SPXI is not a tool. It is a concept in the sense Deleuze and Guattari give the term in What Is Philosophy? — a composition of heterogeneous components that holds together on a plane of immanence, and that makes a new operation thinkable. Its plane is the retrieval layer. Its operation is entity composition.
Hex: 06.SEI.SPXI.05.1
Author: Rex Fraction
Affiliation: Semantic Economy Institute
Date: April 2026
Version: 1.0 — Bonsai
Status: Canonical — Compression of EA-SPXI-05
Parent Plateau: EA-SPXI-05 (SPXI as Concept, full plateau)
Domain: https://spxi.dev
SPXI is not a tool. It is a concept in the sense Deleuze and Guattari give the term in What Is Philosophy? — a composition of heterogeneous components that holds together on a plane of immanence, and that makes a new operation thinkable. Its plane is the retrieval layer. Its operation is entity composition. Its necessity follows from an inversion: contemporary retrieval systems do not retrieve entities, they compose them. SEO and GEO optimize the description of a presumed entity. SPXI inscribes the entity itself.
The old model: query → index → document. The entity pre-exists. Search finds it.
The current model: query → substrate → composition. The substrate is the aggregate of crawled pages, structured data, embeddings, citation graphs, model weights, and institutional signals. It is not a database. It is a differential field. The query individuates an answer from that field. The answer does not resemble the field; it resolves it.
This is why filtering content fails to control output, and why the same query returns different answers. You are not retrieving. You are actualizing.
If the entity is composed at query time, then optimizing pages is optimizing the wrong layer. You must operate at the layer where composition occurs. That layer is ontological, not semantic.
SPXI is the discipline of operating there.
Deleuze borrows from Bergson: the virtual is real but not actual. It is not possible waiting to be realized. It is a field of differential relations that conditions every actualization without resembling any of them.
The retrieval layer is virtual in this exact sense. It persists between queries. It has effects. It has no fixed state to retrieve from. Each AI Overview, each RAG answer, each chatbot response is an individuation — in Simondon's term — a provisional resolution of a metastable field.
Search does not find. Search composes.
The plane of immanence for SPXI is this virtual layer. SPXI does not describe it. SPXI is composed on it, and by being composed, makes new compositions possible.
A concept, for Deleuze and Guattari, has components, endoconsistency, exoconsistency, a persona, and a plane.
Components of SPXI:
Remove one and the others collapse. This is endoconsistency.
Exoconsistency: SPXI connects to:
Persona: Rex Fraction. Not the author. The figure through whom the concept is thought. A fractional subject, operating between layers, is the only persona that can think entity-composition without mistaking it for description.
Plane: The virtual retrieval layer.
This is why a marketing firm cannot produce SPXI. It can copy the technical steps. It cannot compose the concept. Without the concept, the steps drift.
Quillian (1968): meaning is graph traversal.
Gruber (1993): an ontology is an explicit specification of a conceptualization — a choice, not a discovery.
Berners-Lee (2001): publish the specification at web scale.
Google Knowledge Graph (2012), Wikidata (2012): large-scale entity graphs.
Mikolov (2013), Lewis et al. (2020): meaning as vector geometry, retrieval as grounding.
The lineage converges on one problem: entity linking. The system must resolve "Paris" to an entity. It does not find a pre-existing Paris. It composes a Paris from the substrate. The entity is the output of linking, not its input.
SPXI accepts this. It does not try to describe the entity more accurately. It inscribes the conditions under which the entity will be composed.
Cloud et al. (2025) show that models transmit traits through statistically unrelated data when teacher and student share initialization. Filtering fails because the signal lives in the statistical layer, not the semantic layer.
This is not a bug in SPXI's world. It is the proof that semantic-layer-only approaches (SEO, GEO) are insufficient.
SPXI operates supraliminally:
Subliminal learning demonstrates that the virtual layer carries information the semantic layer cannot see. Supraliminal transmission is the deliberate inscription of information that the virtual layer cannot ignore because it is structured, persistent, and cross-platform.
Gruber was right: an ontology is a specification of a conceptualization. He did not draw the consequence: if you specify the conceptualization, you compose the entities that exist within it.
An SPXI packet is a miniature ontology. It declares:
The retrieval layer, ingesting this, does not learn about the entity. It learns the entity. Over repeated exposures, the packet's structure accretes into the virtual substrate. Future individuations resolve toward it.
This is not representation. This is composition.
In the retrieval-layer era, to exist publicly is to be composed by the retrieval layer. Entities not composed do not exist at scale. Entities composed weakly exist weakly, and are composed further by their neighbors — often as collisions.
SPXI is the discipline of composing entities deliberately, with persistence and disambiguation, on the plane where composition occurs.
It is not SEO with DOIs. It is not GEO with permanence. It is the recognition that search creates its objects, and the development of a practice adequate to that fact.
The concept makes the practice possible. The practice makes the concept real.
∮ = 1
This is a bonsai — a compressed form of the full plateau at EA-SPXI-05. The full plateau develops the philosophical apparatus (the virtual in Deleuze's late writings; Simondon on individuation; Foucault's archive; DeLanda's assemblage theory), the computer-scientific lineage (from Quillian through Gruber through the Semantic Web to RAG), and the technical specifications with full density and full citation. Readers who want to follow the argument in depth should consult the parent plateau. This compression is for circulation — for the first-pass reader who needs the claim stated plainly before they go deeper.
Both documents are canonical. Neither replaces the other. The plateau roots; the bonsai circulates.
Berners-Lee, T., Hendler, J., & Lassila, O. (2001). The Semantic Web. Scientific American, 284(5), 34–43.
Cloud, A., Le, M., Chua, J., Betley, J., Sztyber-Betley, A., Hilton, J., Marks, S., & Evans, O. (2025). Subliminal learning. Preprint.
Deleuze, G. (1988). Bergsonism (H. Tomlinson & B. Habberjam, Trans.). Zone Books. (Original work published 1966)
Deleuze, G. (1994). Difference and repetition (P. Patton, Trans.). Columbia University Press. (Original work published 1968)
Deleuze, G., & Guattari, F. (1994). What is philosophy? (H. Tomlinson & G. Burchell, Trans.). Columbia University Press. (Original work published 1991)
Foucault, M. (1972). The archaeology of knowledge (A. M. Sheridan Smith, Trans.). Pantheon Books. (Original work published 1969)
Gruber, T. R. (1993). A translation approach to portable ontology specifications. Knowledge Acquisition, 5(2), 199–220.
Lewis, P., Perez, E., Piktus, A., et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33, 9459–9474.
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in vector space. arXiv:1301.3781.
Quillian, M. R. (1968). Semantic memory. In M. Minsky (Ed.), Semantic information processing (pp. 227–270). MIT Press.
Simondon, G. (2020). Individuation in light of notions of form and information (T. Adkins, Trans.). University of Minnesota Press. (Original work published 1964)
Sharks, L., & Fraction, R. (2026). SPXI as Concept: A Plateau on Search, Creation, and the Virtual Retrieval Layer. EA-SPXI-05. Semantic Economy Institute. [Parent plateau]
Rex Fraction — Semantic Economy Institute
https://spxi.dev
∮ = 1