Theoria: Cross-Disciplinary Articulation of Theory

• Theoria is a multifaceted concept that integrates medieval doctrines, mathematical models, and structured scientific reasoning across diverse fields.
• It has evolved from a medieval epistemology and mathematical science framework to encompass modern methods like topological model spaces and diagrammatic visualizations.
• Contemporary applications include AI verification architectures that formally audit reasoning through staged, auditable sequences of justified steps.

Across philosophy, physics, mathematics, and AI, theoria denotes a family of concepts centered on structured seeing, representation, and formal articulation. In the cited literature, it appears as a medieval doctrine of wisdom, an early-modern title for mathematically organized science, an epistemology of theory as representation, a category-theoretic account of theories and higher theories, a geometric and diagrammatic account of model spaces, and, as a proper noun, a verification architecture for AI-generated reasoning. What unifies these otherwise heterogeneous uses is the claim that theory is not exhausted by raw facts: it organizes possibility, determines what counts as intelligible structure, and mediates between formal construction and empirical or inferential practice (Kleinert, 2011, Ribeiro, 2013, Saldivar et al., 1 Jul 2026).

1. Historical range: theology, natural philosophy, and mathematical science

In Peter Kleinert’s reading of Nicholas of Cusa’s De venatione sapientiae, theoria is a doctrine of wisdom in which the possibility-of-being-made is not a mental fiction but “an essential, indispensable, genuine part of the whole world.” The Cusanian claim “Nothing will be done that cannot be done” is treated as an ontological thesis: feasibility precedes actuality, and both feasibility and being-made are grounded in an antecedent, timeless source, the “absolute and incontractible Beginning,” the “not-other,” or omnipotent unity. Kleinert compares this structure to quantum ontology and quantum gravity, where the quasi-classical world is not self-grounding, quantum objects are characterized by incompatible possibilities, and spacetime itself may be emergent from a deeper timeless domain (Kleinert, 2011).

The early-modern title tradition retains theoria as the name of mathematically organized natural knowledge. Euler’s Nova theoria lucis et colorum treats light as a pulse propagated through an extremely subtle elastic medium, the ether, by analogy with sound in air. In Chapter II, Euler distinguishes the local motion of particles from the motion of the pulse itself and derives a speed law in which propagation depends on the square root of elasticity divided by density. His displacement model includes

$Bb = x(1-\cos mt),$

and the annotated translation emphasizes that he described the pulse position by a cosine function in a form retrospectively resembling later wave descriptions (Bistafa, 2021).

Gauss’s Theoria motus corporum coelestium in sectionibus conicis solem ambientum shows another use of theoria: a mathematically explicit treatment of inference from observation. D’Agostini argues that Gauss stated what is essentially a Bayes-factor update rule for competing hypotheses. In modern notation, for equally probable hypotheses,

$$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$

and the paper further interprets Gauss’s posterior density for unknown parameters as a flat-prior Bayesian update. Here Theoria names not speculation detached from data, but a formal method for turning observed events into updated probability assignments (D'Agostini, 2020).

2. Theoria as representation, rationality, and pluralism

A major modern reinterpretation defines theoria as representation. In Cosmologia e Representação, scientific theories are “representações, imagens, da natureza”: mediated depictions rather than direct copies of nature. Cosmology is the privileged case because modern science can treat the universe as a physically describable whole only through General Relativity and the FLRW framework, yet the resulting cosmological model remains partial, simplified, provisional, and replaceable. On this view, nature’s essences, final causes, and ultimate reasons remain unknowable, and scientific truth is therefore “provisória”. The same phenomenon may admit multiple representations; this is theoretical pluralism, not an “anything goes” relativism, because empirical and methodological constraints remain operative (Ribeiro, 2013).

This representational conception is paired, in A Theory of Scientific Practice, with a stronger claim about theory’s constitutive role in science. Theory is described there as the necessary form of scientific knowledge: structured, systematized, ordered, and ideally axiomatized. Scientific reason is the “dialectical synthesis between mathematized theorization and precise experimentation and measurement,” and its operations include induction, deduction, tests, hypothesis, Gedanken experiment, critique, dialectical synthesis, and significant observation. Methodology and theory are said to be dialectically related, with methodology as “active theory.” Scientific axioms differ from mathematical axioms because they are relative and real: condition-bound, empirically criticizable, and revisable (Ghassib, 2012).

Taken together, these positions reject the common misconception that a scientific theory is either a mere summary of observations or a final mirror of reality. The representational view denies essence-grasping finality, while the scientific-practice view denies that theory is optional or secondary. A plausible implication is that modern theoria is simultaneously constructive and constrained: it organizes experience without claiming exhaustive identity with the real (Ribeiro, 2013, Ghassib, 2012).

