# World Models in Pieces: Structural Certification for General Agents

arXiv: https://arxiv.org/abs/2606.24842 · HTML: https://mlsanity.com/paper/2606.24842/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in World models for language agents, one of 11 directions in World models.
- Builds on 20 of its 86 references that are in the map.
- First posted on arXiv 23 Jun 2026.

## Placed in

- World models / [World models for language agents](https://mlsanity.com/wm/world-models-for-language-agents/)

## Builds on (most cited first, 20 of 20)

- [Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems](https://mlsanity.com/paper/2005.01643/) (2020)
- [When to Trust Your Model: Model-Based Policy Optimization](https://mlsanity.com/paper/1906.08253/) (2019)
- [WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models](https://mlsanity.com/paper/2401.13919/) (2024)
- [Large Language Model Agent: A Survey on Methodology, Applications and Challenges](https://mlsanity.com/paper/2503.21460/) (2025)
- [Understanding World or Predicting Future? A Comprehensive Survey of World Models](https://mlsanity.com/paper/2411.14499/) (2024)
- [Goal-Conditioned Reinforcement Learning: Problems and Solutions](https://mlsanity.com/paper/2201.08299/) (2022)
- [AgentGym: Evolving Large Language Model-based Agents across Diverse Environments](https://mlsanity.com/paper/2406.04151/) (2024)
- [From Word to World: Can Large Language Models be Implicit Text-based World Models?](https://mlsanity.com/paper/2512.18832/) (2025)
- [OSWorld-Human: Benchmarking the Efficiency of Computer-Use Agents](https://mlsanity.com/paper/2506.16042/) (2025)
- [Automatic Failure Attribution and Critical Step Prediction Method for Multi-Agent Systems Based on Causal Inference](https://mlsanity.com/paper/2509.08682/) (2025)
- [Dynamic Model Predictive Shielding for Provably Safe Reinforcement Learning](https://mlsanity.com/paper/2405.13863/) (2024)
- [Rethinking the Foundations for Continual Reinforcement Learning](https://mlsanity.com/paper/2504.08161/) (2025)
- [A Comprehensive Survey of Reinforcement Learning: From Algorithms to Practical Challenges](https://mlsanity.com/paper/2411.18892/) (2024)
- [Simplifying Latent Dynamics with Softly State-Invariant World Models](https://mlsanity.com/paper/2401.17835/) (2024)
- [Probabilistic Subgoal Representations for Hierarchical Reinforcement learning](https://mlsanity.com/paper/2406.16707/) (2024)
- [Provably Efficient Offline Goal-Conditioned Reinforcement Learning with General Function Approximation and Single-Policy Concentrability](https://mlsanity.com/paper/2302.03770/) (2023)
- [Goal-Conditioned Reinforcement Learning from Sub-Optimal Data on Metric Spaces](https://mlsanity.com/paper/2402.10820/) (2024)
- [Look Before Leap: Look-Ahead Planning with Uncertainty in Reinforcement Learning](https://mlsanity.com/paper/2503.20139/) (2025)
- [Subjective Causality](https://mlsanity.com/paper/2401.10937/) (2024)
- [Partial Identifiability in Inverse Reinforcement Learning For Agents With Non-Exponential Discounting](https://mlsanity.com/paper/2412.11155/) (2024)

## Abstract (arXiv)

In the big-world regime, agents cannot be universally capable and their ability is inevitably specialized across a world model in pieces. Consequently, standard uniform guarantees fail to distinguish between the understanding of critical bottlenecks and irrelevant failures. We first formalize this limitation by proving that general agents are not universal, rendering standard worst-case analysis uninformative. To overcome this, we introduce structural certification, a transition-local framework that maps bounded goal-conditioned performance to entry-wise guarantees on the agent's internal world model. Our main contribution is constructive. We provide algorithms that filter specific transitions using deep compositional goals and prove that a general agent on these goals has a structural world model with a O(1/n) + O(δ) error bound. Conversely, this bound is tight in the small-δ regime, whose existence is explicitly guaranteed by our certification. These results enable the certifiable deployment of general agents by localizing the specific transitions where long-horizon planning is reliable.