Procedural Content Metageneration via Program Search and Continual Abstraction Discovery

Matthew Siper; Ahmed Khalifa; Julian Togelius

cs.AIAug 18, 2026

Procedural Content Metageneration via Program Search and Continual Abstraction Discovery

Authors: Matthew Siper, Ahmed Khalifa, Julian Togelius

Organizations: Game Innovation Lab New York University New York, USA · Institute of Digital Games University of Malta Msida, Malta

Abstract

Large language models can generate executable programs, which makes it possible to search directly over procedural content generators rather than individual levels. We study this approach in Sokoban, Zelda, Dangerous Dave, and Lode Runner. Each run evolves complete Python generators through language-model mutation and crossover. We introduce Continual Abstraction Discovery, or CAD, which extracts reusable primitives from high-fitness programs into a run-specific helper module. A 2x2 experiment crosses CAD with access to a fixed hand-written domain API. The completed data set contains 160 complete runs, with at least ten 50-generation runs in every cell. CAD raises mean final best fitness in all eight domain and API comparisons. Across all CAD runs, learned libraries are adopted by most later programs and repeatedly rediscover validation, reachability, and structural utilities. These results support that discovering reusable primitives improves evolutionary program search for content generators.

Figures & tables

Fig. 2: Best-so-far fitness over 50 generations in each domain. Solid lines show CAD and dashed lines show no CAD. Shading reports mean

\pm

standard error, and legends show the completed run count for each cell. The plot uses CL as the implementation label for CAD.

Fig. 3: Representative generated levels from the Base condition without CAD on the left and with CAD on the right. Rows show Zelda, Sokoban, Dangerous Dave, and Lode Runner. Each row contains four samples. The images illustrate output structure and are not used as evidence of playability, which is measured through benchmark execution.

Fig. 4: Mechanism measurements across CAD runs. Panel A shows learned-library growth by domain. Panel B shows the reported adoption rate and helper calls per program. Panel C compares the line count of the best solution under Base and API prompting with and without CAD. Shading and error bars report uncertainty around the mean. The plot uses CL as the implementation label for CAD.

Abstraction	Category	Total Calls
In Bounds	Other	4,693
Entity Count Normalization	Validation	3,848
Ensure One Player	Validation	2,611
Grounded Empty Cells	Utility	465
Flood Fill	Other	569
Deep Copy Grid	Other	118

TABLE I: Most frequently rediscovered learned abstractions across all CAD runs. Semantically equivalent functions are grouped, and Total Calls is the aggregate call count in the supplied audit. The complete audit is provided in the supplementary source.

Fig. 5: Two primitives from the evolved helper library, each applied to a level (before, left; after, right; changed tile circled): guard_gold places an enemy beside a gold piece in Lode Runner, and place_enemies_away_from_critical places an enemy off the player $\rightarrow$ key $\rightarrow$ door critical path (white dots) in Zelda.

Fig. 6: Mutation prompt. The prompt specifies the generate(context_dict) contract, domain constraints, run memory, available helper APIs, mutation strength, and the parent program.

Fig. 7: Crossover prompt. The prompt receives two parent programs and asks the model to combine compatible mechanisms into a complete child program.

Fig. 8: Correction prompt. The prompt is used when a mutation or crossover output fails to compile or crashes during smoke testing. It receives the broken code and the error traceback.

Fig. 9: Memory reflection prompt. The prompt maintains a structured per-run memory with generation observations and global lessons for later variation calls.

Fig. 10: CAD helper extraction prompt. Starting at generation 8 and every five generations thereafter, the prompt extracts reusable domain utilities from programs in the top fitness quartile.

Fig. 11: CAD helper correction prompt. The prompt receives the helper module and its compile or smoke-test error.

Fig. 12: Refactoring prompt. The prompt rewrites a generator to replace duplicated inline logic with calls to tested helper functions.

Fig. 13: Missing-helper scaffold prompt. The prompt is used when a CAD program imports a helper that does not yet exist in its run-specific module.

Procedural Content Generation

Procedural content generation creates game levels, 3D worlds, assets, stories, and simulation environments through algorithms and learned models.

See all 66 papers

SeaEvo: Advancing Algorithm Discovery with Strategy Space Evolution
Apr 27, 2026Sichun Luo, Yi Huang, Haochen Luo +7Automated Algorithm Discovery Evolutionary Optimization
1The University of Hong Kong · 2JIUTIAN Research, China Mobile · 3City University of Hong Kong
AgenticEvolve: Evolutionary Program Search with Persistent Language Model Agents that Design and Run Experiments
Oct 22, 2026Ze-Yu Peng, Qing-Yu Lan, Zhan-He Wang +2LLMs for Evolutionary Optimization LLM-Based Program Synthesis
School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China. · International Centre for Theoretical Physics Asia-Pacific, University of Chinese Academy of Sciences, Beijing 100190, China. · Taiji Laboratory for Gravitational Wave Universe, University of Chinese Academy of Sciences, Beijing 100049, China. +2
PACE: Primitive-Aware Code Evolution for Automated Algorithm Design
Aug 7, 2026Zhuoliang Xie, Ruihao Zheng, Xiang Xu +2Automated Algorithm Discovery Automated Heuristic Design
1Southern University of Science and Technology · 2Shenzhen University

Apr 27, 2026J/K move · Enter open · S save

SeaEvo: Advancing Algorithm Discovery with Strategy Space Evolution

Sichun Luo, Yi Huang, Haochen Luo, Fengyuan Liu+6

1The University of Hong Kong · 2JIUTIAN Research, China Mobile · 3City University of Hong Kong

Automated Algorithm Discovery Evolutionary Optimization Large Language Model-Guided Optimization LLM-Based Program Synthesis LLMs for Evolutionary Optimization

Open paper AskPDF arXiv

Large Language Model (LLM)-guided evolutionary search is increasingly used for automated algorithm discovery, yet most current methods track search progress primarily through executable programs and scalar fitness. Even when natural-language reasoning is used through heuristic descriptions or reflection, it typically remains transient mutation context or unstructured memory, rather than organized as persistent population-level state over strategic directions. As a result, evolutionary search can struggle to distinguish syntactically different implementations of the same idea, preserve lower-fitness but strategically promising directions, or detect when an entire family of strategies has saturated. We introduce \model, a modular strategy-space layer that turns language-level strategic reasoning into first-class population-level evolutionary state in LLM-driven program search. \model represents each candidate program with an explicit natural-language strategy, clusters the archive by strategy semantics, retrieves behaviorally complementary inspirations, and periodically navigates the strategy landscape to avoid saturated directions. Without modifying the underlying evolutionary algorithms, \model improves existing evolutionary backbones across algorithm discovery, systems optimization, and agent-scaffold design tasks in most settings. Across four systems benchmarks, \model achieves a 20.6% average relative improvement, with the best single run on Prism scoring 3

\times

higher. These results suggest that persistent strategy representations provide a practical mechanism for improving the effectiveness and cost-efficiency of LLM-guided evolutionary search, pointing toward compound AI systems whose search capabilities benefit from the structured accumulation and reuse of algorithmic strategies.

Abstract

Procedural Content Generation

Explore similar work

SeaEvo: Advancing Algorithm Discovery with Strategy Space Evolution

AgenticEvolve: Evolutionary Program Search with Persistent Language Model Agents that Design and Run Experiments

PACE: Primitive-Aware Code Evolution for Automated Algorithm Design