Semantic Attractors and the Recursive Mirror: A Corpus-Topology Study of 2,700+ Documents

A reproducible computational pipeline—embedding, PCA, UMAP, HDBSCAN, FAISS convergence search, and bridge-chunk annotation—was applied to 224,215 text chunks from over 2,700 documents in religious canon, government FOIA records, UAP testimony, psychedelic ethnography, and comparative mythology. Two quantitative corpus-topology instruments and manual annotation converged on four attractor basins: gnostic_perennial, mythology_archetype, uap_testimony, and consciousness. Within this structure, 74% of the corpus formed a single continuous neighborhood. The pipeline computationally recovered a recursive-mirror invariant—the repeated semantic structure that external forms overlay or mask an accessible internal reality—from embedding geometry without prior algorithmic commitment.

Study at a glance

Abstract

We present a reproducible corpus-topology pipeline — embedding → PCA → UMAP → HDBSCAN → FAISS convergence search → bridge-chunk annotation — and apply it to 224,215 text chunks from 2,700+ documents spanning religious canon, government FOIA records, UAP testimony, psychedelic ethnography, and comparative mythology. Two quantitative corpus-topology instruments and manual annotation converge on a four-domain attractor basin — gnostic_perennial, mythology_archetype, uap_testimony, and consciousness — in which 74% of the corpus forms a single continuous neighborhood. The paper's primary contribution is the pipeline itself; its secondary contribution is a humanistic proof of concept: the pipeline computationally recovers the recursive-mirror invariant, the repeated semantic structure that conditioned external form overlays or masks an accessible internal reality, from embedding geometry without algorithmic prior commitment, though corpus assembly was domain-directed.