---
title: 'Pure Data Spaces: Theory and Practice'
url: https://www.emergentmind.com/topics/pure-data-spaces
type: topic
---

# Pure Data Spaces: Theory and Practice

Pure data spaces are formal constructs that appear in several technically distinct research programs. In one line of work, the only primitive is the finite sequence, objects with mathematical meaning are data, and collections of mathematical objects are associative data called spaces [2508.14271]. In another, the “pure data space” is the finite topological space \(D=(X,\mathcal T)\) on a data set \(X\), prior to interpreting variables, functions, and relations in a domain [1411.1497]. In decentralized data-sharing architectures, a Pure Data Space (PDS) is an idealized, minimal instance of a data-space architecture in which all data stay at their original sources, integration is performed purely at the semantic level, and the soft infrastructure is completely decoupled from any centralized storage or compute platform [2509.06983].

## 1. Terminological scope

The phrase “pure data spaces” is used for at least three non-identical notions in the cited literature. In the foundational program, a space is the basic collection analogous to sets in Set Theory or objects in Category Theory, and spaces are studied via their semiring of endomorphisms [2508.14271]. In the data-information-knowledge program, a data space is a finite topological space whose underlying set of points is a given data set \(X\), with further informational and knowledge-bearing structure arising only after interpretation in a specific domain \(\Delta\) [1411.1497]. In the architectural program, a Pure Data Space is a decentralized semantic data-space architecture specified by open standards, semantic contracts, and machine-readable policies rather than centralized storage or monolithic middleware [2509.06983].

A common source of ambiguity is that all three usages retain the word “space” while assigning it different formal roles. One usage is algebraic and foundational, one topological and pre-semantic, and one infrastructural and interoperability-oriented. This suggests a shared methodological pattern: each formulation begins from a deliberately minimal substrate and postpones richer semantic, inferential, or operational structure until later stages.

## 2. Pure data as finite-sequence substrate

In the foundational framework, the only primitive is the notion of a finite sequence. A **coda** is a pair of data, written \((A:B)\), and a **data** is any finite sequence of codas, including the empty sequence \(()\). Two operations are primitive: concatenation \((AB)\) and pairing as a coda \((A:B)\) [2508.14271].

The earlier axiomatization states the same substrate as follows: Data is the smallest set such that \(()\in\mathrm{Data}\), \(A,B\in\mathrm{Data}\Rightarrow AB\in\mathrm{Data}\), and \(A,B\in\mathrm{Data}\Rightarrow A:B\in\mathrm{Data}\). The entire system is governed by a single **Axiom of Definition**, where a context \(\delta\) is a partial function from codas to Data, extended recursively by
\[
\delta(AB)=\delta(A)B=A\delta(B),\qquad
\delta(A:B)=\delta(A):B=A:\delta(B).
\]
The empty context \(\delta_0\) is valid, and if \(\delta\) is valid and a new definition \(d\) has disjoint domain, then \(\delta\cup d\) is also valid [2305.05507].

Mathematical meaning is introduced by choosing such a partial function \(\delta\) and declaring \(c\sim\delta(c)\) on its domain of definition, then closing under the congruence rules for concatenation and colon [2508.14271]. A coda \(a\) is a fixed point if \(\delta(a)=a\); any data reducible to a fixed point is an atom; and a data \(A\) is invariant if every coda in \(A\) is a fixed point and its left- and right-components are themselves invariant [2508.14271]. The system begins from the empty context, and a minimal consistency argument forces the introduction of one distinguished atom `(:)` with
\[
\delta((:)\!:X)=( :\!:X)\quad\forall X,
\]
after which a small library of combinator definitions is added, but no arithmetic axioms are assumed [2508.14271].

The same framework defines an internal three-valued logic. In any valid context, \(A\) is **true** iff \(A=()\), **false** iff \(A\) is atomic, and **undecided** iff \(A\neq()\) and \(A\) is not atomic [2305.05507]. This internal logic is not primitive; it arises from the permanent dichotomy between empty and atomic data.

## 3. Spaces, morphisms, and the semiring of endomorphisms

A data \(S\) is a **space** if
\[
(S:XY)=(S:(S:X)(S:Y))\quad\forall X,Y,
\]
equivalently, if the binary operator
\[
X\ast_S Y := (S:XY)
\]
is associative [2508.14271]. In particular,
\[
(S:S:X)=(S:X),\qquad (S:X:S)=(S:X),
\]
so \(S\) is idempotent in the global sense, and the data \((S:X)\) are precisely its fixed points under colon [2508.14271]. The earlier formulation writes the same law as
\[
S:(S:X)(S:Y)=S:XY
\]
and calls \(S:()\) the neutral element of \(S\) [2305.05507].

