World Models in Pieces: Structural Certification for General Agents

arXiv 2026 44.3 theory, method

TLDR

Proposes structural certification for general agents, proving world models can be decomposed with bounded error guarantees for reliable deployment.

Reasoning

Strengths include formalizing limitations of universal guarantees and providing constructive algorithms with tight error bounds. Weaknesses are the lack of empirical validation and a purely theoretical focus, which may limit immediate practical applicability.

Read-first score

Read-first score 44.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 14.

Recency 6%

100

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

Citation impact 18%

94.5

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.94456897

Methodology quality 18%

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

Reproducibility 18%

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

Topical relevance 29%

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

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 365.

Keyword Scores

world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Formal proof that general agents are not universal, rendering standard worst-case analysis uninformative
Introduction of structural certification, a transition-local framework that maps bounded goal-conditioned performance to entry-wise guarantees on the agent's internal world model
Algorithms that filter specific transitions using deep compositional goals
Proof that a general agent on these goals has a structural world model with a O(1/n) + O(δ) error bound, which is tight in the small-δ regime
Enabling certifiable deployment of general agents by localizing specific transitions where long-horizon planning is reliable

Methodology

The paper proposes a structural certification framework that maps bounded goal-conditioned performance to entry-wise guarantees on the agent's internal world model. It provides algorithms to filter specific transitions using deep compositional goals, and proves theoretical error bounds of O(1/n) + O(δ) that are tight in the small-δ regime.

Key Results

The paper proves that general agents are not universal, and introduces structural certification with a O(1/n) + O(δ) error bound that is tight in the small-δ regime, enabling localization of reliable transitions for long-horizon planning.

Resources

Paper