Native Type Universe in Fidelity Framework

• Native Type Universe is a unified type-theoretic substrate integrating algebraic dualities by natively supporting negative and fractional types within the Fidelity Framework.
• It preserves compilation-persistent type-level invariants, ensuring that structural metadata survives elaboration, lowering, and cross-hardware mapping.
• NTU leverages formal dualities to enable reversible computations and constraint discharge, with applications in Bayesian inference, quantum simulation, and adiabatic computation.

Searching arXiv for the cited NTU-related papers and adjacent universe/type-theory work. arXiv search query: "Native Type Universe" OR "Negative and Fractional Types in the Fidelity Framework" Native Type Universe (NTU) denotes, in the Fidelity Framework literature, a unified type-theoretic substrate for compilation, verification, and hardware targeting. In "Negative and Fractional Types in the Fidelity Framework" the term is used for the framework’s core substrate, and the paper’s central NTU-specific claim is that this substrate already has the algebraic and compilation structure to admit negative types and fractional types as native, first-class constructs (Haynes, 3 Jun 2026). In that usage, NTU is not merely a front-end typing convention: it is a semantic and compilation-persistent layer in which type-level structure survives elaboration and lowering, and is intended to remain available for representation choice, memory placement, constraint discharge, and hardware routing.

1. Provenance and core role in the Fidelity Framework

The paper presents the Native Type Universe as the framework’s unified type substrate and treats five earlier components as background: DTS+DMM, PHG, ADM, DBC, and FPS. According to the paper’s provenance summary, these establish dimensional typing plus deterministic memory management as a base substrate, extend the semantic representation from a binary graph to a Program Hypergraph, carry the discipline into training-time gradient flow, develop decidability and verification tiers, and ground the fixed-point combinator that drives nanopass lowering and proof-preserving compilation (Haynes, 3 Jun 2026).

Within that architecture, NTU has three simultaneous roles. First, it is a type algebra used by inference. Second, it is a semantic universe of structural commitments that survive compilation, rather than being erased after type checking. Third, it is a compilation target substrate meant to support cross-hardware lowering. The paper repeatedly emphasizes that algebraic type information is preserved through elaboration and lowering as compilation metadata or codata in the Program Semantic Graph and Program Hypergraph. Negative and fractional types are therefore proposed not as isolated surface-language features, but as structural objects that Baker elaboration records and that later passes are expected to preserve.

This placement distinguishes NTU from ordinary Hindley–Milner extensions that stop at static checking. The stated design goal is to preserve optionality across compute substrates and verification disciplines, with the resulting disciplines persisting through each refinement stage and informing later decisions. A plausible implication is that NTU is intended as an end-to-end invariant carrier rather than a phase-local typing mechanism.

2. Algebraic and categorical basis

The direct conceptual sources for the NTU extension are James and Sabry’s two dualities of computation and Chen and Sabry’s later categorical interpretation in compact closed categories. The paper presents two orthogonal dualities. The first is additive duality, expressed by a negative type $-T$ satisfying

$T + (-T) \leftrightarrow 0.$

The second is multiplicative duality, expressed by a fractional type $1/T$ satisfying

$T \times (1/T) \leftrightarrow 1.$

A value of type $-T$ is described as a value of the same kind as a positive $T$ value, but flowing in the reverse direction of evaluation; a value of type $1/T$ is described as a constraint on context, representing a logic variable whose value is determined by unification at a corresponding annihilation site (Haynes, 3 Jun 2026).

The paper argues that these constructions fit NTU because the framework already adopts a generalized abelian-group discipline. Kennedy’s units-of-measure work is cited as establishing that dimensions form a finitely generated free abelian group and that unification can be handled by Gaussian elimination over integers within Hindley–Milner inference. NTU then generalizes that pattern across multiple disciplines, and the paper explicitly claims that dimensional algebra, memory and coeffect disciplines, capability coeffects, schema disciplines, and grade disciplines share the properties of being decidable in polynomial time, finitely generated, and closed under inversion. On that basis, the negative type constructor is said to introduce an additive inverse element in the type-level algebra, and the fractional type constructor a multiplicative inverse element.

