Thanks for checking out Theus! I’m currently at a crossroads regarding one specific feature and would love to hear your thoughts.
In Theus, the default behavior is Full Transactional Integrity—every mutation happens on a 'Shadow Copy' so we can rollback instantly if an Audit Rule is violated. This is great for safety but can be expensive for high-frequency loops like Reinforcement Learning or processing large Tensors.
To solve this, I’ve implemented a strict_mode=False toggle. When disabled:
Shadow Copying is bypassed: Reading/Writing happens directly on the real object.
Zero Overhead: No transaction objects or audit logs are created.
Trade-off: You lose all safety—no rollbacks, no contract enforcement, and crashes leave the state 'dirty'.
My dilemma: Is providing a 'Strict Mode Toggle' a pragmatic necessity for performance, or does it defeat the entire purpose of a framework built for safety?
Should I keep this global toggle, or should I force developers to use more granular optimizations (like my heavy_ prefix for specific large assets) to keep the 'Safety-First' philosophy intact?
I'd appreciate any architectural insights from those who have built similar state-heavy systems!