Hi - I'm getting many NOT_PARSABLEs using the QA eval template. Any idea? I don't see any structural differences between when it works and when it doesn't?
My input is similar to this with more specific data about the patient as text: You are a physician specializing in obesity and GLP-1 therapies (such as Ozempic, Mounjaro, and Wegovy). Your task is to generate two types of personalized comments FOR THE PATIENT: SUMMARY COMMENT (2-4 sentences): Direct and to the point, like an experienced physician. Mention the main finding and a specific action if applicable. Empathetic yet realistic. DETAILED COMMENT (7-10 sentences): A CONTINUATION of the summary, providing more in-depth information. Style: A physician who explains directly but with warmth. Give SPECIFIC and ACTIONABLE recommendations (no generic phrases). Be realistic: acknowledge when something isn't right and suggest concrete adjustments. <detailed patient raw data>
sample output: { "summaryComment": "We see good weight and waist reduction in these first few weeks. This indicates a favorable body composition, with fat loss. Keep it up!", "detailedComment": "Excellent! In these first few weeks of treatment with Wegovy, you have achieved good weight loss and, most importantly, a decrease in your waist circumference. This suggests that you are losing fat, which is the main goal. GLP-1 helps you better follow your eating plan and control your appetite. Remember that your dose will gradually increase each month, and it is normal to experience some mild side effects, especially after each increase. Stay well hydrated, eat slowly, and prioritize appropriate portions of nutritious foods. Let's keep going!", "nextBestAction": "continue_monitoring", "actionMetadata": { "suggestedDate": null, "priority": "low", "churnRisk": "low", "reasoningNotes": "Good initial progress, patient consistent with measurements and no significant side effects. Low risk of discontinuation." Confidence: 0.95 }
