WisdomTwin Research develops testable frameworks, protocols and evidence concerning governed, role-specific Judgment Twins. Working papers are published open access, label every claim by status, and are revised as evidence accumulates.
Working papers in this program are preprints. They are not peer reviewed and report proposed outcomes require empirical validation.
One working paper and one technical whitepaper are available on this site. Each entry lists its record: identifier, version, date, status, license, author, abstract and citation formats.
Roman Bodnarchuk, Founder and CEO, WisdomTwin
The engines made the answer free. WisdomTwin makes it allowed. Why permission, not inference, is the bottleneck in regulated enterprise decisions.
Roman Bodnarchuk, Co-Founder and CEO, WisdomTwin, Inc..
Technical whitepaper. Proposed architecture. Not peer reviewed. Product outcomes require empirical validation. No DOI assigned.
Enterprise AI can retrieve information and execute work without establishing who is authorized to decide, which evidence remains valid, or how an exception should be reviewed. This whitepaper asks what additional architecture is needed to make expert decision context available while preserving organizational control. It defines a Sovereign Judgment Twin as a proposed decision-support system that combines permission-aware evidence retrieval, explicit expert-informed frameworks, enforceable authority boundaries, human disposition, and a versioned decision record. The contribution is an implementation-neutral reference architecture, a minimum decision-record schema, a sample risk-exception walkthrough, and a comparative evaluation design. The method is a targeted synthesis of primary technical sources followed by design analysis; it is not a systematic review or an empirical study. The paper treats sovereignty as a set of verifiable controls across processing, access, retention, operations, and exit. It makes no claim that a twin reproduces a person's mind or that agentic systems cannot implement the same controls. The proposed value is conditional: reduce avoidable decision delay while maintaining decision quality and accounting for review, rework, and operating effort. ChatGPT assisted with research, drafting, and document production.
Roman Bodnarchuk, Co-Founder and Chief Executive Officer, WisdomTwin, Inc. Toronto, Ontario, Canada. ORCID
Organizations can possess the information required for a decision while still waiting for a particular role to attend to it. This conceptual working paper defines judgment latency as the elapsed time between decision readiness and attention by the person or forum whose judgment is required. It proposes the AI Judgment Twin as a role-specific, evidence-grounded decision-support architecture, distinguishes the concept from personal simulation, and derives governance requirements for provenance, authorization, safe decline, human confirmation, monitoring, and contestability. The paper integrates research on work interruption, meetings, decision rights, organizational memory, digital and human twins, and human reliance on automation. It reproduces the assumptions behind McKinsey's 2019 estimate of about 530,000 manager-days and USD 250 million in annual wages, labels the “about one day per week” restatement as author-derived arithmetic, introduces a non-overlapping model of potentially recovered time, and proposes six falsifiable propositions and a staged empirical evaluation protocol. The proposed outcomes require empirical validation.
Bodnarchuk, R. (2026). Judgment latency in regulated enterprises: A conceptual framework and research agenda for role-specific AI decision support (WisdomTwin Working Paper WT-100-001, Version 2.0). WisdomTwin, Inc.
Working paper / preprint. Not peer reviewed. DOI: pending repository deposit. Proposed outcomes require empirical validation.
The program exists to make the judgment-latency thesis testable, not to manufacture validation.
Every statement is labeled as established evidence, author-derived arithmetic, conceptual proposition, modeled scenario or design requirement. Nothing is presented as a measured product result.
Papers in this series are preprints. They are not peer reviewed, and proposed outcomes require empirical validation.
Papers are published under Creative Commons Attribution 4.0 with stable URLs, open PDFs and machine-readable records. No login and no interaction gate.
Every record below is public, versioned and directly accessible without login.
Questions about the research program go to roman@wisdomtwin.ai.
AI Judgment Twins for regulated enterprises: a sovereign judgment layer so high-consequence decisions move while the question is live, with a named human still accountable.
Watch a Judgment Twin handle a live enterprise decision.