Involving before Evolving: A Vision for Trustworthy Enterprise Digital Twin Engineering
Kérian Fiter, Adil Lagrou, Franck Dervault, Bentley Oakes
cs.SE · cs.AI · cs.HC
Quick verdict
A credible, industrially grounded vision for enterprise digital twins—prototype, federation, observability—but its stakeholder buy-in evidence is anecdotal.
Summary
This MODELS Companion 2026 vision paper proposes the 'involving before evolving' paradigm: secure organizational buy-in for Enterprise Digital Twins (EDTs) by rapidly building a working prototype with foundation models, then evolving toward federated interoperability. Grounded in a collaboration with Michelin, the authors extract five challenges (C1–C5) from informal stakeholder discussions and structure a three-stage roadmap: (1) prototype creation using LLMs to build UIs and models, a locally hosted TimesFM 2.5 time-series foundation model, and Monte Carlo simulation — users edit spline trajectories with 10/90th-percentile uncertainty bands, and uncertainty propagates through expert-developed downstream models; (2) federation through an OML ontological backbone acting as structural wiring, typed interfaces, and semantic constraints; (3) observability via versioned predictions, DTInsight-based reporting, and locally hosted models for data sovereignty. The authors report that the marketing/finance prototype satisfied SME requirements and secured cross-department buy-in, generating new requirements around multi-dataset correlation and TSFM fine-tuning. Evidence is thin: buy-in is anecdotal with no evaluation protocol, all benefits remain aspirational, and the authors candidly flag unaddressed scenario provenance, uncertain scalability of ontology contracts, and the infeasibility of longitudinally validating their core assumption that EDT-informed decisions improve outcomes.
Review complete
Assessment at a glance
3
risk flags identified
● high 0
● medium 2
● low 1
4 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Early organizational buy-in, secured via rapidly built foundation-model prototypes, is the critical precondition for successful long-term EDT evolution — the 'involving before evolving' paradigm.
- EDTs are 'long-loop' digital twins whose loop closes through human decisions at day-to-year time scales with potentially irreversible consequences, raising trust requirements beyond tight-loop DTs.
- A three-stage approach — FM-based prototyping, OML-based ontological federation, and observability tooling — can deliver trustworthy, interoperable EDTs across departmental silos.
- An initial prototype at Michelin (TimesFM 2.5 + spline editing + Monte Carlo) satisfied SME requirements and secured cross-department buy-in in marketing and finance.
Strengths
Genuine industrial grounding with a realized artifact
Unlike many vision papers, stage 1 already exists: a working prototype for Michelin's marketing and finance departments (Section 3, Figure 2), built with agentic LLMs on company infrastructure, a locally hosted TimesFM 2.5, spline editing with 10/90th-percentile uncertainty bands, and Monte Carlo propagation through expert-developed models.
Coherent staged roadmap with explicit challenge mapping
Figure 1 transparently maps the five challenges C1–C5 onto the three stages (prototype → C1–C3, federation → C4, observability → C5), and Section 4 articulates three concrete ontology roles — structural wiring, typed interfaces, semantic constraints — illustrated by the finance-approval rule and the historical-data-fidelity reasoning example.
Candid open-challenges section
Clear differentiation from adjacent work
Weaknesses
Central buy-in claim is unevaluated anecdote
Challenges and benefits rest on informal grounds
Speculative world-model component
No quantitative evidence of any kind
Reproducibility is essentially nil
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a 5-page vision/position paper (MODELS Companion 2026, DOI 10.1145/3837062.3839094; arXiv:2609.13071, cs.SE) from Polytechnique Montréal and Michelin Canada. Its central claims: (1) Enterprise Digital Twins (EDTs) are 'long-loop' digital twins — the loop closes through human decisions at days-to-years scale with potentially irreversible consequences — so trust and early stakeholder involvement, not automation, are the binding constraints (Section 1); (2) an 'involving before evolving' paradigm — rapidly involve stakeholders via a foundation-model prototype, then evolve toward ontology-backed federation and observability — best secures organizational buy-in (Section 2); (3) a working prototype in Michelin's marketing and finance departments has already satisfied SME requirements and secured buy-in (Section 3). The paper explicitly targets the socio-technical adoption problem rather than simulation technique, positioning itself against Barat et al.'s agent-based ESL line [4–7], Franzén's ontology-first SoS exploration [18], and Giabbanelli et al.'s LLM-as-translator [20].
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 3 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What does 'involving before evolving' mean in this paper?
It is the proposed paradigm of first securing organizational buy-in by rapidly building a working Enterprise Digital Twin prototype with foundation models (LLMs plus the TimesFM time-series model), and only then evolving the system toward ontology-based federation and full interoperability across departments.
Was the approach evaluated, and are there any metrics?
No formal evaluation is reported. The authors state that a prototype for Michelin's marketing and finance departments satisfied subject-matter-expert requirements and helped secure buy-in, but no user study, forecast-accuracy numbers, or adoption metrics are provided, and the authors themselves list this evaluability gap among their open challenges.
Is code or data available for this paper?
No paper-specific artifact is released. The only public component referenced is Google Research's TimesFM repository; the prototype was built with company-hosted LLMs on Michelin infrastructure using sensitive business data, so the system cannot be reproduced externally.
What is a 'long-loop' digital twin?
The authors' term for a digital twin whose control loop closes through human decisions over long time scales (days to years) with potentially irreversible consequences — unlike typical digital twins with automatic, tight feedback loops. This is why they argue trust and stakeholder understanding matter more for Enterprise Digital Twins.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1 and 2 are available in the extraction only as inline text labels and captions, not as rendered images, so visual details of the UI and diagram layout were assessed from their textual descriptions alone.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.