Why does Python json.dumps emit NaN that another parser rejects?

Python’s JSON encoder allows non-finite floating-point values by default, although they are outside the strict JSON number syntax. Other implementations may reject them. Set allow_nan=False to fail explicitly, then decide whether the data contract should reject, omit or replace non-finite values. Test float("nan"), positive infinity and a regular finite number. Do not replace the text NaN across serialized output: that can alter ordinary string values. A successful serialization does not validate an API schema, and converting every invalid measurement to zero changes its meaning. Preserve the distinction between unavailable data and a measured zero. Minimal example:

import json
try:
    json.dumps({"value": float("nan")}, allow_nan=False)
except ValueError:
    pass
else:
    raise AssertionError("Non-finite number was accepted")
assert json.loads(json.dumps({"value": 1.5}, allow_nan=False)) == {"value": 1.5}