About the TRACE
framework

TRACE is a cross-domain engineering framework for trustworthy agentic AI in operationally critical contexts. It defines a four-layer reference architecture (with explicit L2a / L2b split into classical ML and LLM validators), seventeen trust metrics including the Computational Parsimony Ratio (CPR), and three reference instances spanning clinical decision support, an industrial multi-domain platform, and judicial decision support.

The framework is grounded in established measurement science — GUM, VIM, ISO/IEC 17025 — and originates in over two decades of work on non-Gaussian signal processing, measurement uncertainty, and applied AI in regulated environments.

Sergii Zabolotnii — framework synthesis and metrological foundation (six design principles, trust-metric set, first formalisation of CPR, mapping to GUM / VIM / ISO/IEC 17025; clinical foundational instance; planned judicial extension). Cherkasy State Business College; State Research Institute of Armament and Military Equipment Testing and Certification, Cherkasy; Uzhhorod National University.

A. Shcherban — ideologist and lead of the industrial multi-domain direction (Instance B): methodology of industrial implementation, investigation across oil & gas sub-domains, and review of the industrial sections.

The clinical foundational instance was developed jointly with V. Holinko and A. Antonenko; the judicial extension operates within the “Legal Positions Database” modernisation programme of the Supreme Court of Ukraine, funded by Expertise France.

For full author contributions, IP status, and acknowledgementssee Authors →

The framework is documented across four interlinked papers. This site is the companion to Paper 1 — the cross-domain framework synthesis, now on arXiv (arXiv:2605.03838).

1. [paper 0]Clinical foundational
   From Black-Box Confidence to Measurable Trust in Clinical AI: A Framework for Evidence, Supervision, and Staged Autonomy
   Zabolotnii, Holinko, Antonenko · IEEE Instrumentation & Measurement Magazine — Special Issue "A Measure of Trust in Healthcare" · Under review · Sep 2026
2. [paper 1]Framework synthesis← you are here
   TRACE: A Metrologically Grounded Engineering Framework for Trustworthy Agentic AI in Operationally Critical Domains
   Zabolotnii (sole author) · arXiv preprint · Published on arXiv (May 2026) — this site is the companion
3. [paper 2]Industrial multi-domain deep-dive
   Industrial Multi-Domain Agentic Platform for Upstream Oil & Gas: A TRACE Instance
   Shcherban (lead), Zabolotnii · Scopus-indexed industrial-AI journal · Planned · Q3 2026
4. [paper 3]Metrological deep-dive
   Layer-wise GUM Propagation in TRACE: A Formal Uncertainty Budget for Agentic AI Systems
   Zabolotnii (lead), Shcherban · IEEE Transactions on Instrumentation and Measurement · Planned (optional)

The TRACE framework methodology is released under Creative Commons Attribution 4.0 International (CC-BY 4.0). Implementation code, when published, will be released under Apache License 2.0. The industrial multi-domain platform (Instance B) carries an independent IP status — see the Authors page for the full notice. This website collects no personal data, sets no cookies, and uses no analytics.

Portions of this site and the surrounding research work are prepared in collaboration with an AI research assistant ("Ayona"). The name is not accidental — Greek αἰών ("duration, epoch") resonates with TRACE as a persistent trace: a record that endures and can be verified, rather than a stream that passes by. This use is disclosed; no content on this site appears without author review.