The categorical claim is that these additions would promote the framework’s PHG and PSG semantics from symmetric monoidal to compact closed structure. The paper gives the corresponding $\eta$ and $\varepsilon$ maps: $0 \to T + (-T), \qquad T + (-T) \to 0,$ and

$T + (-T) \leftrightarrow 0.$0

It also invokes the compact-closed “zigzag” or “snake” pattern as the visual signature of the negative-type discipline. The text does not settle a single operational reading of multiplicative annihilation: it notes both the James–Sabry interpretation as a unification site and the Chen–Sabry interpretation as a structural operation that may reclaim storage or signal an exception on mismatch.

A further bridge to existing NTU machinery is dimensional behavior. The paper states that negative types inherit the same dimension as the corresponding positive type, whereas fractional types invert the dimension. Its examples therefore distinguish, for instance, Neg<float<N>> from Recip<float<N>>, and explicitly connect reciprocal forms such as conductance to inverse dimensions.

3. Formal constructs proposed inside NTU

The paper proposes extending the framework’s Hindley–Milner style with directional judgments: $T + (-T) \leftrightarrow 0.$1 The intended reading is that the forward judgment produces a $T + (-T) \leftrightarrow 0.$2, while the backward judgment demands a $T + (-T) \leftrightarrow 0.$3. This directional layer is the basis for the proposed negative-type rules. In the sketch given by the paper, negative introduction and elimination are

$T + (-T) \leftrightarrow 0.$4

Additive annihilation is written as

$T + (-T) \leftrightarrow 0.$5

For multiplicative duality, the paper proposes pair introduction and annihilation: $T + (-T) \leftrightarrow 0.$6 The notional Clef primitives are correspondingly given as eta_plus, epsilon_plus, eta_times, and epsilon_times. The paper stresses that these are not ordinary runtime functions, but type-level structural transitions recognized by Baker elaboration and carried as codata into the PSG.

At the source-language level, the proposed notation includes both nominal and symbolic forms, such as Neg<'T> or -'T, and Recip<'T> or 1/'T. The paper’s examples include reverse physical quantities and reciprocal dimensional forms. It also states that the resulting isomorphisms would admit concise forms of resolution within the lowering strategy, because cancellation and annihilation sites become explicit structural markers.

4. Compilation persistence and lowering strategy

A distinctive feature of the NTU proposal is that the new type forms are tied directly to the framework’s lowering pipeline. Negative and fractional constructors are to be recognized at source level and elaborated by Baker into the Program Semantic Graph or Program Hypergraph as codata. The fixed-point nanopass lowering discipline is then expected to preserve those commitments rather than rediscovering them later (Haynes, 3 Jun 2026).

The paper sketches differentiated lowering targets for the two dualities. Negative types are associated with rewrite-rule and interaction-net style machinery, especially Coll’s Inet dialect. Fractional types are associated with the SMT dialect, where $T + (-T) \leftrightarrow 0.$7 acts as a constraint discharge or unification site. It also notes that DCont may provide a continuation-oriented target where required. In this design, cancellation structure is not an implementation afterthought: it is an explicit intermediate-language resource that guides lowering toward the proper operational carrier.

This architecture is also used to justify the claim that the proposed extensions are compatible with decidability and principal types. The paper does not offer a complete proof, but argues that Kennedy’s abelian-group pattern already underwrites decidable inference and principal typing for dimensional algebra, and that NTU has generalized the same pattern across several disciplines. The intended conclusion is that additive and multiplicative inverses can be incorporated by the same style of algebraic unification machinery. The text further proposes extending dimensional algebra to rational dimensions, specifically to support reciprocal and fractional forms.