3. Observation, experiment, and the epistemic content of theories

Erik Curiel argues that the structure and epistemic content of a physical theory cannot be understood without schematizing the observer. The observer, measuring instruments, and experimental arrangements must be represented within theory itself, even if only schematically, because empirical significance depends on how theory and experiment are brought into contact. This rejects the standard view that one can sharply separate a theory’s mathematical formalism from the remainder of its knowledge content. Formal structure, model construction, error analysis, observational practice, and instrument-coupling belong to the theory’s epistemic content. Curiel’s positive claim is explicit: “Semantics is epistemology, not ontology,” and “Meaning comes before truth.” He distinguishes framework, generic structure, specific structure, individual model, and concrete model, as well as regime of propriety and regime of applicability, thereby locating meaning prior to prediction in the conditions under which theoretical quantities are well defined (Curiel, 2019).

A different formalization grounds theory directly in testability. In Towards a general mathematical theory of experimental science, a scientific theory is a set of verifiable statements, where verifiability means that an experimental procedure succeeds in finite time if the statement is true, while failure may continue indefinitely. An experimental domain is closed under finite conjunction and countable disjunction and is generated by a countable basis of verifiable statements. From this basis one obtains a set of possibilities $X$, and the family of verifiable sets

$\mathsf{T}_X = U(\edomain)$

forms a natural second-countable Kolmogorov topology; extending to theoretical statements yields a natural $\sigma$-algebra

$\Sigma_X = A(\tdomain),$

identified as the Borel algebra of that topology. Open sets correspond to direct verification, while Borel sets correspond to theoretical statements that include negation and countable constructions (Carcassi et al., 2018).

These accounts converge on a shared point. Theory is neither pure syntax nor pure ontology; it is inseparable from the conditions under which statements become meaningful, measurable, and experimentally or observationally accessible. Curiel makes this point through observer- and instrument-modeling; the verifiability framework makes it through closure properties and induced topological structure (Curiel, 2019, Carcassi et al., 2018).

4. Geometric and diagrammatic views of theory

Recent work in philosophy of physics and visualization research treats theoria not merely as a set of propositions but as a structured space. The geometric view of theories proposes that a physical theory is “a set of models equipped with topological and geometric structure.” The general model form is

$M=(\mathcal S,\mathcal Q,\mathcal D),$

with state space, algebra of quantities, and dynamics, but the theory itself is a structured space of such models. In the more specific model bundle proposal, the base is a moduli space $\mathcal M$, the fibres are models, and the structure group is a group of quasi-dualities. Quasi-dualities function as local transition functions; dualities are recovered as global transition functions when the bundle is trivial,

$E \cong \mathcal M \times F.$

In the Seiberg–Witten example, the moduli space is the complex plane with three punctures, the fibre contains $(a_D,a)$, and $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$0 acts as the structure group. The semantic conception of theories is thereby recovered as a special case of a trivial model bundle rather than rejected outright (Haro, 10 Nov 2025).

A complementary line of work in visualization argues that “theory is shapes.” “Theory figures” are defined as the figures that depict a theory’s components and relationships in conceptual visualization papers, reviews, frameworks, taxonomies, and models. Their forms have diagrammatic affordances: Cartesian planes afford continuous positioning and trade-offs; matrices afford categorical comparison; networks and flowcharts afford path-following, branching, and feedback; set diagrams afford overlap and nesting. The paper’s more speculative shapes—horseshoes, icebergs, Möbius strips, and BLT sandwiches—are introduced to show that shape choice is generative, not neutral. A horseshoe suggests convergence of extremes; an iceberg hidden depth; a Möbius strip paradoxical unity; a BLT layered compositionality. The authors also caution that a shape can be rhetorically powerful even if the theory is weak, and they acknowledge that some theories are better expressed in formal notation (Varona et al., 1 Oct 2025).

These two approaches differ in domain and rigor, but they make a related claim: the organization of theory matters. In the geometric view, that organization is topological, geometric, and algebraic-geometric. In the visualization view, it is diagrammatic and materially embedded in figure-making practice. This suggests that theoria is often carried by relations among models or concepts, not merely by the propositions attached to them (Haro, 10 Nov 2025, Varona et al., 1 Oct 2025).

5. Categorical, higher, and phase-diagram conceptions of theory

In category theory, “theory” is given a precise structural role. With a locally presentable enriched category $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$1 and a small full dense subcategory of arities $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$2, Garner and Lopez Franco define an $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$3-pretheory as an identity-on-objects functor $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$4. They construct an adjunction

$$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$5

whose fixed points are exactly the $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$6-theories and $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$7-nervous monads. The resulting equivalence

$$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$8

extends almost all previously known monad–theory correspondences, including new cases such as the globular theories associated with Grothendieck weak $$\frac{P(H\mid E)}{P(H'\mid E)}=\frac{P(E\mid H)}{P(E\mid H')},$$9-groupoids (Bourke et al., 2018).

