Is your feature request related to a problem? Please describe.
Yes, the current ScenarioSpec-based architecture is overly complex and rigid. The JSON schema with actions, preconditions, effects, random events, and JSONLogic conditions creates a high barrier for both LLMs to generate valid specs and users to understand/modify scenarios. The Monte Carlo simulations add unnecessary computational overhead and complexity without clear benefits for creative storytelling. This makes the system difficult to use and limits creative freedom.
Describe the solution you'd like
I'd like to completely redesign the scenario generation and engine architecture around a simpler, more narrative-driven approach:
Remove ScenarioSpec entirely and replace with:
World Background Generation: Generate rich, descriptive world lore and setting from the user's scenario description
Character Memory System: Create detailed backstories, personalities, and memories for each character/entity
Dynamic Narrative Engine: Use LLM-powered turn-by-turn storytelling without rigid rule systems
Key Changes:
Simplified Generation: Instead of generating complex JSON specs, generate:
World background text
Character profiles with memories and traits
Basic scenario metadata (name, themes, difficulty hints)
Remove Monte Carlo Simulations: No need for statistical validation - let creative scenarios exist without artificial difficulty constraints
Flexible Engine: The orchestrator would use the world background and character memories as context for LLM narrative generation, allowing more organic storytelling
Backward Compatibility: Keep the same API endpoints but change the internal data structures. This might have to also have frontend changes in the api response so I need to make sure that still works.
Describe alternatives you've considered
No response
Additional context
No response
Would you like to contribute this feature?
Is your feature request related to a problem? Please describe.
Yes, the current ScenarioSpec-based architecture is overly complex and rigid. The JSON schema with actions, preconditions, effects, random events, and JSONLogic conditions creates a high barrier for both LLMs to generate valid specs and users to understand/modify scenarios. The Monte Carlo simulations add unnecessary computational overhead and complexity without clear benefits for creative storytelling. This makes the system difficult to use and limits creative freedom.
Describe the solution you'd like
I'd like to completely redesign the scenario generation and engine architecture around a simpler, more narrative-driven approach:
Remove ScenarioSpec entirely and replace with:
World Background Generation: Generate rich, descriptive world lore and setting from the user's scenario description
Character Memory System: Create detailed backstories, personalities, and memories for each character/entity
Dynamic Narrative Engine: Use LLM-powered turn-by-turn storytelling without rigid rule systems
Key Changes:
Simplified Generation: Instead of generating complex JSON specs, generate:
World background text
Character profiles with memories and traits
Basic scenario metadata (name, themes, difficulty hints)
Remove Monte Carlo Simulations: No need for statistical validation - let creative scenarios exist without artificial difficulty constraints
Flexible Engine: The orchestrator would use the world background and character memories as context for LLM narrative generation, allowing more organic storytelling
Backward Compatibility: Keep the same API endpoints but change the internal data structures. This might have to also have frontend changes in the api response so I need to make sure that still works.
Describe alternatives you've considered
No response
Additional context
No response
Would you like to contribute this feature?