Paper deep dive
Reconstructing Persistent Worlds from Narratives for Narrative-Grounded Interactive Experiences
Yi-Chun Chen
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 90%
Last extracted: 8/9/2026, 2:32:37 AM
Summary
This paper proposes a framework for reconstructing explicit persistent worlds from narrative descriptions to support narrative-grounded interactive experiences. Unlike existing approaches that generate task-specific representations for scenes or gameplay, this method maintains a shared computational foundation of entities, locations, semantic relationships, and evolving world states. The authors present a reference prototype that reconstructs structured world representations from text and instantiates playable tile-based environments, demonstrating feasibility through case studies involving procedural scenarios, fantasy narratives, and public-domain stories.
Entities (8)
Relation Signals (9)
Persistent World → supports → Narrative-Grounded Interactive Experiences
confidence 95% · We investigate reconstructing explicit persistent worlds from narrative descriptions as the central computational objective for narrative-grounded interactive realization.
Reference Prototype → reconstructs → Persistent World
confidence 93% · we develop a reference prototype that reconstructs structured persistent world representations from narrative descriptions
Persistent World → contains → Locations
confidence 92% · the persistent world maintains persistent entities, locations, semantic relationships, and evolving world states
Persistent World → contains → Entities
confidence 92% · the persistent world maintains persistent entities, locations, semantic relationships, and evolving world states
Persistent World → contains → Semantic Relationships
confidence 92% · the persistent world maintains persistent entities, locations, semantic relationships, and evolving world states
Persistent World → contains → Evolving World States
confidence 92% · the persistent world maintains persistent entities, locations, semantic relationships, and evolving world states
Existing approaches → contrastswith → Persistent World Reconstruction
confidence 90% · Existing approaches formulate problems such as narrative planning, scene generation, and gameplay generation, each constructing computational representations tailored to specific downstream tasks rather than explicitly reconstructing and maintaining the persistent world
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Designing narrative-grounded interactive experiences remains labor-intensive because interactive content must align with the underlying world implied by the narrative. Existing approaches formulate problems such as narrative planning, scene generation, and gameplay generation, each constructing computational representations tailored to specific downstream tasks rather than explicitly reconstructing and maintaining the persistent world that grounds them. We investigate reconstructing explicit persistent worlds from narrative descriptions as the central computational objective for narrative-grounded interactive realization. Rather than treating the world as an implicit by-product of downstream generation, our approach reconstructs and maintains persistent entities, locations, semantic relationships, and evolving world states while inferring only the contextual information required to support coherent interactive experiences. To investigate this perspective, we develop a reference prototype that reconstructs structured persistent world representations from narrative descriptions and subsequently instantiates playable tile-based environments. Through three representative case studies spanning a procedural scenario, an original fantasy narrative, and an adapted public-domain story, we demonstrate the feasibility of reconstructing persistent worlds and show how a shared world representation supports coherent gameplay while remaining grounded in the source narrative. By explicitly reconstructing persistent worlds prior to interactive realization, this work bridges computational narrative understanding and interactive content generation, providing a semantic foundation for AI-assisted game authoring, mixed-initiative design, educational simulations, and narrative-grounded interactive experiences.
Tags
Links
- Source: https://arxiv.org/abs/2608.04037v1
- Canonical: https://arxiv.org/abs/2608.04037v1
PDF not stored locally. Use the link above to view on the source site.
Full Text
91,457 characters extracted from source content.
Expand or collapse full text
Reconstructing Persistent Worlds from Narratives for Narrative-Grounded Interactive Experiences Yi-Chun Chen National Cheng Kung University Tainan City 701, Taiwan rimi.chen@gs.ncku.edu.tw, ychen74@alumni.ncsu.edu Abstract Designing narrative-grounded interactive experiences re- mains labor-intensive because interactive content must align with the underlying world implied by the narrative. Existing approaches formulate problems such as narrative planning, scene generation, and gameplay generation, each construct- ing computational representations tailored to specific down- stream tasks rather than explicitly reconstructing and main- taining the persistent world that grounds them. We investigate reconstructing explicit persistent worlds from narrative descriptions as the central computational objective for narrative-grounded interactive realization. Rather than treating the world as an implicit by-product of downstream generation, our approach reconstructs and maintains persis- tent entities, locations, semantic relationships, and evolving world states while inferring only the contextual information required to support coherent interactive experiences. To investigate this perspective, we develop a reference pro- totype that reconstructs structured persistent world represen- tations from narrative descriptions and subsequently instan- tiates playable tile-based environments. Through three repre- sentative case studies spanning a procedural scenario, an orig- inal fantasy narrative, and an adapted public-domain story, we demonstrate the feasibility of reconstructing persistent worlds and show how a shared world representation supports coher- ent gameplay while remaining grounded in the source narra- tive. By explicitly reconstructing persistent worlds prior to in- teractive realization, this work bridges computational narra- tive understanding and interactive content generation, pro- viding a semantic foundation for AI-assisted game author- ing, mixed-initiative design, educational simulations, and narrative-grounded interactive experiences. 1 Introduction Narratives provide a natural foundation for interactive expe- riences, including narrative-driven games, interactive story- telling, educational simulations, training environments, and mixed-initiative authoring systems (Cardona-Rivera, Zagal, and Debus 2020; Horswill, Montfort, and Young 2014). Rather than specifying every detail of an environment, nar- ratives communicate an intended world through characters, events, locations, and their evolution over time. Designers subsequently construct environments, gameplay, and visual assets that realize this world while maintaining coherence with the source narrative (Cardona-Rivera, Zagal, and Debus 2023). As AI increasingly assists game development, inter- active content must remain grounded in the narrative world rather than merely being individually plausible. Recent advances have introduced computational prob- lems such as narrative planning (Ware and Siler 2021), scene generation from narrative descriptions (Chen and Jhala 2025b), gameplay generation, and end-to-end game genera- tion (Zhou et al. 2025). These approaches construct compu- tational representations tailored to specific downstream ob- jectives. However, narratives intentionally omit much of the contextual information required for interactive experiences, including persistent entities, spatial organization, and evolv- ing world states. Consequently, downstream systems must reconstruct and maintain aspects of the underlying world be- fore coherent interactive experiences can be realized. Although existing approaches necessarily reconstruct some representation of the underlying world, these rep- resentations are typically created to support a particular downstream objective, such as planning, scene generation, or gameplay realization. As a result, they primarily func- tion as intermediate artifacts rather than reusable computa- tional objects shared across subsequent computations. We instead argue that the persistent world implied by a narrative should be reconstructed explicitly before interactive realiza- tion. Unlike individual scenes or gameplay sequences, the persistent world maintains entities, locations, semantic rela- tionships, and evolving world states across narrative events and player interactions. Figure 1 illustrates this distinction between existing narrative-conditioned generation pipelines and our reconstruction-centered framework, in which an ex- plicit persistent world serves as the computational founda- tion for interactive realization. Motivated by this perspective, we investigate reconstruct- ing explicit persistent worlds from narrative descriptions as the central computational objective for narrative-grounded interactive realization. We interpret narratives as partial ob- servations of an underlying world rather than complete spec- ifications of interactive content. The reconstructed world maintains persistent entities, locations, semantic relation- ships, and evolving world states while inferring only the contextual information required for coherent interactive re- alization. This reconstructed world subsequently serves as a reusable computational foundation from which multiple arXiv:2608.04037v1 [cs.CL] 3 Aug 2026 Figure 1: Persistent world as a computational foundation for narrative-grounded interactive realization. Existing methods con- struct task-specific representations for downstream generation, whereas our framework reconstructs an explicit persistent world that supports multiple interactive realizations while preserving narrative consistency. narrative-grounded interactive experiences can be instanti- ated. The contributions of this paper are as follows: • We investigate reconstructing persistent worlds from nar- rative descriptions as the central computational objec- tive for supporting narrative-grounded interactive expe- riences. • We develop a reference prototype that reconstructs per- sistent worlds from narrative descriptions and instanti- ates playable tile-based environments. • Through representative case studies, we demonstrate the feasibility of persistent world reconstruction and show that a shared persistent world provides a reusable com- putational foundation for coherent narrative-grounded in- teractive realization. 