⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

The Trinity of Consistency as a Defining Principle for General World Models

arXiv 26.2 2026 60.8 theory

TLDR

Proposes Trinity of Consistency (modal, spatial, temporal) as principle for general world models, with CoW-Bench benchmark for evaluation.

Reasoning

The paper provides a principled theoretical framework and introduces a benchmark, which are strengths. However, the abstract lacks concrete empirical results and the framework remains conceptual without experimental validation.

Read-first score

Read-first score 60.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 42.

Recency 8%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 25%
80

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark,evaluation,metric

Topical relevance 42%
60

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%
30

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 267.

Keyword Scores

world model
10
world simulator
8
generative world model
7
video world model
7
world dynamics prediction
6
interactive world model
2
model-based reinforcement learning world model
2

Deep Analysis

Innovations

  • Proposal of the Trinity of Consistency as a defining principle for General World Models, comprising Modal Consistency (semantic interface), Spatial Consistency (geometric basis), and Temporal Consistency (causal engine).
  • Introduction of CoW-Bench, a benchmark centered on multi-frame reasoning and generation scenarios, with a unified evaluation protocol for both video generation models and Unified Multimodal Models (UMMs).

Methodology

The paper systematically reviews the evolution of multimodal learning, tracing a trajectory from loosely coupled specialized modules toward unified architectures that enable internal world simulators. It introduces CoW-Bench, a benchmark designed for multi-frame reasoning and generation, and evaluates both video generation models and UMMs under a unified evaluation protocol.

Key Results

CoW-Bench provides a unified evaluation protocol for assessing video generation models and UMMs on multi-frame reasoning and generation tasks, enabling systematic comparison and highlighting the limitations of current systems.

Limitations

  • The field currently lacks a principled theoretical framework that defines the essential properties requisite for a General World Model.
  • Current systems do not satisfy the Trinity of Consistency (Modal, Spatial, and Temporal Consistency), as clarified by the proposed framework.

Resources

Paper

Tags