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 They are not peer reviewed and report proposed outcomes require empirical validation.
Three working papers 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 Not peer reviewed.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 CEO, WisdomTwin, Inc..
Proposed control specification. Working paper; not peer reviewed. No DOI assigned.
A role-specific AI assistant creates value only if an authorized recipient can use its evidence without mistaking a generated recommendation for the authority of an officeholder. This paper translates the judgment-latency framework of WT-100-001 into a proposed Trust Layer: a set of enforceable boundaries around data admission, role memory, retrieval, response delivery, and delegated action. The design treats authorization as specific to the requesting person, intended recipient, source, purpose, time, and requested action. It separates historical precedent from current policy and decision authority, requires safe decline when evidence or permissions are uncertain, and specifies an audit record without requiring disclosure of private model reasoning. Voice meetings and multi-party conversations receive explicit audience controls. A traceability matrix connects principal failure modes to testable controls and release evidence. The contribution is a falsifiable specification for future implementation and evaluation. No claim is made that the proposed controls have been implemented, independently validated, or shown to satisfy any particular legal regime.
Roman Bodnarchuk, Co-Founder and CEO, WisdomTwin, Inc..
Proposed protocol. Not peer reviewed, not preregistered. Host-dependent fields unresolved. No study conducted and no results reported. No DOI assigned.
Claims that role-specific AI can save hours per day require a design that distinguishes elapsed waiting from human effort, counts review and maintenance burden, and treats decision quality and confidentiality as constraints. This paper proposes a staged evaluation of the AI Judgment Twin concept. Retrospective testing establishes bounded task performance, prospective shadow mode estimates feasibility without affecting decisions, and an authorized controlled pilot compares assisted work with the existing workflow. Decision readiness is defined independently of treatment assignment. The primary outcome is elapsed time to accountable human disposition, supplemented by the narrower judgment-latency interval and net human time per eligible episode. All assigned episodes, including declines and failures, remain in the analysis. The protocol specifies quality review, cluster-aware analysis, missing-data treatment, serious-incident stopping rules, and claim wording tied to observed evidence. It is a research design and does not establish that WisdomTwin reduces time, preserves quality, or is ready for regulated deployment.
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.
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.