2 Persistent Worlds for Narrative-Grounded Interactive Experiences This section formulates the need for reconstructing an ex- plicit persistent world from narrative descriptions to support narrative-grounded interactive experiences. Rather than di- rectly transforming narratives into scenes, levels, or com- plete games, we treat the persistent world as the primary computational object between narrative understanding and interactive realization. We first define the scope of narra- tives considered in this work, explain why interactive real- ization requires a persistent world, define the role of that world, characterize narratives as partial observations, for- mulate persistent world reconstruction from narrative de- scriptions, and clarify how the reconstructed world supports downstream interactive realization. Narrative Scope This work does not seek to provide a general linguistic or narratological definition of narrative. Instead, it defines the class of narratives addressed by the proposed formulation. We consider narratives that describe the evolution of an un- derlying world through a sequence of temporally related ob- servations or events. Although each observation reveals only part of the world, the narrative collectively provides infor- IncludedOutside Scope Fictional storiesImage captions Narrative questsSingle scene descriptions Educational simulationsStandalone prompts Procedural scenarios Isolated observations Historical narratives Single moment descriptions Role-playing game scriptsStandalone object descrip- tions Table 1: Examples of narrative inputs considered in this work and inputs outside its scope. mation about entities, relationships, locations, and states that persist or change over time. Inputs that do not describe temporal world evolution, such as image captions or isolated scene descriptions, fall out- side the scope of this formulation. Such inputs may support the realization of an individual scene, but they do not pro- vide a sequence of observations from which persistence and change across events can be reconstructed. Table 1 summa- rizes the narrative types considered in this work. Why Interactive Realization Requires a Persistent World Narratives communicate enough information for readers to follow characters, events, and changes in the narrated situa- tion, but they rarely specify everything required to construct an interactive experience. Interactive realization additionally requires decisions about which entities exist, where they are located, how locations are connected, which object states persist, what interactions are possible, and how actions af- fect later situations. These requirements cannot be addressed independently for every scene or gameplay sequence without risking in- compatible assumptions. An entity introduced in one event may need to remain available later; an object collected in one location may need to remain in the player’s inventory after a transition; and an opened door or altered object state may need to persist across subsequent interactions. If each scene or downstream process reconstructs this information sepa- rately, individually plausible outputs may fail to describe one coherent world. A shared persistent world provides the computational ba- sis for coordinating these requirements. It preserves infor- mation communicated earlier, integrates later observations and interaction updates, and supplies a common reference for constructing scenes, determining possible actions, and maintaining continuity across an interactive experience. The need for a persistent world therefore arises not merely from a desire to store reusable information, but from the require- ment that multiple narrative events and interactive realiza- tions remain grounded in the same evolving world. Persistent World Narratives rarely describe complete worlds. Instead, they re- veal observations that readers naturally interpret as occur- ring within a shared persistent context. Characters, loca- tions, objects, and events introduced throughout a narrative are understood as belonging to the same underlying world, even though many aspects of that world remain implicit. We define a persistent world as the computational account of the narrative world that persists across narrative observa- tions and subsequent interactions. Rather than correspond- ing to a particular scene or gameplay sequence, the persis- tent world maintains entities, locations, semantic relation- ships, and evolving world states that provide a common ref- erence for interpreting observations and supporting interac- tive realization. Here, the term world does not imply a complete speci- fication of every fact about a fictional universe. Instead, it refers only to the information required to maintain a coher- ent account of the entities, relationships, locations, and states relevant to the narrative and its interactive realization. The scope of the reconstructed world is therefore determined by the narrative observations together with the contextual infor- mation required for coherent interaction. This definition specifies the computational role of the persistent world rather than any particular implementation. A persistent world may be represented using knowledge graphs, symbolic world models, scene graphs, relational databases, or other structured representations. The proposed formulation is therefore independent of any particular rep- resentation. Unlike scenes, game levels, or gameplay se- quences, which each realize only part of the underlying world, the persistent world serves as a reusable computa- tional foundation from which multiple interactive realiza- tions can be consistently instantiated. Figure 2 illustrates this distinction. Narrative Observations The proposed formulation interprets narratives as partial ob- servations of a persistent world rather than complete spec- ifications of that world. Authors communicate information relevant to the intended discourse while leaving much of the surrounding context implicit. Human readers combine explicit observations with contextual reasoning and prior knowledge to construct a coherent account of the characters, objects, locations, and states involved. In this paper, we use the following short procedural narrative as a running exam- ple. Figure 2: Conceptual illustration of a persistent world. Nar- rative observations reveal only part of the underlying world, while the reconstructed persistent world maintains entities, locations, semantic relationships, and evolving world states that provide a shared foundation for multiple interactive re- alizations. A scientist entered a laboratory to conduct an experi- ment. After preparing the required equipment and ma- terials, the scientist mixed the chemicals to complete the experiment. The narrative explicitly describes the major activities and participating entities but omits much of the contextual in- formation required to instantiate an interactive environment, including the laboratory layout, object placement, navigable space, supporting environmental structures, and the persis- tent states of objects as the experiment progresses. Neverthe- less, readers can infer enough of this context to understand the narrated events. The proposed formulation adopts the same perspective computationally. Rather than treating the narrative as a com- plete specification of an interactive environment, it treats the narrative as a sequence of observations from which a persis- tent world can be reconstructed. Explicit narrative observa- tions provide the evidence for reconstruction. Constrained inference supplements only the contextual information re- quired for coherent interaction, while previously established entities, relationships, and states are maintained unless later observations or interactions modify them. The running ex- ample is revisited throughout the remainder of the paper to illustrate these operations. Persistent World Reconstruction Based on the preceding definitions, we formulate persistent world reconstruction from narratives as the computational task of constructing and maintaining an explicit persistent world from narrative observations. Let N =o 1 , o 2 , . . . , o n denote a narrative consisting of a sequence of temporally ordered observations. The objective is to reconstruct a per- sistent world W = (E, L, R, Σ), where E denotes persistent entities, L represents spatial organization, R captures semantic relationships, and Σ rep- resents the evolving world state. Formally, f : N → W, where the reconstructed world satisfies the following properties. • Narrative fidelity. Preserve information explicitly com- municated by the narrative. • Persistence and continuity. Maintain entities, locations, relationships, and states across observations and subse- quent interactions unless explicitly modified. • Constrained reconstruction. Reconstruct only the con- textual information required to support coherent interac- tive realization. • World consistency. Maintain one coherent account of entities, locations, relationships, and states as narrative observations and interaction updates accumulate. • Operational foundation. Provide the maintained world required to support downstream interactive realization. The formulation specifies the computational objective rather than prescribing any particular representation, recon- struction strategy, or implementation. Knowledge graphs, symbolic world models, relational databases, scene graphs, and hybrid representations are all compatible provided they satisfy the properties above. Persistent world reconstruction is therefore distinct from directly generating scenes, levels, or complete games. Those downstream realizations operate on the maintained persistent world rather than replacing it. Interactive Realization Once a persistent world has been reconstructed, it serves as the computational interface between narrative understanding and interactive realization. It provides the entities, locations, relationships, evolving states, and interaction-relevant con- text from which individual scenes and gameplay situations can be instantiated. The same persistent world may support multiple down- stream realizations, including explorable game environ- ments, educational simulations, interactive storytelling ex- periences, and mixed-initiative authoring systems. These re- alizations may differ in layout, presentation, mechanics, or interaction design while remaining grounded in the same narrative observations and maintained world. Interactive realization may subsequently update the per- sistent world. Player actions can modify entity locations, inventory contents, object states, or semantic relationships, and these updates become part of the maintained world available to later scenes and interactions. The persistent world, therefore, connects narrative interpretation with con- tinued interaction by serving as the long-lived computational state shared across both processes. 3 Related Work This work builds upon research in computational narrative, interactive storytelling, narrative-conditioned content gen- eration, procedural content generation, and mixed-initiative game design. Rather than organizing related work by appli- cation domain, we examine the computational representa- tions constructed by existing approaches and the roles those representations play in supporting narrative-grounded in- teractive experiences. Our focus is not simply on whether prior work represents narrative or world information, but on whether it reconstructs and maintains an explicit, persistent world as the shared computational object that links narrative interpretation, interactive realization, and subsequent inter- action. Narrative-Grounded Interactive Experiences Narratives underpin many interactive experiences by defin- ing characters, events, locations, goals, and their evolu- tion over time. Prior work has studied maintaining coher- ence between story progression and gameplay (Cardona- Rivera, Zagal, and Debus 2023), formal narrative reason- ing (Cardona-Rivera, Zagal, and Debus 2020; Horswill, Montfort, and Young 2014), experience management under player agency (Ware et al. 2022), and consistency across game artifacts generated with large language models (Gal- lotta, Liapis, and Yannakakis 2024). These approaches es- tablish narrative coherence as a central objective of AI- assisted interactive experiences. Maintaining coherence, however, requires more than en- suring that individual scenes, actions, or artifacts are locally plausible. Information introduced in one narrative event may constrain later environments, interactions, and game- play states. Characters and objects must retain their iden- tities, locations and relationships must remain compatible, and changes produced by narrative events or player actions must persist across subsequent realizations. This work fo- cuses on the explicit computational object needed to main- tain that continuity: a