Defines ~100+ effect constants (e.g., extra_battle_strength_points, extra_breakthrough_success_rate) with typed values, merge rules, and documentation. This centralized effect registry is a decoupled, reusable pattern for any attribute-based game or system where effects modify numeric/list properties from multiple sources (techniques, roots, personas, equipment). It demonstrates non-obvious skill in designing a maintainable effect catalog that avoids magic strings across the codebase.

Provides base classes (DefineAction, TimedAction, InstantAction, ActualActionMixin) and a cooling decorator for action lifecycle management. Each action encapsulates start/execute/finish phases with cooldown support and parameter validation. This pattern is self-contained and could be extracted for any turn-based or timed action framework (e.g., RPG skills, job scheduling) with minimal adaptation. The registry in __init__.py allows discoverability.

Implements a generic LLM-driven single-choice decision engine with fallback to deterministic logic when LLM fails. Supports accept/reject scenarios, item exchange, and sect recruitment. The engine is decoupled from game-specific models and can be reused in any project requiring structured LLM decisions with recovery strategies.

Defines the ordered pipeline of 29 phases (lifecycle, actions, social, world, etc.) executed each month in the simulation. Each phase is a callable handler registered with a name, and the PhaseRunner orchestrates execution with hooks for before/after. This pattern is a clean, extensible pipeline architecture applicable to any iterative simulation or batch processing system.