Semantic Mapping: Breaks input into core concepts, entities, and intent. Constraint Identification: Finds explicit and implicit constraints. Relevance Scoring: Determines essential vs. tangential information.
Mode Selection: Routes to BUILD, EXPLORE, or HARDEN. Constraint-Subtraction: Prunes unnecessary data. Entropy Injection: Prevents overly deterministic output.
Constraint-Aware Generation: Output adheres to the optimized path. Self-Correction Loop (CEGAR): Feedback loop checks output against constraints — failures inject new constraints into the next pass.