Sanofi's clinical-data ingestion pipeline receives nested payloads containing patients and their observations. Write a function that transforms the payload into a normalized list, keeps only the highest version of duplicate observations, and returns records in a deterministic order.
An observation is identified by the combination of patient_id, code, and collected_at. If multiple observations share that key, retain the one with the greatest version.
Implement transform_observations(payload). The input is a dictionary with a patients key whose value is a list of patient dictionaries. Each patient contains a string patient_id and an observations list. Each observation contains code, value, unit, collected_at, and integer version fields.
Return a list of dictionaries. Each output dictionary must contain exactly patient_id, code, value, unit, and collected_at, excluding version. Sort the result by patient_id, then by collected_at, then by code, all in ascending order.
Example 1
Input: {"patients": [{"patient_id": "P2", "observations": [{"code": "TEMP", "value": 37.1, "unit": "C", "collected_at": "2026-01-02", "version": 1}]}]}
Output: [{"patient_id": "P2", "code": "TEMP", "value": 37.1, "unit": "C", "collected_at": "2026-01-02"}]
The output removes version while preserving the observation data.
Example 2
Two observations with the same identity and versions 1 and 3 produce only the version 3 record.
def transform_observations(payload):