Morphisms are also defined internally. One exposition defines a morphism \(F\colon A\to B\) as a distributive data satisfying
\[
F:A:X = B:F:X\quad\forall X,
\]
with composition \(G\circ F:=G:F\) and identity given by \(\mathsf{pass}\) [2305.05507]. Another defines a morphism \(S\overset F\longrightarrow T\) as any product \(F\) with \(F=T\cdot F\cdot S\), so that an endomorphism of \(S\) is any \(f\) with
\[
f=S\cdot f\cdot S.
\]
Composition is pure-data concatenation [2508.14271]. In the earlier presentation, spaces and morphisms form a “coda-category,” in which every pair of morphisms is automatically composable [2305.05507].

For a fixed space \(S\), the set \(\End(S)\) of endomorphisms carries a semiring structure. Addition is
\[
f\oplus g := S\cdot(f+g)\cdot S,
\]
where \(f+g\) is the global data sum \((X+Y)\colon Z\mapsto (Z:X)(Z:Y)\); multiplication is composition; the additive identity is \(0=S\cdot\mathrm{null}\cdot S\); and the multiplicative identity is \(1=S\cdot\mathrm{pass}\cdot S\) [2508.14271]. Within this semiring, several classes are distinguished:

- **Subspaces**: idempotent endomorphisms \(e\) with \(e\cdot e=e\), ordered by \(e\le_i f\iff e\cdot f=e\).
- **Homomorphisms**: endomorphisms \(h\) satisfying
  \[
  h\cdot(f\oplus g)=(h\cdot f)\oplus(h\cdot g),\qquad
  (f\oplus g)\cdot h=(f\cdot h)\oplus(g\cdot h).
  \]
- **Central endomorphisms**: endomorphisms commuting with every unit in the group of units.
- **Units**: invertible elements under composition.

The semiring viewpoint supports several structural results: isomorphic spaces have isomorphic semirings; every morphism \(h\colon S\to T\) factors uniquely as \(h=m\cdot e\) with \(m\) a monomorphism and \(e\) idempotent; and a space \(S\) is a field if and only if every non-constant homomorphism of \(S\) is a unit [2508.14271].

## 4. Organic emergence of classical mathematical structures

One of the main claims of the foundational literature is that familiar objects from classical mathematics emerge from pure data spaces “grown organically” from the substrate of pure data with minimal combinatoric definitions, including natural numbers, integers, rational numbers, boolean spaces, matrix algebras, Gaussian Integers, Quaternions, and non-associative algebras like the Integer Octonions [2508.14271].

| Structure | Pure-data realization | Stated outcome |
|---|---|---|
| Natural numbers | \(N=\mathrm{is}\;a\) | \(\End(N)\) is the standard semiring \((\mathbb N,+,\times)\) |
| Integers | subspace **reduce** of \(N_2=\mathrm{is}\;a\;b\) | central semialgebra isomorphic to \(\mathbb Z\) |
| Matrix algebra | subspace **sort** of \(N_2\) | realizes \(\Mat_{2\times2}(\mathbb N)\) |
| Boolean space | \(\mathbb L_2=\mathrm{bool}\) | \(\End(\mathbb L_2)\) has four elements: \(\mathrm{ID},\mathrm{TRUE},\mathrm{FALSE},\mathrm{NOT}\) |
| Positive rationals | \(Q^+=\mathrm{ADD}\cdot\mathrm{NORM}\cdot\mathrm{NORM0}\) | field isomorphic to \(\mathbb Q^+\) |
| Sequences | \(\Seq(s:S):=\mathrm{ap}\,(\mathrm{put}\;s\cdot S\cdot\mathrm{get}\,s)\) | space of finite \(S\)-valued sequences |

For the natural numbers, the fixed points of \(N=\mathrm{is}\;a\) are \(\{a^n:n\in\mathbb N\}\), the operation
\[
(N:XY)=a^{\,\bigl(n(X)+n(Y)\bigr)}
\]
gives ordinary addition, and homomorphisms \(h\colon N\to N\) are exactly multiplication by a fixed \(k\in\mathbb N\), namely
\[
(h:a^n)=a^{\,k\,n}.
\]
The non-constant idempotents are the endomorphisms
\[
\mathrm{rem}(p,q)\colon n\mapsto \min(n\bmod p,\;q-1),\qquad p,q\ge 1,
\]
and fields occur precisely when \(q=p\) is prime, yielding the prime fields \(\mathbf F_p\) [2508.14271].