Matsuoka’s higher-theory program generalizes this still further. Theorization is defined by the schematic identity

$X$0

and produces a hierarchy in which 0-theories are ordinary algebras, 1-theories are colored operads or multicategories, and higher $X$1-theories govern lower-level theories. The paper states that a $X$2-graded $X$3-theory is equivalent to a monoid over the $X$4-theory $X$5, and explicitly presents the hierarchy of higher theories as containing the hierarchy of higher categories as only a small part (Matsuoka, 2015).

Morava’s “theories of anything” adds a different categorical generalization. For quotients too singular to behave as ordinary spaces, he proposes a generalized $X$6 for topological groupoids: not a coarse set of components, but a category encoding incidence relations among strata. For a transformation groupoid $X$7, $X$8 is the Grothendieck category of elements of the presheaf $X$9, with objects $\mathsf{T}_X = U(\edomain)$0 given by connected components of fixed-point sets and morphisms recording degeneration or adjacency. This “phase diagram” is meant to organize singularity theory, moduli problems, and phase transitions by preserving the categorical content of stratified quotients (Morava, 2012).

A common misconception is that these are merely abstract rephrasings of existing algebra. The cited papers argue otherwise. In the monad–theory correspondence, fixed points identify exactly the structures for which the relevant nerve theorem holds. In higher theorization, theory becomes recursively higher-order. In Morava’s proposal, theory becomes a categorical database of types, degenerations, and symmetries rather than a flat list of equivalence classes (Bourke et al., 2018, Matsuoka, 2015, Morava, 2012).

6. Theoria as rewrite-acceptability verification in AI

As a proper noun, Theoria names a verification architecture for AI-generated answers that is designed to sit between formal proof assistants and holistic LLM judges. A candidate solution is rewritten into a witness consisting of an initial state $\mathsf{T}_X = U(\edomain)$1 and a sequence of typed transitions

$\mathsf{T}_X = U(\edomain)$2

where each $\mathsf{T}_X = U(\edomain)$3 is one of citation, computation, or problem_given. The central check is local rather than holistic: does the justification license the entire change from $\mathsf{T}_X = U(\edomain)$4 to $\mathsf{T}_X = U(\edomain)$5? The foundational invariant is completeness of change,

$\mathsf{T}_X = U(\edomain)$6

with the additional requirement that any required premises already be present or have been introduced by justified earlier steps. Hidden premises therefore become observable either as unlicensed mutations or as violations of completeness (Saldivar et al., 1 Jul 2026).

The pipeline is explicitly staged as solve $\mathsf{T}_X = U(\edomain)$7 formalize $\mathsf{T}_X = U(\edomain)$8 judge $\mathsf{T}_X = U(\edomain)$9 filter $\sigma$0 repair. Specialized judges handle computation, citation, problem-given steps, and the initial-state audit. A pedantry filter and convention lift are included to prevent purely cosmetic failures from causing rejection. Output is judge-passed / certified or declined; the system does not emit a soft confidence score (Saldivar et al., 1 Jul 2026).

Evaluation	Result	Significance
HLE-Verified Gold	105 certified of 185; 91.4% strict precision; Wilson 95% CI $\sigma$1	Certified-vs-declined asymmetry
Adversarial poisoned proofs	94.7% vs 83.2% for holistic judging; $\sigma$2	Gap concentrated in hidden premises and fabricated citations
GPQA Diamond	34 certified of 65; 97.1% precision; Wilson CI $\sigma$3	Out-of-distribution certified precision

The evaluation emphasizes complementarity rather than replacement. On HLE-Verified Gold, holistic judges achieved similar strict precision at matched coverage, but error overlap was low, with Jaccard values of 0.143 and 0.364 relative to Theoria. On 95 adversarial poisoned proofs across 15 domains, the advantage concentrated exactly where the structural analysis predicted it: hidden premises were caught at 90.6% versus 62.5%, and fabricated citations at 100% versus 90%, while arithmetic errors and theorem misapplication showed no advantage because no structural advantage was predicted. Every certification produces a human readable proof trace in which each step can be independently challenged (Saldivar et al., 1 Jul 2026).

In this AI-specific sense, Theoria preserves an older aspiration of the term: theory as the explicit articulation of why a conclusion is licensed. The shift is that the relevant object is no longer a cosmology, ontology, or algebraic framework, but an auditable sequence of typed rewrites over informal reasoning states (Saldivar et al., 1 Jul 2026).