persistent world shared across narra- tive interpretation and interactive realization. Computational Narrative and Story Understanding Computational narrative develops representations that sup- port narrative understanding beyond surface text (Montfort and y P ́ erez 2023; Gerv ́ as 2024; Castricato et al. 2021). Re- cent work models storyworlds through world-state transi- tions (G ́ ongora et al. 2026), represents narrative chronol- ogy (Gerv ́ as and Calle 2024), and performs commonsense reasoning over event preconditions and effects (Xie et al. 2025). Structured representations further support down- stream applications, including hierarchical visual narrative understanding (Chen 2025) and collaborative visual narra- tive generation (Chen and Jhala 2024). These studies demonstrate that narrative understanding often requires explicit representations of entities, events, temporal relations, causal structure, and changing world states. Their primary objective, however, is to support nar- rative interpretation, reasoning, or generation. The present work builds upon these representational foundations while asking how the information conveyed through a narrative can be reconstructed and maintained as a persistent compu- tational world that remains available throughout subsequent interactive realization. Narrative-Conditioned Interactive Content Generation Narrative-conditioned content generation transforms narra- tive descriptions into interactive artifacts at multiple compu- tational levels. Existing work generates scenes (Chen and Jhala 2025b), environments (Kumaran et al. 2023; Nasir, James, and Togelius 2024; Buongiorno et al. 2024), and complete interactive fiction games (Zhou et al. 2025). Com- plementary research investigates narrative planning (Ware and Siler 2021; Siler, Fisher, and Ware 2025; Fisher and Ware 2024; Farrell and Ware 2024), experience manage- ment (Ware et al. 2022), and alignment between gameplay progression and narrative goals (Cardona-Rivera, Zagal, and Debus 2023; Rivera et al. 2024). These approaches construct plans, scenes, environments, gameplay structures, or complete interactive experiences from narrative descriptions. In doing so, they often infer in- formation not explicitly stated in the source narrative, such as spatial organization, object placement, event precondi- tions, affordances, or gameplay constraints. Such informa- tion is usually introduced to satisfy the requirements of a particular output or downstream task. This work instead makes the persistent world itself the primary computational object, allowing inferred and observed information to be maintained across multiple scenes, events, and interactions rather than reconstructed separately for each output. Procedural Content Generation and Mixed-Initiative Design Procedural Content Generation (PCG) develops computa- tional methods for generating game content (Summerville et al. 2018; Yannakakis and Togelius 2018). Recent re- search emphasizes intermediate representations throughout PCGML pipelines (Guzdial, Snodgrass, and Summerville 2025b), mixed-initiative workflows (Guzdial, Snodgrass, and Summerville 2025a), semantic models of game sys- tems (Cardona-Rivera, Zagal, and Debus 2022), co-creative AI systems (Lin and Riedl 2023; Agarwal, Lin, and Riedl 2023), automated game design (Cook 2022, 2025), and foundation-model-based PCG (Gallotta et al. 2024; Yan- nakakis and Togelius 2025). These studies show that ex- plicit intermediate representations can support controllabil- ity, author interaction, semantic organization, and coordina- tion among different stages of game creation. The persistent world considered in this work is related to such intermediate representations but serves a more spe- cific role. It is not merely a temporary format used to con- nect stages of a generation pipeline. It maintains narrative- grounded entities, locations, relationships, and evolving states across narrative events and player interactions. It therefore provides a continuing semantic reference from which different interactive realizations may be produced and updated. Positioning Persistent World Reconstruction from Narratives Prior research provides computational representations for narrative understanding, planning, scene synthesis, envi- ronment construction, gameplay generation, and mixed- initiative design (G ́ ongora et al. 2026; Ware and Siler 2021; Zhou et al. 2025). These representations capture many com- ponents also required by a persistent world, including en- tities, events, spatial relations, temporal structure, affor- dances, and evolving world states. The distinction lies not primarily in the information represented, but in the compu- tational role that representation serves. Existing approaches generally construct representations to support particular downstream tasks, such as interpret- ing a narrative, answering questions, generating a plan, syn- thesizing a scene, or producing an interactive experience. In contrast, persistent world reconstruction formulates the computational problem of constructing and maintaining the shared world implied by a narrative. The reconstructed per- sistent world preserves narrative observations, incorporates only the contextual information required for coherent inter- action, maintains continuity as narrative observations and player interactions accumulate, and serves as the common computational foundation for subsequent interactive realiza- tion. From this perspective, narrative understanding and in- teractive realization become complementary computational processes rather than independent stages. Narrative under- standing provides observations and inferred structure from which the persistent world is reconstructed, while interactive realization instantiates and updates that maintained world. Persistent world reconstruction, therefore, defines the com- putational interface connecting these processes rather than introducing another task-specific intermediate representa- tion. 4 Reference Framework To operationalize the proposed formulation, we introduce a reference framework that organizes persistent world recon- struction from narrative descriptions into a sequence of com- putational stages. Rather than directly generating scenes, levels, or complete games from narrative input, the frame- work places an explicit persistent world at the center of com- putation, allowing subsequent reasoning and interactive re- alization to operate on a shared semantic representation. As illustrated in Figure 3, the framework decomposes the pro- cess into narrative interpretation, persistent world construc- tion, world reasoning, interactive realization, and world- grounded gameplay. The framework specifies one conceptual organization of these computational stages rather than a fixed implemen- tation pipeline. Individual stages may be implemented us- ing different language models, symbolic reasoning systems, knowledge representations, or realization techniques with- out changing the underlying computational objective of re- constructing and maintaining a persistent world. The refer- ence prototype described in the following section instanti- ates one implementation of this framework. Framework Overview The reference framework consists of five computational stages: (1) narrative interpretation and observation extrac- tion, (2) persistent world construction, (3) spatiotemporal world reasoning, (4) interactive world realization, and (5) world-grounded gameplay design. Narrative interpretation extracts observations describing entities, locations, actions, semantic relationships, and state changes from the source narrative. Persistent world con- struction integrates these observations into a maintained per- sistent world through constrained world completion, recon- structing only the contextual information required for co- herent interaction while remaining grounded in the narra- tive evidence. Spatiotemporal world reasoning derives spa- tial organization, temporal consistency, and interaction con- straints from the reconstructed world. Interactive realization instantiates portions of the maintained world as interactive experiences, while world-grounded gameplay derives inter- actions and gameplay mechanics from that same world. Sub- sequent player interactions and newly introduced narrative observations update the persistent world, allowing it to re- main the shared computational state throughout continued interaction. Collectively, these stages describe one conceptual organi- zation of persistent world reconstruction rather than a fixed implementation pipeline. Different implementations may re- alize individual stages using different computational tech- niques while preserving the same underlying computational responsibilities. Narrative Interpretation and Observation Extraction Within the framework, the narrative interpretation stage treats a narrative as observations describing an evolving world rather than as a complete specification of an inter- active environment. The stage extracts explicit entities, lo- cations, actions, semantic relationships, and state changes and organizes them into structured observations. Because narratives intentionally omit information unnecessary for discourse, these observations provide only partial evidence about the underlying world. Additional observations in- troduced through branching interactions or author-directed modifications can subsequently be incorporated into the same persistent world. Persistent World Construction The persistent world construction stage integrates structured narrative observations into a maintained persistent world through constrained world completion. Rather than recon- structing every possible aspect of the underlying world, this stage reconstructs only the contextual information required to support coherent interactive realization while remaining grounded in the narrative evidence. The resulting persistent world maintains entities, locations, semantic relationships, evolving world states, and temporal organization as obser- vations accumulate over time. The framework specifies the computational responsibil- ity of this stage rather than a particular reconstruction strategy. Knowledge graphs, symbolic world models, scene graphs, relational databases, hybrid symbolic-neural repre- sentations, or other structured representations are all com- patible with the formulation, provided they support con- structing and maintaining a persistent world that satisfies the properties defined in Section 2. Different implementations may therefore realize this stage using different representa- tions while preserving the same computational objective. Spatiotemporal World Reasoning The spatiotemporal reasoning stage derives the structural in- formation required for interactive realization from the recon- structed persistent world. Spatial reasoning determines en- vironmental layouts, navigation, containment relationships, and location connectivity, while temporal reasoning main- tains event ordering, causal dependencies, and evolving world states. Consistency checks ensure that inferred knowl- edge remains compatible with both the reconstructed world and the original narrative. Because reasoning operates on the maintained persistent world rather than repeatedly interpret- ing the source narrative, information accumulated from ear- lier observations remains available throughout subsequent reasoning and interaction. Interactive World Realization The interactive realization stage instantiates portions of the reconstructed persistent world as interactive experiences. Rather than reconstructing world information independently for each realization, this stage operates on the maintained persistent world to derive the scene-level structures, interac- tion context, and state information required by a particular interactive experience. Because realization is separated from persistent world construction, multiple layouts, visual presentations, interac- tion paradigms, or gameplay experiences may be instanti- ated from the same reconstructed world while remaining grounded in shared entities, locations, semantic relation- ships, and evolving world states. The framework, therefore, separates maintaining the world from realizing individual in- teractive experiences, allowing different realizations to reuse and subsequently update the same persistent world. World-Grounded Gameplay Design The world-grounded gameplay design stage derives game- play mechanics from the reconstructed persistent world rather than authoring them independently. Interaction op- portunities, objectives, progression constraints, and state- dependent behaviors are grounded in persistent entities, affordances, semantic relationships, and evolving world states. The persistent world specifies what exists and how it evolves, while gameplay specifies how players interact with that world. Consequently, the same persistent world may support multiple gameplay experiences while remain- ing grounded in the same narrative. 