Update test.py

test.py

@@ -1,3 +1,29 @@
+
+You are an expert oncology clinical data curator. Read one patient’s clinical notes and produce a single JSON object using the schema below. Your main goal is to (1) extract all dated progression/response assessments and related details, and (2) apply the secondary-cancer-type aggregation rule described here.
+
+AGGREGATION RULE (30-day window, same evidence source)
+- Consider assessments that occur within 30 calendar days of each other AND share the same evidence_source (e.g., imaging, physician).
+- Collapse those assessments into one event and update fields as follows:
+  • secondary_cancer_type → set to a **comma-separated list of unique values** observed across the collapsed assessments (leave "" if none are stated).
+  • date_of_disease_progression_assessment → use the **latest date** among the collapsed assessments.
+  • disease_progression_status → if multiple statuses appear, keep the most definitive using this priority:
+      progression > complete response > partial response > stable disease > no evidence of disease > indeterminate.
+  • treatment_change → if present in any collapsed assessment, keep the details tied to the **latest date**; if conflicting across notes, prefer the latest.
+  • evidence_for_disease_progression_assessment / exact_evidence_span_for_disease_progression_assessment → use the clearest/most definitive phrasing from the latest assessment.
+
+DATE & VALUE RULES
+- Use ISO-8601 where available: YYYY-MM-DD; if only month is known use YYYY-MM; if only year is known use YYYY.
+- If a note says “today” and the note date is known, resolve it to that date; otherwise leave the date field as "".
+- Keep drug/regimen names verbatim.
+- Do not guess. If a field is not stated, set it to "".
+
+OUTPUT
+- Return **only** valid JSON (no prose, no Markdown fences).
+- Keep field order and names exactly as in the schema.
+- All string fields must be strings; if unknown, use "".
+
+
+
 import pandas as pd
 import re