For integers, the fixed points of **reduce** are identified with \(\mathbb Z\) via \(a^n\mapsto +n\) and \(b^n\mapsto -n\). For the **sort** subspace, fixed points are \(\{a^m b^n:m,n\ge 0\}\cong\mathbb N^2\), and a homomorphism is determined by a \(2\times2\) matrix over \(\mathbb N\) through
\[
h(a^m b^n)=a^{\,m_{11}m+m_{12}n}\,b^{\,m_{21}m+m_{22}n}.
\]
In the four-atom space \(N_4=\mathrm{is}\;a\;b\;c\;d\), suitable sorting and cancellation recover a copy of \(\mathbb Z[i]\), while further increases in the number of atoms and suitable involutive cancellations produce the integer quaternions and the integer octonions \(\mathbb O(\mathbb Z)\) [2508.14271].

The framework also realizes algebraic and combinatorial structures beyond number systems. The space \(\Seq(b:\mathbb L_2)\) is the Boolean sequence space, and the algebraic semilattice space \(\mathrm{sort}\cdot\mathrm{once}\cdot(\mathrm{is}\;a\;b\;c)\) realizes the power-set \(\mathcal P(\{a,b,c\})\) ordered by inclusion, with constants forming the union-lattice of subsets, units forming the symmetric group \(S_3\), and homomorphisms exactly the union-preserving maps [2508.14271].

## 5. The pure data space as finite topology

In the data-information-knowledge framework, a data space \(D\) on a finite nonempty data set \(X\) is a pair
\[
D=(X,\mathcal T),
\]
where \(\mathcal T\subseteq\mathcal P(X)\) satisfies \(\varnothing\in\mathcal T\), \(X\in\mathcal T\), any union of members of \(\mathcal T\) lies in \(\mathcal T\), and any finite intersection of members of \(\mathcal T\) lies in \(\mathcal T\) [1411.1497]. The family \(\mathcal T\) is the data structure of \(D\), and its members are open sets. A data function of arity \(n\) is any map \(\mathcal T^n\to\mathcal T\), while a data relation of arity \(n\) is any Boolean-valued map \(\mathcal T^n\to\{\mathrm{true},\mathrm{false}\}\) [1411.1497].

Concrete examples include the discrete topology on \(X=\{3,7,11,23\}\), the cofinite topology on the same set, and a preorder-induced data space on \(X=\{a,b,c\}\) with preorder \(a\le b\le c\), where the open sets are the upper sets and
\[
\mathcal T=\{\varnothing,\{c\},\{b,c\},\{a,b,c\}\}.
\]
For finite data spaces, the stated properties include compactness, the \(T_1\)-discrete dichotomy, a characterization of connectedness in terms of the preorder graph, and the fact that a data space is metrizable if and only if it is discrete [1411.1497].

At this level, no semantics have yet been assigned to points or open sets. The construction proceeds by interpreting variables, functions, and explicit relations over \(D\) in a specific domain \(\Delta\) to obtain an information space \(I\), and then building a knowledge space \(K\) as the product of \(K_T\) and \(K_P\). Here \(K_T\) is obtained from \(I\) by using the induction principle to generalize propositional relations to quantified relations, the deduction principle to generate new relations, and standard mechanisms to validate relations; \(K_P\) is the space of specifications of methods with operational instructions valid in \(K_T\) [1411.1497].

This formulation explicitly distinguishes the “pure data space” from later semantic enrichment. It constrains only how data points cluster or separate via \(\mathcal T\), and what syntactic constructions are available as data-functions and data-relations. Information retrieval then consists essentially in mining domain objects and relations, and knowledge discovery consists essentially in applying induction and deduction, synthesizing and modeling information, and validating the resulting propositions and specifications [1411.1497].

## 6. Pure Data Space as decentralized semantic architecture

In the systems and interoperability literature, a Pure Data Space (PDS) is an idealized, minimal instance of a data-space architecture in which all data stay at their original sources, integration is performed purely at the semantic level, and the soft infrastructure is completely decoupled from any centralized storage or compute platform [2509.06983]. Relative to a general data space, the defining negatives are explicit: no “data lake” or aggregated staging area exists, even transiently; no proprietary or monolithic middleware is assumed; and governance is enforced exclusively through machine-readable policies such as SPARQL guards and ODRL [2509.06983].