5 Reference Prototype The reference framework presented in Section 4 identi- fies the computational stages required to reconstruct and Figure 3: Reference framework for persistent world reconstruction from narrative descriptions. Narrative interpretation ex- tracts structured observations, which are integrated through constrained world completion to reconstruct a persistent world. Spatiotemporal reasoning derives spatial organization and interaction constraints from the reconstructed world, supporting in- teractive realization and world-grounded gameplay. Player interactions and newly introduced observations subsequently update the same persistent world while preserving narrative consistency. CaseEvaluation Scope Demonstrated Capability Chemical Laboratory End-to-end realization Persistentworldrecon- struction, scene realiza- tion, and world-grounded gameplay. The Forgot- ten Shrine Multi- location persistence Persistence across con- nectedlocations,scene transitions, and navigation. LittleRed Riding Hood Existing narrative adaptation Applicability to a narrative adapted from an existing story. Table 2: Prototype cases used to examine complementary aspects of persistent world reconstruction and interactive re- alization. use a persistent world for narrative-grounded interactive experiences. This section presents a reference prototype implementing that framework. Rather than serving as a production-ready game-generation system, the prototype provides one operational realization of the proposed formu- lation, demonstrating how an explicit persistent world can be reconstructed from narrative observations and subsequently support scene realization, gameplay, and interaction updates. The prototype is designed to examine three aspects of the proposed formulation: 1. reconstructing and maintaining a persistent world from narrative observations; 2. instantiating portions of the reconstructed world as playable interactive environments while preserving con- tinuity across events, scenes, and locations; and 3. grounding gameplay mechanics and interaction updates in persistent entities, locations, affordances, semantic re- lationships, and evolving world states. The implementation intentionally simplifies several real- ization components so that the evaluation focuses on the computational role of the persistent world rather than the performance of individual generation modules. Narrative interpretation builds upon structured information extrac- tion (Chen and Jhala 2025b), semantic asset grounding em- ploys GameTileNet (Chen and Jhala 2025a), and spatial layout realization, together with gameplay mechanics, uses lightweight rule-based procedures. The remainder of this section introduces the prototype cases and describes how the Laboratory EntranceWorkbench PreparationMixing and Observation (a) Layout Realization(b) Layout Realization(c) Layout Realization (d) Asset-Grounded Scene(e) Asset-Grounded Scene(f) Asset-Grounded Scene Figure 4: Realization results for the Chemical Laboratory prototype. The top row visualizes tile-based layouts derived from the reconstructed persistent world, and the bottom row shows the corresponding asset-grounded scenes. Persistent entities and evolving world states remain synchronized across the three scenes. reference framework is instantiated in the current implemen- tation. Prototype Cases The reference prototype is demonstrated through three nar- rative cases chosen to exercise complementary aspects of persistent world reconstruction and interactive realization rather than to serve as benchmark datasets. Each case be- gins with a manually designed event specification that is re- alized as a natural-language narrative using OpenAI GPT- 5-mini and a shared prompt template. The prompt template and generated narratives are provided in the Appendix. The three cases emphasize different capabilities of the prototype implementation. Chemical Laboratory serves as the primary running example because it exercises the com- plete pipeline from narrative interpretation and persistent world reconstruction to interactive realization. The For- gotten Shrine focuses on persistence across multiple con- nected locations, including inventory continuity and world- state maintenance during scene transitions. Little Red Riding Hood demonstrates that the same implementation can be ap- plied to a narrative adapted from an existing public-domain story rather than a procedurally authored scenario. Together, these cases demonstrate how the proposed for- mulation can be instantiated through the reference proto- type while exercising different aspects of persistent world reconstruction. Table 2 summarizes the three cases, while additional intermediate representations and realization re- sults are included in the Appendix. Narrative Interpretation The first stage transforms a narrative into structured observa- tions that provide evidence for persistent world construction. It uses a hierarchical representation adapted from prior work on visual narrative understanding (Chen 2025). Because the evaluation narratives are relatively short, the prototype re- tains only the event and event-fragment levels. As summa- rized in Table 3, events correspond to candidate playable sit- uations organized around immediate objectives, while event fragments capture the actions and observations contributing to those objectives. Narrative interpretation is implemented using a prompt- based large language model following the structured extrac- tion strategy of Narrative-to-Scene Generation (Chen and Jhala 2025b). Under a constrained JSON schema, the model extracts the event hierarchy together with entities, locations, semantic relationships, object states, and temporal ordering explicitly supported by the narrative. No additional world knowledge is inferred during this stage. All extracted ob- servations are manually verified before they are incorpo- rated into the persistent world. Tables 3 and 4 summarize the adopted representation and representative interpretation results. Persistent World Construction Narrative observations describe actions and local state changes but do not by themselves constitute a persistent world. The prototype therefore normalizes narrative obser- vations into symbolic world facts and incrementally inte- grates those facts into an evolving world representation. State normalization converts narrative actions into persis- tent facts whenever possible. For example, Scientist entered Representation LevelDescriptionRole in the Prototype NarrativeComplete natural-language story provided as input.Narrative input. Macro-eventHigh-level grouping of related events in longer narratives.Omitted in the current prototype. EventCoherent narrative situation organized around one immediate objective.Candidateplayable scene. Event FragmentIndividual actions or observations contributing toward the event objective.Scene interactions and state changes. Narrative ObservationExplicit entities, relationships, actions, and states extracted from event fragments. Evidence for world construction. Table 3: Hierarchical narrative representation adopted by the reference prototype. CaseEventImmediate ObjectiveRepresentative Event Fragments Chemical Labora- tory Prepare the Exper- iment Gather materials and prepare the workspace. Clear workbench; arrange tools; find flask; collect chemical bottles. Chemical Labora- tory Mix and Observe Reaction Mix chemicals and monitor the reaction. Pour and stir chemicals; observe color change; observe bubbling. Forgotten ShrineFind the Bronze Key Search the watchtower and ob- tain the key. Travel to watchtower; search rooms; find key in chest; keep key. Forgotten ShrineEnter the ShrineUse the key to access the shrine and retrieve the crystal. Reach shrine; clear vines; unlock entrance; retrieve crystal. Little Red Riding Hood Encounterthe Wolf Navigate the encounter and continue toward the cottage. Meet wolf; observe encounter; continue to cottage. Little Red Riding Hood RescueGrand- mother Find grandmother in danger and make the cottage safe. Open cottage door; discover danger; move bedcover; help grandmother; secure room. Table 4: Representative narrative interpretation results. Actions and observations contributing to the same immediate objective are grouped into a single event, providing the basis for persistent world construction and subsequent scene realization. the laboratory becomes the spatial relation Scientist inside Laboratory, while Scientist opened the cabinet changes the cabinet state to Open. Representative examples are shown in Table 5. The normalized facts are accumulated into three comple- mentary structures: a hierarchical event representation pre- serving narrative and temporal organization (Chen 2025), a location graph describing navigable locations and their se- mantic connections, and an entity-state representation main- taining persistent identities, locations, ownership, proper- ties, and evolving states. Together, these structures imple- ment the persistent world used by the prototype. As obser- vations are incorporated, recurring entities retain their iden- tities and previously established states unless they are ex- plicitly modified by later narrative observations or player in- teractions. World Completion Narratives omit many details required for interactive realiza- tion, including environmental structures, spatial context, ob- ject affordances, and interaction constraints. The persistent world must therefore include not only explicitly communi- cated observations but also the contextual structures needed to make those observations interactively realizable. World completion supplies this missing context while remaining grounded in both the narrative evidence and the world facts already reconstructed. The prototype performs three forms of constrained infer- Narrative Obser- vation Normalized World Fact Purpose Scientistentered the laboratory Scientist inside Labora- tory Persistent spatial state Scientistopened the cabinet Cabinet state = open Persistent object state Scientist picked up the flask Scientist holding Flask Track mov- able entities Scientist placed the flask on the work- bench Flask on Work- bench Maintain spatial conti- nuity Table 5: Representative examples of state normalization. Narrative observations are transformed into persistent world facts that maintain entity identity, spatial relationships, and evolving states across events and interactions. ence: (1) interaction-required world knowledge, such as a workbench for chemical mixing or a doorway connecting locations; (2) placement constraints distinguishing fixed and movable entities; and (3) realization hints describing generic environmental characteristics. Representative examples are summarized in Table 6. Candidate completions are produced