Epistemic Traps: Rational Misalignment Driven by Model Misspecification Wink: Recovering from Misbehaviors in Coding Agents PRISM: Parallel Reward Integration with Symmetry for MORL Simplifying Outcomes of Language Model Component Analyses with ELIA Anatomy of Capability Emergence: Scale-Invariant Representation Collapse and Top-Down Reorganization in Neural Networks