Publication registry / provenanceRecords, corrections,
and archival continuity.
This registry distinguishes current research records from historical titles, corrected interpretations, access-history records, and unresolved deposit identities.
ZenodoOSFCorrections preserved
Registry ruleFor current scientific interpretation, VARZIN prioritizes the latest relevant public Zenodo manuscript and its version-matched reproducibility artifacts. Earlier version-specific DOI records and historical titles remain visible for citation continuity and provenance, but they do not override later evidence for the same claim.
01 / current recordsPublic research records.
Seven public Zenodo records are listed by version-specific DOI. Each VARZIN record page now exposes the research question, tested methods, strongest supported findings, negative results, limitations, full-text artifact, verified code/data package where present, checksums, and scholarly machine-readable metadata. A repository deposit is citable, but is not by itself a claim of peer review or independent validation.
01Morphemes vs. Manifolds: Diagnosing Structural Blindness and Morphological Hijacking in Large Language Models via Finite Affine Orbits
NIROUYAR, R. (2026). Morphemes vs. Manifolds: Diagnosing Structural Blindness and Morphological Hijacking in Large Language Models via Finite Affine Orbits (Version v1.0.0). Zenodo.
Direct frozen-model audit: Mistral 7B ARI −0.0337 ± 0.039 and Llama 3 8B ARI 0.0152 ± 0.028 against the designed orbit labels; the axes were not recovered under the tested protocol.
10.5281/zenodo.22101179 ↗ VARZIN record →VARZIN V2 — Recoverable but Not Composable: Representation Recovery and Systematic Composition in a Frozen Language Model
Nirouyar, R. (2026). VARZIN V2 — Recoverable but Not Composable: Representation Recovery and Systematic Composition in a Frozen Language Model (Version V1.0). Zenodo.
Frozen Qwen structural information is strongly linearly decodable (mean BA 0.9666), while the preregistered composition phase reports 0/58 passes; recovery did not establish systematic composition in this experiment.
10.5281/zenodo.22679978 ↗ VARZIN record →LUXVAR: A Constructed Language with Distinctive Morphemic Structure and AI-Recoverable Morpheme Families
Reza, N. (2026). LUXVAR: A Constructed Language with Distinctive Morphemic Structure and AI-Recoverable Morpheme Families (Version v2.2). Zenodo.
Morphology and phonotactics remain the supported signal; semantic-axis tests are negative, and the manuscript addendum explicitly rejects the historical “AI-Recoverable” axis interpretation.
10.5281/zenodo.22115483 ↗ VARZIN record →VARZIN Level-1 Computational Stack: Finite Affine Core, Genomic Scanner, and Torus Manifold Solver
Nirouyar, R. (2026). VARZIN Level-1 Computational Stack: Finite Affine Core, Genomic Scanner, and Torus Manifold Solver (Version v2.0.0) [Computer software]. Zenodo.
Open software + companion monograph: finite-affine core, torus solver, deterministic genomic scanner, tectonic null-model audit, and versioned reproducibility files.
10.5281/zenodo.22036769 ↗ VARZIN record → Companion monograph →Morphological Hijacking in Frozen Language Models: A Contrastive, Symmetry-Regularized Projection Head for Algebraic Structure Recovery
Nirouyar, R. (2026). Morphological Hijacking in Frozen Language Models: A Contrastive, Symmetry-Regularized Projection Head for Algebraic Structure Recovery (Version v1). Zenodo.
Projection-head intervention record; explicit intervention is distinct from spontaneous recovery in an unmodified frozen model.
10.5281/zenodo.22258644 ↗ VARZIN record →Morphological Hijacking in Frozen Language Models: A Contrastive, Symmetry-Regularized Projection Head for Algebraic Structure Recovery
Nirouyar, R. (2026). Morphological Hijacking in Frozen Language Models: A Contrastive, Symmetry-Regularized Projection Head for Algebraic Structure Recovery (Version v2). Zenodo.
Second public version of the projection-head research record; preserve version-specific DOI citation.
10.5281/zenodo.22262388 ↗ VARZIN record →Morphological Hijacking in Frozen Language Models: A Contrastive, Symmetry-Regularized Projection Head for Algebraic Structure Recovery
Nirouyar, R. (2026). Morphological Hijacking in Frozen Language Models: A Contrastive, Symmetry-Regularized Projection Head for Algebraic Structure Recovery (Version v3). Zenodo.
v3 adds the harder composition test: a seen-pair MLP fit ≈0.996, but held-out TRUE accuracy 0.106 falls below SHUFFLED 0.281 and WRONG_OP 0.175.
10.5281/zenodo.22287006 ↗ VARZIN record →OSF research record
Maintained as a separate research record. Protocol identity relationships among historical OSF entries remain subject to the cautions below.
10.17605/OSF.IO/YEGJ8 ↗ Version provenance: the three “Morphological Hijacking in Frozen Language Models” deposits are retained as separate v1, v2 and v3 records because each has its own DOI. The current evidence hierarchy still distinguishes direct frozen-model tests from later projection/training interventions.