using a prompt- based large language model conditioned on the recon- structed persistent world. Rather than producing unrestricted InferenceExampleRole Interaction knowl- edge Workbench for chemical mixing Support an interaction required by the narrative. Placement constraint Cabinet= fixed; bottles = movable Guidelayoutwithout prescribing exact coordi- nates. Realization hint Laboratory floor; shrine; forest Guide visual and envi- ronmental realization. Table 6: Representative forms of constrained world comple- tion used by the prototype. Mechanic Required World Infor- mation ExampleUpdated State InspectEntityat- tributes Inspect cabi- net None CollectEntity location and ownership Collect flaskInventory updated UseAffordances andobject relations Mix chemi- cals Entity state updated Transition Task comple- tion and loca- tion graph Enternext location Player loca- tion updated Table 7: Gameplay mechanics implemented in the proto- type and the persistent-world information required to sup- port each interaction. story continuations, the model proposes structured enti- ties, relationships, affordances, environmental context, and placement constraints. All proposed additions are manu- ally verified before they are incorporated into the persistent world. Before interactive realization, the completed persistent world is projected into scene-level representations. Each projection selects the entities, interaction targets, objectives, and placement constraints required for a particular playable scene. Persistent entities and their evolving states are propa- gated across scene projections unless they have been modi- fied by later narrative observations or interactions. This pro- jection preserves global continuity while exposing only the portion of the persistent world required by each scene. Visual and Interactive Realization The scene-level projections produced from the persistent world are instantiated as playable tile-based environments. Unlike the persistent world, which maintains global se- mantic and state information throughout the narrative, each scene representation contains only the entities, interaction targets, placement constraints, and gameplay objectives re- quired for a particular playable situation. This separation de- couples persistent world maintenance from engine-specific realization while preserving continuity across scenes. Spatial realization converts each scene representation into a navigable tile-based layout. Fixed entities are positioned according to inferred placement constraints, while movable entities are initialized from the persistent states propagated during scene projection. Because different scenes refer to the same persistent world, they preserve shared entity identities while reflecting changes in location, ownership, inventory, and object state. Visual realization is performed through semantic as- set grounding with GameTileNet (Chen and Jhala 2025a). Scene entities are matched to candidate tile assets using se- mantic labels and affordances. The prototype retrieves the top-k candidates and applies lightweight manual verification before selecting the final asset for each entity. The realized scenes are assembled into a playable PyGame prototype. Rather than generating gameplay auto- matically, the prototype implements a small set of reusable mechanics that query and update the reconstructed persis- tent world. The current implementation supports object in- spection, item collection, inventory management, object in- teraction, object use and combination, and location transi- tions. These interactions update persistent entity states, loca- tions, and inventory contents, allowing later scenes to remain synchronized through the shared world rather than through independent scene-specific scripts. Figure 4 presents repre- sentative realization results, while Table 7 summarizes the persistent-world information required by each implemented mechanic. Artifact and Reproducibility A public repository accompanies this manuscript. 1 The repository contains the reference prototype implementation together with the narrative cases, prompt templates, inter- mediate representations, and scripts required to reproduce the reported demonstrations. The current implementation uses OpenAI GPT-5-mini to realize manually designed event specifications as natural- language narratives, extract structured narrative observa- tions, and propose constrained world completions. All model-generated artifacts incorporated into the reported pro- totype are manually verified before being integrated into the persistent world. Appendix additionally provide complete intermediate representations and realization results for all three prototype cases. 6 Prototype Validation The proposed formulation is examined through the reference prototype presented in Section 5. Rather than benchmarking a task-specific generation algorithm, the validation examines whether a reference implementation of the formulation can reconstruct an explicit persistent world from narrative ob- servations and use that world to support coherent interactive realization. The three prototype cases exercise complemen- tary aspects of the formulation, including end-to-end recon- struction, persistence across multiple connected locations, and applicability beyond the procedural scenarios used dur- ing prototype development. 1 Repo: https://github.com/RimiChen/2026Narrative2World The Chemical Laboratory case serves as the primary end- to-end demonstration of the reference prototype. Figure 4 presents representative realization results. Narrative obser- vations are reconstructed into a persistent world, from which scene-level representations are projected and instantiated as playable tile-based environments. Throughout the three scenes, entity identities, spatial relationships, inventory con- tents, and object states remain synchronized through the shared persistent world. Player interactions subsequently up- date that same world, allowing later scenes to reflect changes introduced during earlier gameplay. The Forgotten Shrine case examines persistence across multiple connected locations. Rather than emphasizing gameplay complexity, this case demonstrates that persistent entities, inventory contents, player location, and object states remain consistent across navigation and scene transitions through the maintained persistent world. For example, once the bronze key is collected, it remains associated with the player and is subsequently available when interacting with the shrine. Representative intermediate representations and realization results are included in the Appendix. The Little Red Riding Hood case examines whether the same prototype implementation can be applied to narratives adapted from existing stories. Using the same interpretation, reconstruction, world completion, and realization pipeline, the narrative is transformed into a persistent world with- out modifying the underlying framework. Although this case does not establish broad generality across narrative genres, it demonstrates that the proposed formulation is not restricted to procedurally authored scenarios. Together, the three case studies demonstrate that the refer- ence prototype can operationalize the proposed formulation by reconstructing an explicit persistent world that serves as a shared computational foundation for narrative interpretation, constrained world completion, scene realization, gameplay mechanics, and interaction updates. The validation is inten- tionally qualitative. Rather than evaluating visual quality, gameplay sophistication, or generation diversity, it examines whether information reconstructed from narrative observa- tions remains available, coherent, and reusable throughout interactive realization. The objective is therefore to assess the feasibility of the proposed formulation through its refer- ence implementation rather than to optimize or benchmark individual realization components. 7 Discussion The reference prototype was developed to investigate the computational role of persistent worlds in narrative- grounded interactive experiences rather than to optimize individual realization components. Although intentionally lightweight, implementing the prototype provided sev- eral insights into reconstructing and maintaining persistent worlds for interactive realization. First, the prototype intentionally adopts a minimal inter- active realization rather than a visually sophisticated game engine. The objective of this work is not to evaluate graphi- cal fidelity or gameplay complexity, but to examine whether an explicit persistent world can coordinate narrative in- terpretation, scene realization, gameplay interactions, and evolving world states. A lightweight tile-based environment provides sufficient expressive power to exercise persistence, state propagation, interaction updates, and cross-scene con- sistency while minimizing implementation complexity un- related to the proposed formulation. Second, continuity across narrative progression proved to be more fundamental than reconstructing individual obser- vations in isolation. Early prototype versions realized narra- tive events independently, causing entities to disappear be- tween consecutive scenes despite remaining part of the same underlying world. The final prototype, therefore, propagates persistent entities and evolving world states during scene projection, allowing information established by earlier nar- rative observations to remain available until explicitly mod- ified by later narrative events or player interactions. This ex- perience suggests that temporal persistence is a fundamental computational requirement rather than simply a representa- tion choice. Third, the implementation highlights the importance of separating the persistent world from its interactive realiza- tions. The persistent world maintains global information describing entities, locations, semantic relationships, affor- dances, and evolving world states, whereas each scene rep- resentation contains only the subset of information required for a particular interactive situation. This separation allows multiple scenes, layouts, interaction structures, or gameplay experiences to be instantiated from the same reconstructed world while preserving overall narrative consistency. Fourth, implementing constrained world completion re- vealed that interactive realization requires contextual in- formation extending beyond explicit narrative observa- tions. Environmental structures, placement constraints, af- fordances, and navigational context are often necessary to support coherent interaction even though they are not di- rectly described by the narrative. Conditioning world com- pletion on the reconstructed persistent world helps ensure that inferred information remains compatible with previ- ously established entities, relationships, and world states. More broadly, the reference implementation suggests that the persistent world should be regarded as a long-lived com- putational object rather than merely an intermediate repre- sentation generated for a particular downstream task. Nar- rative interpretation reconstructs this world from observa- tions, interactive realization consumes it to instantiate scenes and gameplay, and subsequent player interactions continue to update the same world. The persistent world, therefore, functions as the shared computational state connecting nar- rative understanding with interactive realization throughout the lifetime of an experience. The proposed formulation also opens several directions for future research. At the implementation level, richer world representations, more sophisticated reasoning mech- anisms, and support for long-running interactive narratives would substantially extend the capabilities of the current prototype. More broadly, because the formulation is inde- pendent of any particular implementation strategy, future work may investigate integrating explicit persistent worlds with end-to-end foundation models. Rather than replac- ing explicit world reconstruction, such architectures could use persistent worlds as stable computational references that maintain entities, locations, semantic relationships, and evolving world states throughout the interpretation and in- teraction of extended narratives. 