The Structural Invariant Framework developed through a sequence of investigations into invariant-preserving neural machine translation, structural diagnostics, output-level admissibility and governing-state assurance.
This page presents selected research publications and reports documenting that progression. Journal publication status, public archival release and accompanying research artifacts are distinguished explicitly.
Principal research outputs
Current publication status
SIF 1.0
Registry-based terminology preservation
Published in a peer-reviewed journal · 15 October 2026
Measurements Infrastructure, Vol. 12 (2026), article MI_096_170626. The published article records review on 16 July 2026 and publication on 15 October 2026.
Structural Invariant Framework 2.0 for Post-Generation Structural Admissibility Inspection in English-to-Ukrainian Technical Translation
Published on Zenodo · October 2026
SIF 2.0 introduces a registry-based approach to checking the structural admissibility of neural technical translations after generation. It separates translation generation from deterministic inspection of specified terminology, formulas, numerical values and units.
The published research presents experimental results, identified limitations and the verification approach implemented in SIF Inspector, a bounded research instrument for post-generation inspection.
The journal manuscript is under review; the Zenodo publication does not establish journal peer-review acceptance.
Extended research manuscript in preparation · Public release and journal submission planned for late 2026
The technical note presents research motivation and a source-governed architecture. It precedes later evaluation and does not establish independently validated general correctness. The extended manuscript is a separate planned output and is not yet publicly available.
Chronological bibliography
Selected research record
Four stages organize 14 selected Zenodo records by the research questions they document. “Publication” describes the Zenodo resource type and does not imply journal peer review.
Stage I · Foundations: structural preservation and TRIZ-aware training
2025
01
Unified Metrology–TRIZ Neural Machine Translation (UMT–NMT) v1.0: Proof-of-Concept Study
An initial English-to-Ukrainian metrology proof of concept using Corpus A, focused on preserving placeholders, formulas and constants under low-data conditions.
02
TRIZ-Aware Regularization for Invariant-Preserving Neural Machine Translation in Metrological and Scientific Domains (v1.0)
Examines TRIZ-aware regularization for retaining structural invariants in low-resource technical translation.
03
TRIZ-Aware Loss v2.0 with Casimir-Regularized Fields: Geometric Invariance for Explainable Neural Machine Translation in Metrological and Scientific Domains
Presents a physics-inspired semantic-field and geometric regularization formulation with bounded experiments; it does not establish guaranteed geometric invariance.
Stage II · Structural diagnostics and controlled regimes
2026
04
Structural Invariant Framework in Low-Data Contradiction-Aware Neural Machine Translation: Architecture and Diagnostic Pipeline (v1.0)
Reports a controlled corpus-ordering ablation: ordering-sensitive effects were observed, while contradiction magnitude was not dominant within the tested protocol.
06
Tunable Operating Regimes in Contradiction-Aware Neural Machine Translation: From Invariant Preservation to Controlled Sensitivity (Article 2)
Examines operating regimes and loss–geometry decoupling under fixed low-data conditions.
07
Contradiction-Conditioned Response Structure in Low-Resource Neural Machine Translation: Representation-Level Diagnostics, Plurality Analysis, and the Transfer Limit
Reports representation-level diagnostics and a boundary between representation differentiation and output admissibility, including a negative output-transfer finding.
Stage III · Output failures, bounded repair and SIF 1.0
2026
08
Corpus D2 — SIF v1.0 Layer 4: Output Failure Taxonomy for Anchor-Induced Fragility in Contradiction-Aware NMT
Describes class-aware post-decoding repair and decision routing; it is not a general-purpose translation model.
10
Structural Invariant Framework (SIF 1.0): A Registry-Based Framework for Empirically Verified Terminology Preservation in Neural EN→UA Translation of Metrological Standards
Documents Registry-based terminology admissibility and bounded deterministic enforcement; the frozen article-level evaluation is distinct from the public demonstration subset.
Stage IV · Research synthesis and Governing-State Assurance
2026
11
When the Loss Is Right and the System Is Wrong: Order Effects in TRIZ-Aware Neural Machine Translation — A Representational Limit of Pointwise Optimization
Presents research motivation and a source-governed architecture for bounded technical assurance; it predates later evaluation and does not establish independently validated general correctness.
Related archival outputs
Supporting artifacts
These records accompany the research chronology and are listed as datasets or models, not as peer-reviewed journal articles.
UMT-NMT v1.0: Proof-of-Concept Study on Normative Documents in Metrology (EN→UK)