Reading JSON and the many faces of missing

Most web data arrives as JSON, a text format of nested objects and arrays that maps cleanly onto Python's dicts and lists:

import json
with open("data/weather.json", encoding="utf-8") as f:
    doc = json.load(f)          # a tree of dicts and lists
temp = doc.get("temp_c")        # None if the key is absent

Notice .get(key) rather than doc["key"]: .get returns None for a missing key instead of crashing, which is the safer way to read data you didn't create. JSON's null becomes Python's None, an array becomes a list, an object becomes a dict — so reading nested JSON is really just walking dicts and lists.

The subtle part of real data is that the token for "missing" changes as a value moves through your stack, and each form has its own trap:

The headline trap: a single NaN turns an ordinary mean into nan. Use np.nanmean (or pandas' skip-na behaviour) when you mean "average the values that are present".
leads into “Handling missing data and validating input”.*