Mallea: Building, Steering, and Memorizing Worlds with Coding Agent

arXiv 2026 35 method

TLDR

Mallea uses a coding agent to build, steer, and memorize interactive code-based world models from text, with object rules, state, and texture memory rendered generatively.

Reasoning

The paper presents a coherent system emphasizing object-centric code world models, interactive editing, and visual memory. However, the abstract lacks concrete experimental results or benchmarks, so its empirical contribution and real-world validity remain unclear.

Read-first score

Read-first score 35, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 37.

Recency 5%

100

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

Methodology quality 15%

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

Topical relevance 25%

52.9

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 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,github

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 313.

Keyword Scores

world model

interactive world model

world simulator

generative world model

world dynamics prediction

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Mallea introduces a system for building, steering, and memorizing text-described worlds using a coding agent.
A coding agent infers object interaction requirements from the prompt and jointly plans object structure, properties, and executable rules, favoring compact representations.
Object texture memory retains reusable appearance only for whitebox-defined objects and adapts as those objects change.
World-state advancement by the program is separated from a generative renderer that combines current object structure with remembered appearance.
Construction, intervention, and memory are organized around the same objects, enabling subsequent observations to be conditioned on the latest state and compatible appearance.

Methodology

Mallea uses a coding agent to infer object interaction requirements from a text prompt, then jointly plans compact object structure, properties, and executable rules. The planned rules advance world state, while optional user actions steer the world and text requests can edit properties or extend capabilities while preserving unaffected state. A generative renderer combines the current object structure with object texture memory, which stores and adapts reusable appearance only for whitebox-defined objects.

Key Results

The abstract does not report quantitative or comparative experimental results. The main claimed outcome is that organizing construction, intervention, and memory around the same objects conditions subsequent observations on the latest world state and compatible appearance.

Resources

Paper