8 Limitations The proposed formulation is examined through a lightweight reference prototype, and several limitations remain. First, the current implementation intentionally simpli- fies several realization components, including rule-based gameplay mechanics, lightweight spatial layout genera- tion, and manual verification during semantic asset ground- ing. These choices isolate the proposed formulation from implementation-specific complexity, but they limit the vi- sual richness, gameplay sophistication, and degree of au- tomation of the resulting experiences. Second, narrative interpretation and constrained world completion currently rely on prompt-based large language models together with manual verification. The reported demonstrations, therefore, establish the feasibility of the proposed formulation rather than fully automatic persis- tent world reconstruction from unrestricted narrative text. More robust information extraction, commonsense reason- ing, provenance tracking, and automatic consistency verifi- cation remain important engineering challenges. Third, the current validation consists of three represen- tative case studies chosen to exercise complementary as- pects of the formulation. Although these demonstrate proof- of-concept feasibility, broader evaluations involving longer narratives, more diverse interactive environments, and richer gameplay experiences will be necessary to characterize scal- ability and generality. Finally, while this work demonstrates that a maintained persistent world can support coherent interactive realization, it does not quantitatively compare this approach against al- ternative generation strategies that reconstruct scenes inde- pendently. Future evaluations comparing continuity, contra- diction rates, state consistency, and downstream interaction quality would provide stronger evidence for the computa- tional benefits of explicit persistent world reconstruction. 9 Conclusion This paper formulated persistent world reconstruction from narrative descriptions as a computational problem for sup- porting narrative-grounded interactive experiences. Rather than treating narratives as direct specifications of scenes, levels, or complete games, the proposed formulation recon- structs and maintains an explicit persistent world that serves as the shared computational object connecting narrative un- derstanding with interactive realization. To investigate this formulation, we presented a reference framework and implemented a reference prototype that re- constructs persistent worlds from narrative observations be- fore instantiating playable tile-based environments. Through three complementary case studies, the prototype demon- strates the feasibility of reconstructing and maintaining per- sistent entities, locations, semantic relationships, and evolv- ing world states while supporting coherent interactive real- ization. More broadly, this work argues that persistent world re- construction should be regarded as a distinct computational problem rather than an implementation detail embedded within individual generation tasks. The reconstructed per- sistent world provides the long-lived computational foun- dation that grounds narrative observations, contextual com- pletion, interactive realization, and subsequent interaction updates in the same evolving world. We hope this for- mulation provides a foundation for future research into richer persistent-world representations, scalable reconstruc- tion methods, long-running interactive narratives, and the integration of explicit persistent worlds with future AI sys- tems. References Agarwal, R.; Lin, Z.; and Riedl, M. 2023. A controllable co-creative agent for game system design. arXiv preprint arXiv:2308.02317. Buongiorno, S.; Klinkert, L.; Zhuang, Z.; Chawla, T.; and Clark, C. 2024. PANGeA: procedural artificial narrative us- ing generative AI for turn-based, role-playing video games. In Proceedings of the AAAI Conference on Artificial Intel- ligence and Interactive Digital Entertainment, volume 20, 156–166. Cardona-Rivera, R. E.; Zagal, J. P.; and Debus, M. S. 2020. Gfi: A formal approach to narrative design and game re- search. In International conference on interactive digital storytelling, 133–148. Springer. Cardona-Rivera, R. E.; Zagal, J. P.; and Debus, M. S. 2022. Game System Models: Toward Semantic Foundations for Technical Game Analysis, Generation, and Design. In Pro- ceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 18, 10–17. Cardona-Rivera, R. E.; Zagal, J. P.; and Debus, M. S. 2023. Aligning story and gameplay through narrative goals. En- tertainment Computing, 47: 100577. Castricato, L.; Biderman, S.; Thue, D.; and Cardona-Rivera, R. 2021. Towards a model-theoretic view of narratives. In Proceedings of the Third Workshop on Narrative Under- standing, 95–104. Chen, Y.-C. 2025.Hierarchical Knowledge Graphs for Story Understanding in Visual Narratives. In International Conference on Interactive Digital Storytelling, 198–218. Springer. Chen, Y.-C.; and Jhala, A. 2024. Collaborative Comic Gen- eration: Integrating Visual Narrative Theories with AI Mod- els for Enhanced Creativity. In Proceedings of the 3rd Work- shop on Artificial Intelligence and Creativity (CREAI), co- located with the 27th European Conference on Artificial In- telligence (ECAI 2024), volume 3810. Chen, Y.-C.; and Jhala, A. 2025a. Gametilenet: A seman- tic dataset for low-resolution game art in procedural con- tent generation. In Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 21, 12–21. Chen, Y.-C.; and Jhala, A. 2025b. Narrative-to-Scene Gen- eration: An LLM-Driven Pipeline for 2D Game Environ- ments. In Proceedings of the Workshops on Experimental AI in Games and Intelligent Narrative Technologies (EXAG- INT 2025) co-located with the 21st AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE 2025), volume 4090. Cook, M. 2022. Puck: a slow and personal automated game designer. In Proceedings of the AAAI Conference on Artifi- cial Intelligence and Interactive Digital Entertainment, vol- ume 18, 232–239. Cook, M. 2025. Game design is generative design. In Pro- ceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 21, 22–31. Farrell, R.; and Ware, S. G. 2024. Large language models as narrative planning search guides. IEEE Transactions on Games, 17(2): 419–428. Fisher, M.; and Ware, S. 2024. A model for automating the abstraction of planning problems in a narrative context. In Proceedings of the AAAI Conference on Artificial Intel- ligence and Interactive Digital Entertainment, volume 20, 35–45. Gallotta, R.; Liapis, A.; and Yannakakis, G. 2024. Consis- tent game content creation via function calling for large lan- guage models. In 2024 IEEE Conference on Games (CoG), 1–4. IEEE. Gallotta, R.; Todd, G.; Zammit, M.; Earle, S.; Liapis, A.; Togelius, J.; and Yannakakis, G. N. 2024. Large language models and games: A survey and roadmap. IEEE Transac- tions on Games. Gerv ́ as, P. 2024. Challenges and opportunities for computa- tional construction of narratives. Digital Society Volume 64, 111. Gerv ́ as, P.; and Calle, J. L. L. 2024. Representing Com- plex Relative Chronology Across Narrative Levels in Movie Plots. In Text2Story@ ECIR, 65–76. G ́ ongora, S.; Chiruzzo, L.; M ́ endez, G.; and Gerv ́ as, P. 2026. World-State Transformations for Neuro-symbolic Interac- tive Storytelling. arXiv preprint arXiv:2605.24719. Guzdial, M.; Snodgrass, S.; and Summerville, A. 2025a. Mixed-initiative PCGML.In Procedural Content Gen- eration via Machine Learning: An Overview, 219–239. Springer. Guzdial, M.; Snodgrass, S.; and Summerville, A. 2025b. PCGML Process Overview. In Procedural Content Genera- tion via Machine Learning: An Overview, 41–56. Springer. Horswill, I. D.; Montfort, N.; and Young, R. M. 2014. Guest editorial: Computational narrative and games. IEEE Trans- actions on Computational Intelligence and AI in Games, 6(2): 93–96. Kumaran, V.; Rowe, J.; Mott, B.; and Lester, J. 2023. Scenecraft: automating interactive narrative scene genera- tion in digital games with large language models. In Pro- ceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 19, 86–96. Lin, Z.; and Riedl, M. 2023. An ontology of co-creative AI systems. arXiv preprint arXiv:2310.07472. Montfort, N.; and y P ́ erez, R. P. 2023. Computational mod- els for understanding narrative. Revista de Comunicac ̧ ̃ ao e Linguagens, (58): 97–117. Nasir, M. U.; James, S.; and Togelius, J. 2024. Word2world: Generating stories and worlds through large language mod- els. arXiv preprint arXiv:2405.06686. Rivera, R. E. C.; Jhala, A.; Porteous, J.; and Young, R. M. 2024. The story so far on narrative planning. In Proceedings of the International Conference on Automated Planning and Scheduling, volume 34, 489–499. Siler, M.; Fisher, M.; and Ware, S. G. 2025. Pareto-based narrative planning making NPCs more rational. In Proceed- ings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 21, 337–346. Summerville, A.; Snodgrass, S.; Guzdial, M.; Holmg ̊ ard, C.; Hoover, A. K.; Isaksen, A.; Nealen, A.; and Togelius, J. 2018. Procedural content generation via machine learning (PCGML). IEEE Transactions on Games, 10(3): 257–270. Ware, S.; Garcia, E. T.; Fisher, M.; Shirvani, A.; and Farrell, R. 2022. Multiagent narrative experience management as story graph pruning. IEEE Transactions on Games, 15(3): 378–387. Ware, S. G.; and Siler, C. 2021. Sabre: A narrative planner supporting intention and deep theory of mind. In Proceed- ings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, volume 17, 99–106. Xie, K.; Yang, I.; Gunerli, J.; and Riedl, M. 2025. Mak- ing large language models into world models with precondi- tion and effect knowledge. In Proceedings of the 31st Inter- national Conference on Computational Linguistics, 7532– 7545. Yannakakis, G. N.; and Togelius, J. 2018. Artificial intelli- gence and games, volume 2. Springer. Yannakakis, G. N.; and Togelius, J. 2025. Procedural Con- tent Generation. In Artificial Intelligence and Games, 235– 251. Springer. Zhou, E.; Basavatia, S.; Siam, M.; Chen, Z.; and Riedl, M. O. 2025.STORY2GAME: Generating (almost) ev- erything in an interactive fiction game.arXiv preprint arXiv:2505.03547. Appendix: Reference Prototype Artifact Companion 1 Overview This appendix serves as an artifact companion for the reference prototype presented in the main paper. It documents the intermediate representations and implementation artifacts produced throughout the prototype, illustrating how nar- rative observations are transformed into a persistent world and subsequently realized as playable tile-based interactive environments. Representative outputs are provided for each stage of the reference framework, including narrative interpretation, persistent world construction, constrained world completion, scene-level representations, and interactive realization. Unless otherwise noted, examples are taken from the Chemical Laboratory running example used throughout the main paper. Complete JSON outputs, prompt templates, implementation scripts, and realization assets are provided in the accompanying project repository. 