jBOKResearch compendium
Work in developmentPublic revision

Research notes on trustworthy simulation and AI

jBOK is an evolving compendium of methods, design principles, and open questions for simulation systems used by people and AI agents. It connects physics-based modeling, computational mathematics, verification, validation, uncertainty quantification, provenance, and accountable decision-making.

The question: what evidence is needed to rely on a model for a particular decision—and how can that evidence remain understandable as models and software change?

Research in development. This public edition presents sourced synthesis and proposed evaluation methods. JWM, the experimental implementation, remains private while it develops toward a shareable MVP. This edition adds no experimental results. Historical implementation observations are labeled separately.

Start with the reasoning

What this work contributes today

The present contribution is the selection and connection of established ideas into an emerging implementation framework. Each recommendation distinguishes what comes from the literature, what is an architectural judgment, and what still needs evaluation. Citations establish intellectual provenance; they do not validate this framework by association.

The first reader is the author: an AI researcher and applied developer working in modeling, simulation, and analysis (MS&A). This is a working reference for making and revisiting implementation decisions. Other researchers and engineers may find the same reasoning useful. Decision-makers benefit through the evidence attached to the resulting work, rather than through a jBOK label alone.

The long-term aim is to support assessment of physical, learned, and hybrid world models. Broad applicability is a research objective: different models expose different information and permit different assurance claims. A future world-model architecture may emerge from this work, but it is not a result claimed here.

Working philosophy

AI-driven, physics-based, and trusted remain the organizing concerns, grounded in computational and mathematical methods. Here, “trusted” means a bounded claim supported by evidence for a stated use—not a property conferred by a label.

From synthesis to experiment

jBOK organizes the reasoning. JWM is the private experimental workspace used to explore implementation choices. Experience from one implementation can suggest a useful pattern; it cannot establish that the pattern generalizes.

The JWM profile preserves previously published, dated implementation observations, including limitations and negative results. Its latest historical snapshot is August 11, 2026. The source and supporting execution artifacts are not publicly available for independent reproduction.

The draft contract concepts and decision records document the developing architecture. They are working proposals and project decisions, with their scope stated explicitly.

Latest public revision

November 3, 2026 — benchmark workbench and visual guide. Added a sourced two-benchmark register, a comparison interface for local result files, and animated explanations of the measurements. Local pendulum evaluation tooling is implemented; no experimental scores are published. Read the research log.