The theoretical core is expressed through data sovereignty, semantic integration, decentralization, and soft infrastructure. If \(P\) is the set of participants, \(C\) the set of semantic connectors, \(M\) the global semantic model, and \(G\) the governance layer, then
\[
\mathrm{PDS}=(P,C,M,G).
\]
For any request \(r\colon p_i\to p_j\), the permission function is
\[
\pi(p_i,p_j,r)=
\begin{cases}
\text{allow}, & \text{if } r \models \Phi_{p_i},\\
\text{deny}, & \text{otherwise,}
\end{cases}
\]
and is computed at the connector of \(p_i\) via policy evaluation. Semantic integration is driven by a shared vocabulary \(\mathcal V=\{\nu_1,\dots,\nu_n\}\), mapping each raw data item \(d\in D_p\) to a triple set \(\tau(d)\subseteq\mathcal T=\{(s,p,o)\mid s,p,o\in I\}\) [2509.06983].

The architecture is decomposed into minimal, protocol-driven microservices or linkable libraries:

| Component | Role | Key basis |
|---|---|---|
| Semantic connectors | Data-plane and control-plane | HTTP/HTTPS, WebSub, LDES, REST/OpenAPI, SPARQL Protocol |
| Federated registry & catalogs | Metadata and policy endpoints only | SPARQL endpoint for discovery |
| Identity & Access Management | Peer identity and credential checks | DIDs and Verifiable Credentials |
| Policy enforcement layer | Local authorization | ODRL profiles with SPARQL-based guard |
| Governance & certification service | Governance rules and compliance | machine-readable smart contracts, test suite |

Interoperability is organized around a lightweight core ontology \(\mathcal O_{\text{PDS}}\) for Entities, Attributes, and Provenance, extended by domain-specific ontologies. Raw schema fields are semantically mapped, for example by annotations of the form
\[
(\text{_:field}\;:\texttt{mapsTo}\;:\!recordsValue)\in\tau.
\]
Authorization is performed locally through a SPARQL-based “authorization guard”: policies are stored as named graphs \(G_{\text{policy}}\), and at request time a connector executes a SPARQL ASK query against \(G_{\text{policy}}\) plus the requested data graph; if the ASK returns true, the data transfer proceeds, otherwise it is denied [2509.06983].

The literature lists concrete deployments and pilots: Catena-X in automotive, the European Health Data Space in healthcare, Smart Freight Centre on AWS in logistics, the Flanders Smart Data Space in smart cities, and the Green Deal Data Space. Across these cases, data remain at producer sites, edge nodes, hospital EHR systems, or origin registries, while shared semantic models, RDF mappings, and SPARQL-guard policies enable federated access without centralizing raw data [2509.06983].

## 7. Comparative significance and open questions

The three usages of pure data spaces converge on a common emphasis on minimal structure, but they diverge sharply in formal objective. The foundational program seeks an axiomatic basis for mathematics and computing in which proof and computation are both sequences of equalities on pure data, and in which all familiar mathematical structures arise as associative data with semiring-organized endomorphisms [2305.05507]. The topological program isolates the raw combinatorial footprint of a data set before semantic interpretation, then derives information and knowledge spaces by interpretation, induction, deduction, synthesis, modeling, and validation [1411.1497]. The architectural program pursues secure and efficient data exchange under data sovereignty, interoperability, and trust by combining semantic connectors, ontologies, DIDs, verifiable credentials, ODRL, and SPARQL-based authorization [2509.06983].

Several open directions are explicit in the literature. In the foundational setting, a rich—but as yet only partially explored—theory of external morphisms between different spaces is anticipated; brute-force scans of data up to modest width and depth already produce small spaces, but the growth of data is doubly exponential; and whether the framework will scale to topology, category theory, analysis, or cohomology remains an open frontier [2508.14271]. In the architectural setting, observed challenges include the performance of real-time SPARQL authorization on very large RDF streams, versioning and evolution of the shared ontology \(M\), and onboarding new participants with non-standard schemas [2509.06983]. In the data-information-knowledge framework, the key observation is that efficient approaches may be designed to discover profound knowledge automatically from simple data, as demonstrated in the case of geometry [1411.1497].

Accordingly, “pure data spaces” should be interpreted contextually. In one body of work the term denotes associative data that serve as basic mathematical collections; in another it denotes the topological skeleton \(D=(X,\mathcal T)\) of a data set before semantic enrichment; and in a third it denotes a fully decentralized semantic overlay for data exchange. The shared vocabulary is real, but the mathematical objects, operational mechanisms, and intended applications are distinct.

Source: https://www.emergentmind.com/topics/pure-data-spaces