolution of humanity…. The abstraction “made of” substracts all of the “made with” .