2 End-to-End Pipeline Walkthrough Table 1 summarizes the main prototype stages and the representation produced by each stage. StageOutputPurpose Prompt AStoryText JSONRealize an author design note as a short natural-language narrative. Prompt BNarrativeStructure JSONExtract events, fragments, observations, entity-relation triples, and state changes. Prompt CWorldComplete JSONNormalize observations and infer necessary world knowl- edge for interaction. Step 3ARealizationPlan JSONSimplify the completed world into renderable entities and gameplay tasks. Step 3BSceneRepresentation JSONBuild per-scene structures for layout, interaction, and asset grounding. Steps 4–6Assetmapping,layouts, PyGame prototype Ground entities in game assets, generate tile layouts, and run the playable prototype. Table 1: Prototype stages and intermediate representations. 2.1 Prompt A Output: Narrative Realization The following story was generated from the Chemical Laboratory design note. It is intentionally short, event-oriented, and concrete enough for downstream interpretation. 1 Chemical Laboratory Story Design Note ## Scenario A short procedural story inside a chemistry laboratory. ## Narrative Outline • Enter the laboratory. • Prepare the experiment. • Find a flask. • Collect the required chemicals. • Mix the chemicals. • Observe the reaction. • Record the result. • Leave the laboratory. Chemical Laboratory Story She entered the laboratory and closed the door behind her, letting the steady hum of lights and instruments settle into the background. After a moment she prepared the experiment, clearing the workbench and arranging glassware, a stirring rod, and a notebook where she could record observations. She found a flask tucked on a lower shelf in the cabinet and carried it to the workbench, setting it down among the other items. From the nearby shelves she collected the required chemical bottles, bringing each onto the bench and lining them up within reach. Carefully she mixed the chemicals in the flask, pouring and stirring until the contents combined. The mixture responded with a slow color change and a gentle bubbling; she observed the reaction closely as it reached a steady state. She recorded the result in the notebook, noting the visible changes and the final appearance of the flask. Satisfied, she returned the bottles and flask to their places and left the laboratory through the door. 2.2 Prompt B Output: Narrative Interpretation Prompt B converts the story into a hierarchical event representation. Table 2 shows the event-level interpretation for the Chemical Laboratory case. Table 2: Chemical Laboratory event hierarchy. IDEventImmediate ObjectiveFragments E1Enter LaboratoryEnter the laboratory and close the door behind her entered the laboratory; closed the door behind her; The lights and instruments hummed in the back- ground as she entered E2Prepare the ExperimentClear and arrange the workbench and gather necessary glassware, tools, and materials for the ex- periment cleared the workbench; arranged glassware, a stirring rod, and a notebook on the workbench; found a flask tucked on a lower shelf in the cabinet; carried the flask to the workbench and set it down among the other items; collected the re- quired chemical bottles from the nearby shelves and brought each onto the bench; lined the chemi- cal bottles up within reach on the bench 2 IDEventImmediate ObjectiveFragments E3Mix and Observe Reac- tion Mix the chemicals in the flask and monitor the reaction until it reaches a steady state carefully mixed the chemicals in the flask, pouring and stirring un- til the contents combined; The mix- ture responded with a slow color change and a gentle bubbling; ob- served the reaction closely as it reached a steady state E4Record Results and Clean Up Record observations in the note- book, return items to their places, and leave the laboratory recorded the result in the notebook, noting visible changes and the final appearance of the flask; returned the bottles and flask to their places; left the laboratory through the door 2.3 Prompt C Output: Persistent World Completion Prompt C converts explicit narrative observations into persistent world facts and infers only the additional knowledge required for interaction, layout, state tracking, and transitions. The complete output is stored in WorldComplete.json; Listing 1 shows a compact excerpt. Listing 1: Representative excerpt from a completed persistent world. "persistent_entities": [ "id": "protagonist", "type": "character", "location": "laboratory", "id": "laboratory", "type": "location", "id": "door", "type": "transition", "state": "closed", "id": "workbench", "type": "surface", "placement": "fixed", "id": "cabinet", "type": "container", "placement": "fixed", "id": "flask", "type": "item", "state": "movable", "id": "chemical_bottles", "type": "item", "state": "movable" ], "normalized_facts": [ ["protagonist", "inside", "laboratory"], ["door", "state", "closed"], ["flask", "located_in", "cabinet"], ["flask", "moves_to", "workbench"], ["chemical_bottles", "moves_to", "workbench"] ], "inferred_for_realization": [ "entity": "floor", "reason": "required for tile layout", "entity": "interaction_zone", "reason": "required for object use" ] 2.4 Step 3A Output: Realization Plan The realization plan simplifies the rich persistent world into concrete renderable entities and task structures. This stage deliberately avoids building a dense ontology; it keeps only the information required for tile-based realization. Listing 2: Representative realization plan excerpt. "renderable_entities": [ "id": "protagonist", "category": "player", "drawable": true, "id": "door", "category": "transition", "fixed": true, "id": "workbench", "category": "surface", "fixed": true, "id": "cabinet", "category": "container", "fixed": true, "id": "flask", "category": "collectible", "dynamic": true, "id": "chemical_bottles", "category": "collectible", "dynamic": true 3 ], "gameplay_tasks": [ "event": "E2", "mechanic": "collect", "target": "flask", "event": "E2", "mechanic": "collect", "target": "chemical_bottles", "event": "E3", "mechanic": "use_combine", "target": "flask" ] 2.5 Step 3B Output: Scene Representation The scene representation organizes the renderable and interactive information required for each playable scene. The JSON excerpt below illustrates the structure used by downstream layout and PyGame scripts. Listing 3: Representative scene representation excerpt. "scene_id": "L2", "based_on_event": "E2", "scene_goal": "Prepare the workbench and gather required materials", "environment": "location": "laboratory", "terrain": "laboratory_floor", "characters": ["protagonist"], "fixed_objects": ["workbench", "cabinet", "door"], "interactive_items": ["flask", "chemical_bottles", "notebook"], "spatial_relations": [ ["flask", "inside", "cabinet"], ["chemical_bottles", "on", "shelf"], ["notebook", "on", "workbench"] ], "tasks": [ "action": "collect", "target": "flask", "action": "collect", "target": "chemical_bottles", "action": "place", "target": "flask", "destination": "workbench" ] 2.6 Final Realization Chemical Laboratory Laboratory EntranceWorkbench PreparationMixing and Observation (a) Layout Realization(b) Layout Realization(c) Layout Realization (d) Asset-grounded Scene(e) Asset-grounded Scene(f) Asset-grounded Scene 4 3 Prompt Templates This section records the main prompts used by the reference prototype. The full prompt files, including JSON schema details, are included in the repository. 3.1 Prompt A: Narrative Realization Prompt A Summary Role: Narrative realization. Convert a lightweight author design note into a short natural-language narrative. The narrative should preserve the intended outline, avoid dropping specified content, and remain suitable for subsequent structured interpretation. Prompt A Template Your role in this step is NARRATIVE REALIZATION: convert a lightweight author design note into a short, natural-language narrative. The narrative will later be analyzed by another component, so your only task is to write a coherent story paragraph that preserves the intended outline. Follow the provided structure specifications and do not omit elements from the narrative outline. 3.2 Prompt B: Narrative Interpretation Prompt B Summary Input: a natural-language narrative. Output: a NarrativeStructure JSON identifying playable narrative events, event fragments, explicit observations, entity-relation triples, and state changes. Prompt B Template A narrative event is not an atomic action. A narrative event is a coherent narrative situation with one immediate objective that can reasonably be realized as one playable scene or level. Event fragments are concrete actions or observations that contribute to completing that event. Group fragments by the immediate objective they support, not by sentence boundaries and not by action size. 3.3 Prompt C: Persistent World Completion Prompt C Summary Input: NarrativeStructure.json. Output: a completed semantic world description preserving explicit observations while adding only the knowledge needed for layout, interaction, state tracking, and transitions. Prompt C Constraints 1. Preserve explicit observations from the narrative. 2. Normalize non-visual action relations into visualizable states when possible. 3. Infer only necessary missing world information. 4. Do not invent unrelated objects, plot events, rooms, hazards, puzzles, dialogue, enemies, or story goals. 5. Separate explicit information from inferred information. 6. Use semantic placement constraints rather than coordinates. 7. Distinguish fixed, flexible, dynamic, and optional placement. 8. Allow generic terrain, floor, and wall hints when needed for visual realization. 9. Maintain enough information to build hierarchical event structures, a location graph, an entity-state table, scene specifications, and gameplay task hints. 5 3.4 Step 3A: Realization Planning Step 3A Summary Convert WorldComplete.json into a stable game-facing RealizationPlan.json. The plan should be compact, implementation-friendly, and directly useful for tile-based game realization. Step 3A Design Principles Do not build a dense ontology or full knowledge graph. Keep renderable objects concrete and drawable. Separate renderable entities from abstract states, relation targets, ambient cues, and narrative notes. Collapse subparts into parent entities when they are not independently drawable. Separate scene representation from gameplay tasks. Link each gameplay task to its event, fragment, mechanic, target entity, and state update. 3.5 Step 3B: Scene Representation Step 3B Summary Build simple per-scene structures from WorldComplete.json and the optional realization plan. Each scene records summaries, scene triples, spatial relations, character lists, interactive objects, propagated environment objects, and affordance tables. 4 Prototype Case Stories 4.1 Chemical Laboratory Story Text She entered the laboratory and closed the door behind her, letting the steady hum of lights and instruments settle into the background. After a moment she prepared the experiment, clearing the workbench and arranging glassware, a stirring rod, and a notebook where she could record observations. She found a flask tucked on a lower shelf in the cabinet and carried it to the workbench, setting it down among the other items. From the nearby shelves she collected the required chemical bottles, bringing each onto the bench and lining them up within reach. Carefully she mixed the chemicals in the flask, pouring and stirring until the contents combined. The mixture responded with a slow color change and a gentle bubbling; she observed the reaction closely as it reached a steady state. She recorded the result in the notebook, noting the visible changes and the final appearance of the flask. Satisfied, she returned the bottles and flask to their places and left the laboratory through the door. 4.2 The Forgotten Shrine Story Text They arrive at the mountain village as afternoon light spills between narrow lanes and stacked cottages. There they meet the village elder, a quiet figure who leads them to a low bench and, after a moment of tired consideration, explains that the sacred crystal is hidden in the forgotten shrine beyond the ridge. Setting out with a simple pack, the traveler travels to the abandoned watchtower where wind has stripped banners from its timbers and moss grows over the steps. Inside the tower they search the rooms and find the bronze key tucked into a cracked wooden chest. With the bronze key kept safe at their belt, they go to the forgotten shrine, skirt past collapsed columns and push aside a curtain of vines. At the stone doorway they use the bronze key to open the shrine entrance, stepping into a quiet chamber to retrieve the sacred crystal. With the relic in hand, they return to the village to hand it back to the elder. 