Separate missing, invalid and zero values

Missing, invalid and zero answer different questions. A blank amount may mean no value was supplied. “bad” is malformed. $0.00 is a legitimate value. A refund such as-$1.25 is negative but valid under the existing currency grammar.

A nullable integer column can represent known cents and pd.NA, but it cannot explain why a value is absent. Keep a separate status column. The final audit still rejects a missing required sales amount; this exercise shows how to retain the distinction before applying that business policy.

The supplied currency parser uses a documented grammar and Decimal-to-integer conversion. The classifier accepts its exact cents only when representable in this lesson’s nullable Int64 schema. Out-of-range data is invalid here, not rounded.

The final report has a different, deliberate representation: its previous records policy allows arbitrarily large Python integer cents, so that capstone uses object storage for exact Python ints. A dtype is an application decision, not a rule that every monetary table must use the same representation.

import pandas as pd
values=pd.Series(pd.array([0,None,29],dtype="Int64"))
print(values.isna().tolist())
print(values.eq(0).fillna(False).tolist())

errors=coerce can be useful only when the lost distinctions are accounted for elsewhere. Here the status column is the audit evidence: missing and invalid both have pd.NA cents, but their causes remain different. Zero is never used as a catch-all replacement.

The tests include nondefault and duplicate index labels. Constructing typed arrays against the copied source index preserves one result per source position instead of accidentally aligning a new Series by unrelated labels.

labels=["missing","invalid","valid"]
cents=[None,None,0]
print(list(zip(labels,cents)))
print("absence cause is a separate column")

Implement classification and prove that zero, missing and invalid remain distinct. Then carry that evidence into an explicit aggregation grain.