5. Domain interpretations

The paper motivates NTU’s extension through several application domains, while presenting these sections as sketches rather than completed formal developments. In Bayesian inference, a value of type Recip<Evidence> is proposed as an unsatisfied conditioning obligation. The illustrative model includes a posterior_demand : Recip<'D>, and conditioning is represented by an epsilon_times discharge. In this reading, multiplicative duality expresses evidence demand and later fulfillment (Haynes, 3 Jun 2026).

In quantum computation and simulation, negative types are proposed as a type-level adjoint. The paper’s example packages a unitary gate with both a forward map and an adjoint of type Neg<'Out> -> Neg<'In>. The stated goal is to preserve adjointability or unitarity as a structural invariant of compilation rather than as a programmer assertion. The paper also suggests that fractional types could represent measurement obligations.

For adiabatic computation, the proposal combines both dualities. Negative types represent reversible Hamiltonian deformation, while fractional types represent a ground-state constraint that must be maintained and later discharged. The example AdiabaticStep<'Hamiltonian> includes both an adjoint on negative Hamiltonians and a ground_state_constraint : Recip<GroundState>. The paper is explicit that this section is speculative and framed as a proposal.

Across these domains, the common theme is that NTU is meant to host unconventional computational disciplines natively while retaining the framework’s compilation-persistent invariants. A plausible implication is that NTU is being positioned less as a language-specific universe in the Martin-Löf sense and more as a broad structural substrate for reversible flow, constraint propagation, and cross-substrate lowering.

6. Relation to adjacent research, acronym ambiguity, and limitations

The phrase “Native Type Universe” is specific to the Fidelity Framework paper, but several adjacent arXiv works bear directly on the broader question of native universes and native type structure. "Type Theory with Explicit Universe Polymorphism" develops a theory in which universe levels and constraints are part of the core judgmental structure, with products indexed by levels and constraints; for NTU design, that work is relevant because it treats universes as a native syntactic and judgmental layer rather than as purely meta-level machinery (Bezem et al., 2022). "Yet another cubical type theory, but via a semantic approach" does not use the term NTU, but it gives a universe-centered semantic presentation with an ambient universe of fibrations, an internal universe of small fibrations, explicit decoding $T + (-T) \leftrightarrow 0.$8, closure under $T + (-T) \leftrightarrow 0.$9, and heterogeneous $1/T$0, making it strongly related semantically to native-universe architectures (Kapulkin et al., 19 Dec 2025). "Native Type Theory" uses “native type system” in a different but related categorical sense: the native type system of a $1/T$1-theory $1/T$2 is presented as the internal language of the presheaf topos $1/T$3, deriving type constructors from term constructors plus logic and dependent type structure (Williams et al., 2021).

The acronym NTU is also overloaded in unrelated arXiv literatures. In fault-tolerant quantum computing, NTU denotes neural transfer unification, a transfer-learning framework for foundation decoders across code distances (Yan et al., 25 Jun 2026). In cooperative game theory for multi-winner voting, NTU denotes non-transferable utility (Chen et al., 21 May 2026). Those usages are terminologically independent of the Native Type Universe.

The Fidelity Framework paper is careful about its own limits. It does not provide a full proof that Hindley–Milner plus NTU plus negative and fractional constructors preserves principal types. It does not fully specify the algebra of rational exponents across all NTU disciplines. It does not choose definitively between alternative operational semantics for fractional closure. Several lowering strategies are presented as working hypotheses that require empirical validation. The paper also leaves open whether duality is best represented as a fourth axis of the cell complex or as its own sheaf over the compilation poset, and notes that a new MLIR dialect may be required if existing Inet or SMT vocabularies prove insufficient.

Taken together, these constraints make the current NTU proposal architectural and programmatic rather than a finished metatheory. Its strongest claim is not that every formal detail has been settled, but that the existing NTU substrate already has the generalized algebraic discipline, compilation persistence, and categorical headroom needed to host compact-closed duality as a native part of the framework (Haynes, 3 Jun 2026).