The SymptomWise framework restricts LLMs to symptom extraction, delegating diagnostic reasoning to a deterministic codex. This architecture prevents hallucinations by mapping free-text inputs to a finite hypothesis space of expert-curated medical knowledge. The approach ensures traceability in safety-critical settings. Practitioners can now deploy more reliable differential diagnoses without relying on generative guesswork.