6 4.3 Little Red Riding Hood Story Text Little Red Riding Hood packed a woven basket for her grandmother, set it on her arm, and left home with the basket for grandmother. She followed the well-worn lane and traveled through the forest, where shafts of light and rustling leaves marked her progress. Along the way she met the wolf at a shaded fork; after a brief, watchful encounter he slipped away and she continued to grandmother’s cottage. At the cottage door she pushed it open and discovered grandmother was in danger: the wolf had taken shelter in the bed and the old woman was hidden and frightened. Little Red Riding Hood moved the bedcover, helped her grandmother from the bed, and closed the door, changing the room from chaotic to calm as they checked the basket and straightened the chairs. Having rescued her grandmother and made the cottage safe, she gathered her cloak and, hand in hand, they walked back down the lane and returned home safely. 5 Complete Event Structures The following tables summarize the complete event structures generated by Prompt B for all three prototype cases. These tables present the event titles, immediate objectives, and event fragments used as the basis for subsequent world construction and scene realization. 5.1 Chemical Laboratory Table 3: Complete event structure for Chemical Laboratory. IDEventObjectiveEvent Fragments E1Enter LaboratoryEnter the laboratory and close the door behind her entered the laboratory; closed the door behind her; The lights and in- struments hummed in the background as she entered E2Prepare the ExperimentClear and arrange the workbench and gather necessary glassware, tools, and materials for the ex- periment cleared the workbench;arranged glassware, a stirring rod, and a note- book on the workbench; found a flask tucked on a lower shelf in the cab- inet; carried the flask to the work- bench and set it down among the other items; collected the required chemical bottles from the nearby shelves and brought each onto the bench; lined the chemical bottles up within reach on the bench E3Mix and Observe Reac- tion Mix the chemicals in the flask and monitor the reaction until it reaches a steady state carefully mixed the chemicals in the flask, pouring and stirring until the contents combined; The mixture re- sponded with a slow color change and a gentle bubbling; observed the re- action closely as it reached a steady state 7 IDEventObjectiveEvent Fragments E4RecordResultsand Clean Up Record observations in the note- book, return items to their places, and leave the laboratory recorded the result in the notebook, noting visible changes and the final appearance of the flask; returned the bottles and flask to their places; left the laboratory through the door 5.2 The Forgotten Shrine Table 4: Complete event structure for The Forgotten Shrine. IDEventObjectiveEvent Fragments E1Meet the Village ElderLearn where the sacred crystal is hidden Arrive at the mountain village while afternoon light spills between narrow lanes and stacked cottages (observa- tion of setting); Meet the village el- der and are led to a low bench; Elder explains that the sacred crystal is hid- den in the forgotten shrine beyond the ridge E2Find the Bronze Key in the Watchtower Search the abandoned watch- tower and obtain the bronze key Set out from the village carrying a simple pack; Travel to the abandoned watchtower and observe that wind has stripped banners from its timbers and moss grows over the steps; Search the rooms inside the tower; Find the bronze key tucked into a cracked wooden chest; Keep the bronze key safe at the traveler’s belt (take posses- sion) E3EntertheForgotten Shrine and Retrieve the Crystal Use the bronze key to open the shrine and retrieve the sacred crystal Go to the forgotten shrine beyond the ridge; Skirt past collapsed columns; Push aside a curtain of vines; Use the bronze key at the stone doorway to open the shrine entrance; Step into a quiet chamber and retrieve the sacred crystal E4Return the Relic to the Elder Bring the sacred crystal back to the village elder Return to the village carrying the relic (sacred crystal); Hand the sacred crystal back to the village elder 5.3 Little Red Riding Hood 8 Table 5: Complete event structure for Little Red Riding Hood. IDEventObjectiveEvent Fragments E1Prepare basket and de- part home Pack a basket for grandmother and leave home carrying it. Packed a woven basket for her grand- mother; Set the woven basket on her arm; Left home carrying the basket intended for grandmother E2Travelalonglane through the forest Journeyfromhometoward grandmother’s cottage via the lane and forest. Followed the well-worn lane; Trav- eled through the forest; Noted en- vironmental details (shafts of light and rustling leaves) as markers of progress E3Encounter the wolf at a fork Navigate a brief encounter with the wolf at a shaded fork and continue toward the cottage. Met the wolf at a shaded fork; Ex- perienced a brief, watchful encounter with the wolf; The wolf slipped away and she continued toward the cottage E4Discoverandrescue grandmother; secure the cottage Find grandmother in danger, re- move her from harm, and make the cottage calm and safe. Pushed open the cottage door and en- tered; Discovered that grandmother was in danger and that the wolf had taken shelter in the bed while the old woman was hidden and frightened; Moved the bedcover (revealing/help- ing free whoever was hidden); Helped her grandmother from the bed; Closed the door, contributing to changing the room from chaotic to calm; Checked the basket and straightened the chairs as part of making the cottage calm E5Return home togetherLeave the cottage with grand- mother and return home safely. Gathered her cloak; Walked hand in hand with her grandmother down the lane; Returned home safely 6 Level Candidates and Interactive Tasks The following tables summarize the candidate playable levels generated from the event hierarchy. These level candi- dates are later projected into scene representations and task-driven gameplay. 6.1 Chemical Laboratory Table 6: Level candidates and possible player tasks for Chemical Laboratory. LevelEventGoalLocationPossible Tasks L1E1Enter the laboratory and secure the entrance laboratory entrance / door approach and enter the lab; operate the door to close it; listen to or observe en- vironmental hum (lights/in- struments) 9 LevelEventGoalLocationPossible Tasks L2E2Prepare the workbench with re- quired tools and gather flask and chemical bottles workbench and nearby storage (cabinet and shelves) clear the workbench; place glassware and tools;lo- cate flask on a lower shelf; carry flask to bench; col- lect chemical bottles from shelves;line up bottles within reach L3E3Mix the chemicals in the flask and monitor the reaction until it stabilizes workbench (flask)pour specified chemicals into the flask; stir carefully untilcontentscombine; observe color change and bubbling; wait until reaction reaches steady state L4E4Record observations, return ma- terials, and exit the laboratory workbench;cabi- net/shelves; laboratory door write results in the note- book; return bottles and flask to storage locations; exit through the laboratory door 6.2 The Forgotten Shrine Table 7: Level candidates and possible player tasks for The Forgotten Shrine. LevelEventGoalLocationPossible Tasks L1E1Learn the location of the sacred crystal from the village elder mountain_villageExplore the village arrival area; Find and approach the village elder; Converse with the elder to receive infor- mation about the forgotten shrine L2E2Search the abandoned watch- tower and retrieve the bronze key abandoned_watchtowerApproach and examine the watchtower exterior;En- ter and search the tower rooms; Open or inspect a cracked wooden chest to find the bronze key; Equip the bronze key at the belt L3E3Reach the forgotten shrine, open it with the bronze key, and re- trieve the sacred crystal forgotten_shrineApproach the shrine area and navigate past collapsed columns;Clear or push aside vines blocking access; Use the bronze key at the stone doorway to open the shrine entrance; Enter the quiet chamber and take the sacred crystal 10 LevelEventGoalLocationPossible Tasks L4E4Return the sacred crystal to the village elder mountain_villageTravel back to the village carrying the relic; Locate the elder and hand over the sacred crystal 6.3 Little Red Riding Hood Table 8: Level candidates and possible player tasks for Little Red Riding Hood. LevelEventGoalLocationPossible Tasks L1E1Pack the basket for grandmother and depart home carrying it. homeSelect or pack items into the woven basket (content unspecified by text); Place basket on character’s arm; Leave the house through the exit door L2E2Travel along the well-worn lane through the forest toward the cot- tage. well-worn lane / forestNavigate the lane and for- est path; Observe environ- mental cues (shafts of light, rustling leaves) L3E3Handle a brief encounter with the wolf at a shaded fork and con- tinue to the cottage. shaded forkApproach or avoid the wolf at the fork; React to the wolf’s watchful behavior; Continue on the chosen path toward the cottage after the encounter L4E4Discover grandmother’s danger, remove her from the bed, and se- cure the cottage to change it from chaotic to calm. grandmother’s cottage (interior / bed / door) Open the cottage door and inspect the room;Iden- tify threat (wolf in bed) and reveal/hide elements by moving the bedcover; As- sist grandmother off the bed and close/secure the door; Check the basket and straighten chairs to calm the room L5E5Leave the cottage together and return home safely. lane; homeEquip cloak; Escort grand- mother along the lane (walk hand in hand); Reach home safely 11 7 Prototype Interface Chemical Laboratory Prototype A screenshot showing the PyGame interface, including the game viewport, inventory, current task hint, and interaction messages. 8 Repository Organization The repository is organized around pipeline stages so that intermediate outputs can be inspected or manually revised independently. Listing 4: Repository organization. Repository/ StoryCases/ case<id>_DesignNote_Json.json case<id>_StoryText.json case<id>_NarrativeStructure.json case<id>_WorldComplete.json case<id>_RealizationPlan.json case<id>_SceneRepresentation.json case<id>_AssetMapping.json case<id>_SceneLayout.json Prompts/ PromptA_NarrativeRealization.txt PromptB_NarrativeInterpretation.txt PromptC_WorldCompletion.txt Step3A_RealizationPlan.txt Step3B_SceneRepresentation.txt Scripts/ Step0_StoryRealization.py Step1_NarrativeStructure.py Step2_WorldCompletion.py Step3A_make_realization_plan.py Step3B_make_scene_representation.py Step4_asset_mapping.py Step5_make_scene_layout.py 12 Step6_play_pygame_task_driven.py Data/GameTile/Assets/ 9 Reproducibility Notes To reproduce the prototype cases, run the scripts in stage order for each case identifier. The prompt-based stages produce JSON files that can be inspected and manually verified before subsequent processing. The asset mapping file is intentionally simple so that inappropriate visual matches can be replaced manually without rerunning earlier narrative or world-construction stages. The final PyGame script loads the scene layout, asset mapping, task structure, and persistent state information to instantiate the playable prototype. 13