Principlev1
Distinguish between increasing complexity that improves
Distinguish between increasing complexity that improves explanatory power and complexity that merely preserves a failing model, because the latter signals the need for architectural replacement rather than parameter tuning.
Why This Is a Principle
Grounded in Abstraction Necessarily Discards Information (abstraction discards information), Simple decision rules using less information can outperform (simple rules outperform complex models in uncertainty), and The performance of an agent is bounded by the accuracy of (performance bounded by model accuracy). The principle prescribes evaluating complexity against predictive improvement to detect when optimization has hit architectural limits.